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 #include <linux/cred.h> /* for sg_check_file_access() */
  55
  56 #include "scsi.h"
  57 #include <scsi/scsi_dbg.h>
  58 #include <scsi/scsi_host.h>
  59 #include <scsi/scsi_driver.h>
  60 #include <scsi/scsi_ioctl.h>
  61 #include <scsi/sg.h>
  62
  63 #include "scsi_logging.h"
  64
  65 #ifdef CONFIG_SCSI_PROC_FS
  66 #include <linux/proc_fs.h>
  67 static char *sg_version_date = "20140603";
  68
  69 static int sg_proc_init(void);
  70 static void sg_proc_cleanup(void);
  71 #endif
  72
  73 #define SG_ALLOW_DIO_DEF 0
  74
  75 #define SG_MAX_DEVS 32768
  76
  77 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  78  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  79  * than 16 bytes are "variable length" whose length is a multiple of 4
  80  */
  81 #define SG_MAX_CDB_SIZE 252
  82
  83 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  84
  85 int sg_big_buff = SG_DEF_RESERVED_SIZE;
  86 /* N.B. This variable is readable and writeable via
  87    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  88    of this size (or less if there is not enough memory) will be reserved
  89    for use by this file descriptor. [Deprecated usage: this variable is also
  90    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
  91    the kernel (i.e. it is not a module).] */
  92 static int def_reserved_size = -1;      /* picks up init parameter */
  93 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
  94
  95 static int scatter_elem_sz = SG_SCATTER_SZ;
  96 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
  97
  98 #define SG_SECTOR_SZ 512
  99
 100 static int sg_add_device(struct device *, struct class_interface *);
 101 static void sg_remove_device(struct device *, struct class_interface *);
 102
 103 static DEFINE_IDR(sg_index_idr);
 104 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
 105                                                            file descriptor list for device */
 106
 107 static struct class_interface sg_interface = {
 108         .add_dev        = sg_add_device,
 109         .remove_dev     = sg_remove_device,
 110 };
 111
 112 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 113         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 114         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 115         unsigned bufflen;       /* Size of (aggregate) data buffer */
 116         struct page **pages;
 117         int page_order;
 118         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 119         unsigned char cmd_opcode; /* first byte of command */
 120 } Sg_scatter_hold;
 121
 122 struct sg_device;               /* forward declarations */
 123 struct sg_fd;
 124
 125 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 126         struct list_head entry; /* list entry */
 127         struct sg_fd *parentfp; /* NULL -> not in use */
 128         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 129         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 130         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 131         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 132         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 133         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 134         /* done protected by rq_list_lock */
 135         char done;              /* 0->before bh, 1->before read, 2->read */
 136         struct request *rq;
 137         struct bio *bio;
 138         struct execute_work ew;
 139 } Sg_request;
 140
 141 typedef struct sg_fd {          /* holds the state of a file descriptor */
 142         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 143         struct sg_device *parentdp;     /* owning device */
 144         wait_queue_head_t read_wait;    /* queue read until command done */
 145         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 146         struct mutex f_mutex;   /* protect against changes in this fd */
 147         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 148         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 149         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 150         struct list_head rq_list; /* head of request list */
 151         struct fasync_struct *async_qp; /* used by asynchronous notification */
 152         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 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         char res_in_use;        /* 1 -> 'reserve' array in use */
 159         struct kref f_ref;
 160         struct execute_work ew;
 161 } Sg_fd;
 162
 163 typedef struct sg_device { /* holds the state of each scsi generic device */
 164         struct scsi_device *device;
 165         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 166         struct mutex open_rel_lock;     /* held when in open() or release() */
 167         int sg_tablesize;       /* adapter's max scatter-gather table size */
 168         u32 index;              /* device index number */
 169         struct list_head sfds;
 170         rwlock_t sfd_lock;      /* protect access to sfd list */
 171         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 172         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 173         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 174         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 175         struct gendisk *disk;
 176         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 177         struct kref d_ref;
 178 } Sg_device;
 179
 180 /* tasklet or soft irq callback */
 181 static void sg_rq_end_io(struct request *rq, int uptodate);
 182 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 183 static int sg_finish_rem_req(Sg_request * srp);
 184 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 185 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 186                            Sg_request * srp);
 187 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 188                         const char __user *buf, size_t count, int blocking,
 189                         int read_only, int sg_io_owned, Sg_request **o_srp);
 190 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 191                            unsigned char *cmnd, int timeout, int blocking);
 192 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 193 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 194 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 195 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 196 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 197 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 198 static void sg_remove_sfp(struct kref *);
 199 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 200 static Sg_request *sg_add_request(Sg_fd * sfp);
 201 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 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 /*
 215  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 216  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 217  * to trigger read() and write() calls from various contexts with elevated
 218  * privileges. This can lead to kernel memory corruption (e.g. if these
 219  * interfaces are called through splice()) and privilege escalation inside
 220  * userspace (e.g. if a process with access to such a device passes a file
 221  * descriptor to a SUID binary as stdin/stdout/stderr).
 222  *
 223  * This function provides protection for the legacy API by restricting the
 224  * calling context.
 225  */
 226 static int sg_check_file_access(struct file *filp, const char *caller)
 227 {
 228         if (filp->f_cred != current_real_cred()) {
 229                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 230                         caller, task_tgid_vnr(current), current->comm);
 231                 return -EPERM;
 232         }
 233         if (unlikely(segment_eq(get_fs(), KERNEL_DS))) {
 234                 pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
 235                         caller, task_tgid_vnr(current), current->comm);
 236                 return -EACCES;
 237         }
 238         return 0;
 239 }
 240
 241 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 242 {
 243         struct sg_fd *sfp = filp->private_data;
 244
 245         if (sfp->parentdp->device->type == TYPE_SCANNER)
 246                 return 0;
 247
 248         return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
 249 }
 250
 251 static int
 252 open_wait(Sg_device *sdp, int flags)
 253 {
 254         int retval = 0;
 255
 256         if (flags & O_EXCL) {
 257                 while (sdp->open_cnt > 0) {
 258                         mutex_unlock(&sdp->open_rel_lock);
 259                         retval = wait_event_interruptible(sdp->open_wait,
 260                                         (atomic_read(&sdp->detaching) ||
 261                                          !sdp->open_cnt));
 262                         mutex_lock(&sdp->open_rel_lock);
 263
 264                         if (retval) /* -ERESTARTSYS */
 265                                 return retval;
 266                         if (atomic_read(&sdp->detaching))
 267                                 return -ENODEV;
 268                 }
 269         } else {
 270                 while (sdp->exclude) {
 271                         mutex_unlock(&sdp->open_rel_lock);
 272                         retval = wait_event_interruptible(sdp->open_wait,
 273                                         (atomic_read(&sdp->detaching) ||
 274                                          !sdp->exclude));
 275                         mutex_lock(&sdp->open_rel_lock);
 276
 277                         if (retval) /* -ERESTARTSYS */
 278                                 return retval;
 279                         if (atomic_read(&sdp->detaching))
 280                                 return -ENODEV;
 281                 }
 282         }
 283
 284         return retval;
 285 }
 286
 287 /* Returns 0 on success, else a negated errno value */
 288 static int
 289 sg_open(struct inode *inode, struct file *filp)
 290 {
 291         int dev = iminor(inode);
 292         int flags = filp->f_flags;
 293         struct request_queue *q;
 294         Sg_device *sdp;
 295         Sg_fd *sfp;
 296         int retval;
 297
 298         nonseekable_open(inode, filp);
 299         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 300                 return -EPERM; /* Can't lock it with read only access */
 301         sdp = sg_get_dev(dev);
 302         if (IS_ERR(sdp))
 303                 return PTR_ERR(sdp);
 304
 305         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 306                                       "sg_open: flags=0x%x\n", flags));
 307
 308         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 309         /* Prevent the device driver from vanishing while we sleep */
 310         retval = scsi_device_get(sdp->device);
 311         if (retval)
 312                 goto sg_put;
 313
 314         retval = scsi_autopm_get_device(sdp->device);
 315         if (retval)
 316                 goto sdp_put;
 317
 318         /* scsi_block_when_processing_errors() may block so bypass
 319          * check if O_NONBLOCK. Permits SCSI commands to be issued
 320          * during error recovery. Tread carefully. */
 321         if (!((flags & O_NONBLOCK) ||
 322               scsi_block_when_processing_errors(sdp->device))) {
 323                 retval = -ENXIO;
 324                 /* we are in error recovery for this device */
 325                 goto error_out;
 326         }
 327
 328         mutex_lock(&sdp->open_rel_lock);
 329         if (flags & O_NONBLOCK) {
 330                 if (flags & O_EXCL) {
 331                         if (sdp->open_cnt > 0) {
 332                                 retval = -EBUSY;
 333                                 goto error_mutex_locked;
 334                         }
 335                 } else {
 336                         if (sdp->exclude) {
 337                                 retval = -EBUSY;
 338                                 goto error_mutex_locked;
 339                         }
 340                 }
 341         } else {
 342                 retval = open_wait(sdp, flags);
 343                 if (retval) /* -ERESTARTSYS or -ENODEV */
 344                         goto error_mutex_locked;
 345         }
 346
 347         /* N.B. at this point we are holding the open_rel_lock */
 348         if (flags & O_EXCL)
 349                 sdp->exclude = true;
 350
 351         if (sdp->open_cnt < 1) {  /* no existing opens */
 352                 sdp->sgdebug = 0;
 353                 q = sdp->device->request_queue;
 354                 sdp->sg_tablesize = queue_max_segments(q);
 355         }
 356         sfp = sg_add_sfp(sdp);
 357         if (IS_ERR(sfp)) {
 358                 retval = PTR_ERR(sfp);
 359                 goto out_undo;
 360         }
 361
 362         filp->private_data = sfp;
 363         sdp->open_cnt++;
 364         mutex_unlock(&sdp->open_rel_lock);
 365
 366         retval = 0;
 367 sg_put:
 368         kref_put(&sdp->d_ref, sg_device_destroy);
 369         return retval;
 370
 371 out_undo:
 372         if (flags & O_EXCL) {
 373                 sdp->exclude = false;   /* undo if error */
 374                 wake_up_interruptible(&sdp->open_wait);
 375         }
 376 error_mutex_locked:
 377         mutex_unlock(&sdp->open_rel_lock);
 378 error_out:
 379         scsi_autopm_put_device(sdp->device);
 380 sdp_put:
 381         scsi_device_put(sdp->device);
 382         goto sg_put;
 383 }
 384
 385 /* Release resources associated with a successful sg_open()
 386  * Returns 0 on success, else a negated errno value */
 387 static int
 388 sg_release(struct inode *inode, struct file *filp)
 389 {
 390         Sg_device *sdp;
 391         Sg_fd *sfp;
 392
 393         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 394                 return -ENXIO;
 395         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 396
 397         mutex_lock(&sdp->open_rel_lock);
 398         scsi_autopm_put_device(sdp->device);
 399         kref_put(&sfp->f_ref, sg_remove_sfp);
 400         sdp->open_cnt--;
 401
 402         /* possibly many open()s waiting on exlude clearing, start many;
 403          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 404         if (sdp->exclude) {
 405                 sdp->exclude = false;
 406                 wake_up_interruptible_all(&sdp->open_wait);
 407         } else if (0 == sdp->open_cnt) {
 408                 wake_up_interruptible(&sdp->open_wait);
 409         }
 410         mutex_unlock(&sdp->open_rel_lock);
 411         return 0;
 412 }
 413
 414 static ssize_t
 415 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 416 {
 417         Sg_device *sdp;
 418         Sg_fd *sfp;
 419         Sg_request *srp;
 420         int req_pack_id = -1;
 421         sg_io_hdr_t *hp;
 422         struct sg_header *old_hdr = NULL;
 423         int retval = 0;
 424
 425         /*
 426          * This could cause a response to be stranded. Close the associated
 427          * file descriptor to free up any resources being held.
 428          */
 429         retval = sg_check_file_access(filp, __func__);
 430         if (retval)
 431                 return retval;
 432
 433         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 434                 return -ENXIO;
 435         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 436                                       "sg_read: count=%d\n", (int) count));
 437
 438         if (!access_ok(VERIFY_WRITE, buf, count))
 439                 return -EFAULT;
 440         if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
 441                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 442                 if (!old_hdr)
 443                         return -ENOMEM;
 444                 if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
 445                         retval = -EFAULT;
 446                         goto free_old_hdr;
 447                 }
 448                 if (old_hdr->reply_len < 0) {
 449                         if (count >= SZ_SG_IO_HDR) {
 450                                 sg_io_hdr_t *new_hdr;
 451                                 new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
 452                                 if (!new_hdr) {
 453                                         retval = -ENOMEM;
 454                                         goto free_old_hdr;
 455                                 }
 456                                 retval =__copy_from_user
 457                                     (new_hdr, buf, SZ_SG_IO_HDR);
 458                                 req_pack_id = new_hdr->pack_id;
 459                                 kfree(new_hdr);
 460                                 if (retval) {
 461                                         retval = -EFAULT;
 462                                         goto free_old_hdr;
 463                                 }
 464                         }
 465                 } else
 466                         req_pack_id = old_hdr->pack_id;
 467         }
 468         srp = sg_get_rq_mark(sfp, req_pack_id);
 469         if (!srp) {             /* now wait on packet to arrive */
 470                 if (atomic_read(&sdp->detaching)) {
 471                         retval = -ENODEV;
 472                         goto free_old_hdr;
 473                 }
 474                 if (filp->f_flags & O_NONBLOCK) {
 475                         retval = -EAGAIN;
 476                         goto free_old_hdr;
 477                 }
 478                 retval = wait_event_interruptible(sfp->read_wait,
 479                         (atomic_read(&sdp->detaching) ||
 480                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 481                 if (atomic_read(&sdp->detaching)) {
 482                         retval = -ENODEV;
 483                         goto free_old_hdr;
 484                 }
 485                 if (retval) {
 486                         /* -ERESTARTSYS as signal hit process */
 487                         goto free_old_hdr;
 488                 }
 489         }
 490         if (srp->header.interface_id != '\0') {
 491                 retval = sg_new_read(sfp, buf, count, srp);
 492                 goto free_old_hdr;
 493         }
 494
 495         hp = &srp->header;
 496         if (old_hdr == NULL) {
 497                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 498                 if (! old_hdr) {
 499                         retval = -ENOMEM;
 500                         goto free_old_hdr;
 501                 }
 502         }
 503         memset(old_hdr, 0, SZ_SG_HEADER);
 504         old_hdr->reply_len = (int) hp->timeout;
 505         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 506         old_hdr->pack_id = hp->pack_id;
 507         old_hdr->twelve_byte =
 508             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 509         old_hdr->target_status = hp->masked_status;
 510         old_hdr->host_status = hp->host_status;
 511         old_hdr->driver_status = hp->driver_status;
 512         if ((CHECK_CONDITION & hp->masked_status) ||
 513             (DRIVER_SENSE & hp->driver_status))
 514                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 515                        sizeof (old_hdr->sense_buffer));
 516         switch (hp->host_status) {
 517         /* This setup of 'result' is for backward compatibility and is best
 518            ignored by the user who should use target, host + driver status */
 519         case DID_OK:
 520         case DID_PASSTHROUGH:
 521         case DID_SOFT_ERROR:
 522                 old_hdr->result = 0;
 523                 break;
 524         case DID_NO_CONNECT:
 525         case DID_BUS_BUSY:
 526         case DID_TIME_OUT:
 527                 old_hdr->result = EBUSY;
 528                 break;
 529         case DID_BAD_TARGET:
 530         case DID_ABORT:
 531         case DID_PARITY:
 532         case DID_RESET:
 533         case DID_BAD_INTR:
 534                 old_hdr->result = EIO;
 535                 break;
 536         case DID_ERROR:
 537                 old_hdr->result = (srp->sense_b[0] == 0 &&
 538                                   hp->masked_status == GOOD) ? 0 : EIO;
 539                 break;
 540         default:
 541                 old_hdr->result = EIO;
 542                 break;
 543         }
 544
 545         /* Now copy the result back to the user buffer.  */
 546         if (count >= SZ_SG_HEADER) {
 547                 if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 548                         retval = -EFAULT;
 549                         goto free_old_hdr;
 550                 }
 551                 buf += SZ_SG_HEADER;
 552                 if (count > old_hdr->reply_len)
 553                         count = old_hdr->reply_len;
 554                 if (count > SZ_SG_HEADER) {
 555                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 556                                 retval = -EFAULT;
 557                                 goto free_old_hdr;
 558                         }
 559                 }
 560         } else
 561                 count = (old_hdr->result == 0) ? 0 : -EIO;
 562         sg_finish_rem_req(srp);
 563         sg_remove_request(sfp, srp);
 564         retval = count;
 565 free_old_hdr:
 566         kfree(old_hdr);
 567         return retval;
 568 }
 569
 570 static ssize_t
 571 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 572 {
 573         sg_io_hdr_t *hp = &srp->header;
 574         int err = 0, err2;
 575         int len;
 576
 577         if (count < SZ_SG_IO_HDR) {
 578                 err = -EINVAL;
 579                 goto err_out;
 580         }
 581         hp->sb_len_wr = 0;
 582         if ((hp->mx_sb_len > 0) && hp->sbp) {
 583                 if ((CHECK_CONDITION & hp->masked_status) ||
 584                     (DRIVER_SENSE & hp->driver_status)) {
 585                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 586                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 587                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 588                         len = (len > sb_len) ? sb_len : len;
 589                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 590                                 err = -EFAULT;
 591                                 goto err_out;
 592                         }
 593                         hp->sb_len_wr = len;
 594                 }
 595         }
 596         if (hp->masked_status || hp->host_status || hp->driver_status)
 597                 hp->info |= SG_INFO_CHECK;
 598         if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
 599                 err = -EFAULT;
 600                 goto err_out;
 601         }
 602 err_out:
 603         err2 = sg_finish_rem_req(srp);
 604         sg_remove_request(sfp, srp);
 605         return err ? : err2 ? : count;
 606 }
 607
 608 static ssize_t
 609 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 610 {
 611         int mxsize, cmd_size, k;
 612         int input_size, blocking;
 613         unsigned char opcode;
 614         Sg_device *sdp;
 615         Sg_fd *sfp;
 616         Sg_request *srp;
 617         struct sg_header old_hdr;
 618         sg_io_hdr_t *hp;
 619         unsigned char cmnd[SG_MAX_CDB_SIZE];
 620         int retval;
 621
 622         retval = sg_check_file_access(filp, __func__);
 623         if (retval)
 624                 return retval;
 625
 626         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 627                 return -ENXIO;
 628         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 629                                       "sg_write: count=%d\n", (int) count));
 630         if (atomic_read(&sdp->detaching))
 631                 return -ENODEV;
 632         if (!((filp->f_flags & O_NONBLOCK) ||
 633               scsi_block_when_processing_errors(sdp->device)))
 634                 return -ENXIO;
 635
 636         if (!access_ok(VERIFY_READ, buf, count))
 637                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 638         if (count < SZ_SG_HEADER)
 639                 return -EIO;
 640         if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 641                 return -EFAULT;
 642         blocking = !(filp->f_flags & O_NONBLOCK);
 643         if (old_hdr.reply_len < 0)
 644                 return sg_new_write(sfp, filp, buf, count,
 645                                     blocking, 0, 0, NULL);
 646         if (count < (SZ_SG_HEADER + 6))
 647                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 648
 649         if (!(srp = sg_add_request(sfp))) {
 650                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 651                                               "sg_write: queue full\n"));
 652                 return -EDOM;
 653         }
 654         buf += SZ_SG_HEADER;
 655         __get_user(opcode, buf);
 656         mutex_lock(&sfp->f_mutex);
 657         if (sfp->next_cmd_len > 0) {
 658                 cmd_size = sfp->next_cmd_len;
 659                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 660         } else {
 661                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 662                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 663                         cmd_size = 12;
 664         }
 665         mutex_unlock(&sfp->f_mutex);
 666         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 667                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 668 /* Determine buffer size.  */
 669         input_size = count - cmd_size;
 670         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 671         mxsize -= SZ_SG_HEADER;
 672         input_size -= SZ_SG_HEADER;
 673         if (input_size < 0) {
 674                 sg_remove_request(sfp, srp);
 675                 return -EIO;    /* User did not pass enough bytes for this command. */
 676         }
 677         hp = &srp->header;
 678         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 679         hp->cmd_len = (unsigned char) cmd_size;
 680         hp->iovec_count = 0;
 681         hp->mx_sb_len = 0;
 682         if (input_size > 0)
 683                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 684                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 685         else
 686                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 687         hp->dxfer_len = mxsize;
 688         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 689             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 690                 hp->dxferp = (char __user *)buf + cmd_size;
 691         else
 692                 hp->dxferp = NULL;
 693         hp->sbp = NULL;
 694         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 695         hp->flags = input_size; /* structure abuse ... */
 696         hp->pack_id = old_hdr.pack_id;
 697         hp->usr_ptr = NULL;
 698         if (__copy_from_user(cmnd, buf, cmd_size)) {
 699                 sg_remove_request(sfp, srp);
 700                 return -EFAULT;
 701         }
 702         /*
 703          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 704          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 705          * is a non-zero input_size, so emit a warning.
 706          */
 707         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 708                 printk_ratelimited(KERN_WARNING
 709                                    "sg_write: data in/out %d/%d bytes "
 710                                    "for SCSI command 0x%x-- guessing "
 711                                    "data in;\n   program %s not setting "
 712                                    "count and/or reply_len properly\n",
 713                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 714                                    input_size, (unsigned int) cmnd[0],
 715                                    current->comm);
 716         }
 717         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 718         return (k < 0) ? k : count;
 719 }
 720
 721 static ssize_t
 722 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 723                  size_t count, int blocking, int read_only, int sg_io_owned,
 724                  Sg_request **o_srp)
 725 {
 726         int k;
 727         Sg_request *srp;
 728         sg_io_hdr_t *hp;
 729         unsigned char cmnd[SG_MAX_CDB_SIZE];
 730         int timeout;
 731         unsigned long ul_timeout;
 732
 733         if (count < SZ_SG_IO_HDR)
 734                 return -EINVAL;
 735         if (!access_ok(VERIFY_READ, buf, count))
 736                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 737
 738         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 739         if (!(srp = sg_add_request(sfp))) {
 740                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 741                                               "sg_new_write: queue full\n"));
 742                 return -EDOM;
 743         }
 744         srp->sg_io_owned = sg_io_owned;
 745         hp = &srp->header;
 746         if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
 747                 sg_remove_request(sfp, srp);
 748                 return -EFAULT;
 749         }
 750         if (hp->interface_id != 'S') {
 751                 sg_remove_request(sfp, srp);
 752                 return -ENOSYS;
 753         }
 754         if (hp->flags & SG_FLAG_MMAP_IO) {
 755                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 756                         sg_remove_request(sfp, srp);
 757                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 758                 }
 759                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 760                         sg_remove_request(sfp, srp);
 761                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 762                 }
 763                 if (sfp->res_in_use) {
 764                         sg_remove_request(sfp, srp);
 765                         return -EBUSY;  /* reserve buffer already being used */
 766                 }
 767         }
 768         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 769         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 770         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 771                 sg_remove_request(sfp, srp);
 772                 return -EMSGSIZE;
 773         }
 774         if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
 775                 sg_remove_request(sfp, srp);
 776                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 777         }
 778         if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 779                 sg_remove_request(sfp, srp);
 780                 return -EFAULT;
 781         }
 782         if (read_only && sg_allow_access(file, cmnd)) {
 783                 sg_remove_request(sfp, srp);
 784                 return -EPERM;
 785         }
 786         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 787         if (k < 0)
 788                 return k;
 789         if (o_srp)
 790                 *o_srp = srp;
 791         return count;
 792 }
 793
 794 static int
 795 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 796                 unsigned char *cmnd, int timeout, int blocking)
 797 {
 798         int k, at_head;
 799         Sg_device *sdp = sfp->parentdp;
 800         sg_io_hdr_t *hp = &srp->header;
 801
 802         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 803         hp->status = 0;
 804         hp->masked_status = 0;
 805         hp->msg_status = 0;
 806         hp->info = 0;
 807         hp->host_status = 0;
 808         hp->driver_status = 0;
 809         hp->resid = 0;
 810         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 811                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 812                         (int) cmnd[0], (int) hp->cmd_len));
 813
 814         if (hp->dxfer_len >= SZ_256M) {
 815                 sg_remove_request(sfp, srp);
 816                 return -EINVAL;
 817         }
 818
 819         k = sg_start_req(srp, cmnd);
 820         if (k) {
 821                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 822                         "sg_common_write: start_req err=%d\n", k));
 823                 sg_finish_rem_req(srp);
 824                 sg_remove_request(sfp, srp);
 825                 return k;       /* probably out of space --> ENOMEM */
 826         }
 827         if (atomic_read(&sdp->detaching)) {
 828                 if (srp->bio) {
 829                         if (srp->rq->cmd != srp->rq->__cmd)
 830                                 kfree(srp->rq->cmd);
 831
 832                         blk_end_request_all(srp->rq, -EIO);
 833                         srp->rq = NULL;
 834                 }
 835
 836                 sg_finish_rem_req(srp);
 837                 sg_remove_request(sfp, srp);
 838                 return -ENODEV;
 839         }
 840
 841         hp->duration = jiffies_to_msecs(jiffies);
 842         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 843             (SG_FLAG_Q_AT_TAIL & hp->flags))
 844                 at_head = 0;
 845         else
 846                 at_head = 1;
 847
 848         srp->rq->timeout = timeout;
 849         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 850         blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
 851                               srp->rq, at_head, sg_rq_end_io);
 852         return 0;
 853 }
 854
 855 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 856 {
 857         unsigned long flags;
 858         int ret;
 859
 860         read_lock_irqsave(&sfp->rq_list_lock, flags);
 861         ret = srp->done;
 862         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 863         return ret;
 864 }
 865
 866 static int max_sectors_bytes(struct request_queue *q)
 867 {
 868         unsigned int max_sectors = queue_max_sectors(q);
 869
 870         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 871
 872         return max_sectors << 9;
 873 }
 874
 875 static void
 876 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 877 {
 878         Sg_request *srp;
 879         int val;
 880         unsigned int ms;
 881
 882         val = 0;
 883         list_for_each_entry(srp, &sfp->rq_list, entry) {
 884                 if (val >= SG_MAX_QUEUE)
 885                         break;
 886                 rinfo[val].req_state = srp->done + 1;
 887                 rinfo[val].problem =
 888                         srp->header.masked_status &
 889                         srp->header.host_status &
 890                         srp->header.driver_status;
 891                 if (srp->done)
 892                         rinfo[val].duration =
 893                                 srp->header.duration;
 894                 else {
 895                         ms = jiffies_to_msecs(jiffies);
 896                         rinfo[val].duration =
 897                                 (ms > srp->header.duration) ?
 898                                 (ms - srp->header.duration) : 0;
 899                 }
 900                 rinfo[val].orphan = srp->orphan;
 901                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 902                 rinfo[val].pack_id = srp->header.pack_id;
 903                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 904                 val++;
 905         }
 906 }
 907
 908 static long
 909 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
 910 {
 911         void __user *p = (void __user *)arg;
 912         int __user *ip = p;
 913         int result, val, read_only;
 914         Sg_device *sdp;
 915         Sg_fd *sfp;
 916         Sg_request *srp;
 917         unsigned long iflags;
 918
 919         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 920                 return -ENXIO;
 921
 922         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 923                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 924         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 925
 926         switch (cmd_in) {
 927         case SG_IO:
 928                 if (atomic_read(&sdp->detaching))
 929                         return -ENODEV;
 930                 if (!scsi_block_when_processing_errors(sdp->device))
 931                         return -ENXIO;
 932                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
 933                         return -EFAULT;
 934                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 935                                  1, read_only, 1, &srp);
 936                 if (result < 0)
 937                         return result;
 938                 result = wait_event_interruptible(sfp->read_wait,
 939                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 940                 if (atomic_read(&sdp->detaching))
 941                         return -ENODEV;
 942                 write_lock_irq(&sfp->rq_list_lock);
 943                 if (srp->done) {
 944                         srp->done = 2;
 945                         write_unlock_irq(&sfp->rq_list_lock);
 946                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 947                         return (result < 0) ? result : 0;
 948                 }
 949                 srp->orphan = 1;
 950                 write_unlock_irq(&sfp->rq_list_lock);
 951                 return result;  /* -ERESTARTSYS because signal hit process */
 952         case SG_SET_TIMEOUT:
 953                 result = get_user(val, ip);
 954                 if (result)
 955                         return result;
 956                 if (val < 0)
 957                         return -EIO;
 958                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 959                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 960                                     INT_MAX);
 961                 sfp->timeout_user = val;
 962                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 963
 964                 return 0;
 965         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 966                                 /* strange ..., for backward compatibility */
 967                 return sfp->timeout_user;
 968         case SG_SET_FORCE_LOW_DMA:
 969                 /*
 970                  * N.B. This ioctl never worked properly, but failed to
 971                  * return an error value. So returning '0' to keep compability
 972                  * with legacy applications.
 973                  */
 974                 return 0;
 975         case SG_GET_LOW_DMA:
 976                 return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
 977         case SG_GET_SCSI_ID:
 978                 if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
 979                         return -EFAULT;
 980                 else {
 981                         sg_scsi_id_t __user *sg_idp = p;
 982
 983                         if (atomic_read(&sdp->detaching))
 984                                 return -ENODEV;
 985                         __put_user((int) sdp->device->host->host_no,
 986                                    &sg_idp->host_no);
 987                         __put_user((int) sdp->device->channel,
 988                                    &sg_idp->channel);
 989                         __put_user((int) sdp->device->id, &sg_idp->scsi_id);
 990                         __put_user((int) sdp->device->lun, &sg_idp->lun);
 991                         __put_user((int) sdp->device->type, &sg_idp->scsi_type);
 992                         __put_user((short) sdp->device->host->cmd_per_lun,
 993                                    &sg_idp->h_cmd_per_lun);
 994                         __put_user((short) sdp->device->queue_depth,
 995                                    &sg_idp->d_queue_depth);
 996                         __put_user(0, &sg_idp->unused[0]);
 997                         __put_user(0, &sg_idp->unused[1]);
 998                         return 0;
 999                 }
1000         case SG_SET_FORCE_PACK_ID:
1001                 result = get_user(val, ip);
1002                 if (result)
1003                         return result;
1004                 sfp->force_packid = val ? 1 : 0;
1005                 return 0;
1006         case SG_GET_PACK_ID:
1007                 if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
1008                         return -EFAULT;
1009                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1010                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1011                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1012                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1013                                                        iflags);
1014                                 __put_user(srp->header.pack_id, ip);
1015                                 return 0;
1016                         }
1017                 }
1018                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1019                 __put_user(-1, ip);
1020                 return 0;
1021         case SG_GET_NUM_WAITING:
1022                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1023                 val = 0;
1024                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1025                         if ((1 == srp->done) && (!srp->sg_io_owned))
1026                                 ++val;
1027                 }
1028                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1029                 return put_user(val, ip);
1030         case SG_GET_SG_TABLESIZE:
1031                 return put_user(sdp->sg_tablesize, ip);
1032         case SG_SET_RESERVED_SIZE:
1033                 result = get_user(val, ip);
1034                 if (result)
1035                         return result;
1036                 if (val < 0)
1037                         return -EINVAL;
1038                 val = min_t(int, val,
1039                             max_sectors_bytes(sdp->device->request_queue));
1040                 mutex_lock(&sfp->f_mutex);
1041                 if (val != sfp->reserve.bufflen) {
1042                         if (sfp->mmap_called ||
1043                             sfp->res_in_use) {
1044                                 mutex_unlock(&sfp->f_mutex);
1045                                 return -EBUSY;
1046                         }
1047
1048                         sg_remove_scat(sfp, &sfp->reserve);
1049                         sg_build_reserve(sfp, val);
1050                 }
1051                 mutex_unlock(&sfp->f_mutex);
1052                 return 0;
1053         case SG_GET_RESERVED_SIZE:
1054                 val = min_t(int, sfp->reserve.bufflen,
1055                             max_sectors_bytes(sdp->device->request_queue));
1056                 return put_user(val, ip);
1057         case SG_SET_COMMAND_Q:
1058                 result = get_user(val, ip);
1059                 if (result)
1060                         return result;
1061                 sfp->cmd_q = val ? 1 : 0;
1062                 return 0;
1063         case SG_GET_COMMAND_Q:
1064                 return put_user((int) sfp->cmd_q, ip);
1065         case SG_SET_KEEP_ORPHAN:
1066                 result = get_user(val, ip);
1067                 if (result)
1068                         return result;
1069                 sfp->keep_orphan = val;
1070                 return 0;
1071         case SG_GET_KEEP_ORPHAN:
1072                 return put_user((int) sfp->keep_orphan, ip);
1073         case SG_NEXT_CMD_LEN:
1074                 result = get_user(val, ip);
1075                 if (result)
1076                         return result;
1077                 if (val > SG_MAX_CDB_SIZE)
1078                         return -ENOMEM;
1079                 sfp->next_cmd_len = (val > 0) ? val : 0;
1080                 return 0;
1081         case SG_GET_VERSION_NUM:
1082                 return put_user(sg_version_num, ip);
1083         case SG_GET_ACCESS_COUNT:
1084                 /* faked - we don't have a real access count anymore */
1085                 val = (sdp->device ? 1 : 0);
1086                 return put_user(val, ip);
1087         case SG_GET_REQUEST_TABLE:
1088                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1089                         return -EFAULT;
1090                 else {
1091                         sg_req_info_t *rinfo;
1092
1093                         rinfo = kzalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
1094                                         GFP_KERNEL);
1095                         if (!rinfo)
1096                                 return -ENOMEM;
1097                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1098                         sg_fill_request_table(sfp, rinfo);
1099                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1100                         result = __copy_to_user(p, rinfo,
1101                                                 SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1102                         result = result ? -EFAULT : 0;
1103                         kfree(rinfo);
1104                         return result;
1105                 }
1106         case SG_EMULATED_HOST:
1107                 if (atomic_read(&sdp->detaching))
1108                         return -ENODEV;
1109                 return put_user(sdp->device->host->hostt->emulated, ip);
1110         case SCSI_IOCTL_SEND_COMMAND:
1111                 if (atomic_read(&sdp->detaching))
1112                         return -ENODEV;
1113                 if (read_only) {
1114                         unsigned char opcode = WRITE_6;
1115                         Scsi_Ioctl_Command __user *siocp = p;
1116
1117                         if (copy_from_user(&opcode, siocp->data, 1))
1118                                 return -EFAULT;
1119                         if (sg_allow_access(filp, &opcode))
1120                                 return -EPERM;
1121                 }
1122                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1123         case SG_SET_DEBUG:
1124                 result = get_user(val, ip);
1125                 if (result)
1126                         return result;
1127                 sdp->sgdebug = (char) val;
1128                 return 0;
1129         case BLKSECTGET:
1130                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1131                                 ip);
1132         case BLKTRACESETUP:
1133                 return blk_trace_setup(sdp->device->request_queue,
1134                                        sdp->disk->disk_name,
1135                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1136                                        NULL,
1137                                        (char *)arg);
1138         case BLKTRACESTART:
1139                 return blk_trace_startstop(sdp->device->request_queue, 1);
1140         case BLKTRACESTOP:
1141                 return blk_trace_startstop(sdp->device->request_queue, 0);
1142         case BLKTRACETEARDOWN:
1143                 return blk_trace_remove(sdp->device->request_queue);
1144         case SCSI_IOCTL_GET_IDLUN:
1145         case SCSI_IOCTL_GET_BUS_NUMBER:
1146         case SCSI_IOCTL_PROBE_HOST:
1147         case SG_GET_TRANSFORM:
1148         case SG_SCSI_RESET:
1149                 if (atomic_read(&sdp->detaching))
1150                         return -ENODEV;
1151                 break;
1152         default:
1153                 if (read_only)
1154                         return -EPERM;  /* don't know so take safe approach */
1155                 break;
1156         }
1157
1158         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1159                         cmd_in, filp->f_flags & O_NDELAY);
1160         if (result)
1161                 return result;
1162         return scsi_ioctl(sdp->device, cmd_in, p);
1163 }
1164
1165 #ifdef CONFIG_COMPAT
1166 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1167 {
1168         Sg_device *sdp;
1169         Sg_fd *sfp;
1170         struct scsi_device *sdev;
1171
1172         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1173                 return -ENXIO;
1174
1175         sdev = sdp->device;
1176         if (sdev->host->hostt->compat_ioctl) {
1177                 int ret;
1178
1179                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1180
1181                 return ret;
1182         }
1183
1184         return -ENOIOCTLCMD;
1185 }
1186 #endif
1187
1188 static unsigned int
1189 sg_poll(struct file *filp, poll_table * wait)
1190 {
1191         unsigned int res = 0;
1192         Sg_device *sdp;
1193         Sg_fd *sfp;
1194         Sg_request *srp;
1195         int count = 0;
1196         unsigned long iflags;
1197
1198         sfp = filp->private_data;
1199         if (!sfp)
1200                 return POLLERR;
1201         sdp = sfp->parentdp;
1202         if (!sdp)
1203                 return POLLERR;
1204         poll_wait(filp, &sfp->read_wait, wait);
1205         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1206         list_for_each_entry(srp, &sfp->rq_list, entry) {
1207                 /* if any read waiting, flag it */
1208                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1209                         res = POLLIN | POLLRDNORM;
1210                 ++count;
1211         }
1212         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1213
1214         if (atomic_read(&sdp->detaching))
1215                 res |= POLLHUP;
1216         else if (!sfp->cmd_q) {
1217                 if (0 == count)
1218                         res |= POLLOUT | POLLWRNORM;
1219         } else if (count < SG_MAX_QUEUE)
1220                 res |= POLLOUT | POLLWRNORM;
1221         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1222                                       "sg_poll: res=0x%x\n", (int) res));
1223         return res;
1224 }
1225
1226 static int
1227 sg_fasync(int fd, struct file *filp, int mode)
1228 {
1229         Sg_device *sdp;
1230         Sg_fd *sfp;
1231
1232         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1233                 return -ENXIO;
1234         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1235                                       "sg_fasync: mode=%d\n", mode));
1236
1237         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1238 }
1239
1240 static int
1241 sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1242 {
1243         Sg_fd *sfp;
1244         unsigned long offset, len, sa;
1245         Sg_scatter_hold *rsv_schp;
1246         int k, length;
1247
1248         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1249                 return VM_FAULT_SIGBUS;
1250         rsv_schp = &sfp->reserve;
1251         offset = vmf->pgoff << PAGE_SHIFT;
1252         if (offset >= rsv_schp->bufflen)
1253                 return VM_FAULT_SIGBUS;
1254         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1255                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1256                                       offset, rsv_schp->k_use_sg));
1257         sa = vma->vm_start;
1258         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1259         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1260                 len = vma->vm_end - sa;
1261                 len = (len < length) ? len : length;
1262                 if (offset < len) {
1263                         struct page *page = nth_page(rsv_schp->pages[k],
1264                                                      offset >> PAGE_SHIFT);
1265                         get_page(page); /* increment page count */
1266                         vmf->page = page;
1267                         return 0; /* success */
1268                 }
1269                 sa += len;
1270                 offset -= len;
1271         }
1272
1273         return VM_FAULT_SIGBUS;
1274 }
1275
1276 static const struct vm_operations_struct sg_mmap_vm_ops = {
1277         .fault = sg_vma_fault,
1278 };
1279
1280 static int
1281 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1282 {
1283         Sg_fd *sfp;
1284         unsigned long req_sz, len, sa;
1285         Sg_scatter_hold *rsv_schp;
1286         int k, length;
1287         int ret = 0;
1288
1289         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1290                 return -ENXIO;
1291         req_sz = vma->vm_end - vma->vm_start;
1292         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1293                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1294                                       (void *) vma->vm_start, (int) req_sz));
1295         if (vma->vm_pgoff)
1296                 return -EINVAL; /* want no offset */
1297         rsv_schp = &sfp->reserve;
1298         mutex_lock(&sfp->f_mutex);
1299         if (req_sz > rsv_schp->bufflen) {
1300                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1301                 goto out;
1302         }
1303
1304         sa = vma->vm_start;
1305         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1306         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1307                 len = vma->vm_end - sa;
1308                 len = (len < length) ? len : length;
1309                 sa += len;
1310         }
1311
1312         sfp->mmap_called = 1;
1313         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1314         vma->vm_private_data = sfp;
1315         vma->vm_ops = &sg_mmap_vm_ops;
1316 out:
1317         mutex_unlock(&sfp->f_mutex);
1318         return ret;
1319 }
1320
1321 static void
1322 sg_rq_end_io_usercontext(struct work_struct *work)
1323 {
1324         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1325         struct sg_fd *sfp = srp->parentfp;
1326
1327         sg_finish_rem_req(srp);
1328         sg_remove_request(sfp, srp);
1329         kref_put(&sfp->f_ref, sg_remove_sfp);
1330 }
1331
1332 /*
1333  * This function is a "bottom half" handler that is called by the mid
1334  * level when a command is completed (or has failed).
1335  */
1336 static void
1337 sg_rq_end_io(struct request *rq, int uptodate)
1338 {
1339         struct sg_request *srp = rq->end_io_data;
1340         Sg_device *sdp;
1341         Sg_fd *sfp;
1342         unsigned long iflags;
1343         unsigned int ms;
1344         char *sense;
1345         int result, resid, done = 1;
1346
1347         if (WARN_ON(srp->done != 0))
1348                 return;
1349
1350         sfp = srp->parentfp;
1351         if (WARN_ON(sfp == NULL))
1352                 return;
1353
1354         sdp = sfp->parentdp;
1355         if (unlikely(atomic_read(&sdp->detaching)))
1356                 pr_info("%s: device detaching\n", __func__);
1357
1358         sense = rq->sense;
1359         result = rq->errors;
1360         resid = rq->resid_len;
1361
1362         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1363                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1364                                       srp->header.pack_id, result));
1365         srp->header.resid = resid;
1366         ms = jiffies_to_msecs(jiffies);
1367         srp->header.duration = (ms > srp->header.duration) ?
1368                                 (ms - srp->header.duration) : 0;
1369         if (0 != result) {
1370                 struct scsi_sense_hdr sshdr;
1371
1372                 srp->header.status = 0xff & result;
1373                 srp->header.masked_status = status_byte(result);
1374                 srp->header.msg_status = msg_byte(result);
1375                 srp->header.host_status = host_byte(result);
1376                 srp->header.driver_status = driver_byte(result);
1377                 if ((sdp->sgdebug > 0) &&
1378                     ((CHECK_CONDITION == srp->header.masked_status) ||
1379                      (COMMAND_TERMINATED == srp->header.masked_status)))
1380                         __scsi_print_sense(sdp->device, __func__, sense,
1381                                            SCSI_SENSE_BUFFERSIZE);
1382
1383                 /* Following if statement is a patch supplied by Eric Youngdale */
1384                 if (driver_byte(result) != 0
1385                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1386                     && !scsi_sense_is_deferred(&sshdr)
1387                     && sshdr.sense_key == UNIT_ATTENTION
1388                     && sdp->device->removable) {
1389                         /* Detected possible disc change. Set the bit - this */
1390                         /* may be used if there are filesystems using this device */
1391                         sdp->device->changed = 1;
1392                 }
1393         }
1394         /* Rely on write phase to clean out srp status values, so no "else" */
1395
1396         /*
1397          * Free the request as soon as it is complete so that its resources
1398          * can be reused without waiting for userspace to read() the
1399          * result.  But keep the associated bio (if any) around until
1400          * blk_rq_unmap_user() can be called from user context.
1401          */
1402         srp->rq = NULL;
1403         if (rq->cmd != rq->__cmd)
1404                 kfree(rq->cmd);
1405         __blk_put_request(rq->q, rq);
1406
1407         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1408         if (unlikely(srp->orphan)) {
1409                 if (sfp->keep_orphan)
1410                         srp->sg_io_owned = 0;
1411                 else
1412                         done = 0;
1413         }
1414         srp->done = done;
1415         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1416
1417         if (likely(done)) {
1418                 /* Now wake up any sg_read() that is waiting for this
1419                  * packet.
1420                  */
1421                 wake_up_interruptible(&sfp->read_wait);
1422                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1423                 kref_put(&sfp->f_ref, sg_remove_sfp);
1424         } else {
1425                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1426                 schedule_work(&srp->ew.work);
1427         }
1428 }
1429
1430 static const struct file_operations sg_fops = {
1431         .owner = THIS_MODULE,
1432         .read = sg_read,
1433         .write = sg_write,
1434         .poll = sg_poll,
1435         .unlocked_ioctl = sg_ioctl,
1436 #ifdef CONFIG_COMPAT
1437         .compat_ioctl = sg_compat_ioctl,
1438 #endif
1439         .open = sg_open,
1440         .mmap = sg_mmap,
1441         .release = sg_release,
1442         .fasync = sg_fasync,
1443         .llseek = no_llseek,
1444 };
1445
1446 static struct class *sg_sysfs_class;
1447
1448 static int sg_sysfs_valid = 0;
1449
1450 static Sg_device *
1451 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1452 {
1453         struct request_queue *q = scsidp->request_queue;
1454         Sg_device *sdp;
1455         unsigned long iflags;
1456         int error;
1457         u32 k;
1458
1459         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1460         if (!sdp) {
1461                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1462                             "failure\n", __func__);
1463                 return ERR_PTR(-ENOMEM);
1464         }
1465
1466         idr_preload(GFP_KERNEL);
1467         write_lock_irqsave(&sg_index_lock, iflags);
1468
1469         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1470         if (error < 0) {
1471                 if (error == -ENOSPC) {
1472                         sdev_printk(KERN_WARNING, scsidp,
1473                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1474                                     scsidp->type, SG_MAX_DEVS - 1);
1475                         error = -ENODEV;
1476                 } else {
1477                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1478                                     "allocation Sg_device failure: %d\n",
1479                                     __func__, error);
1480                 }
1481                 goto out_unlock;
1482         }
1483         k = error;
1484
1485         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1486                                         "sg_alloc: dev=%d \n", k));
1487         sprintf(disk->disk_name, "sg%d", k);
1488         disk->first_minor = k;
1489         sdp->disk = disk;
1490         sdp->device = scsidp;
1491         mutex_init(&sdp->open_rel_lock);
1492         INIT_LIST_HEAD(&sdp->sfds);
1493         init_waitqueue_head(&sdp->open_wait);
1494         atomic_set(&sdp->detaching, 0);
1495         rwlock_init(&sdp->sfd_lock);
1496         sdp->sg_tablesize = queue_max_segments(q);
1497         sdp->index = k;
1498         kref_init(&sdp->d_ref);
1499         error = 0;
1500
1501 out_unlock:
1502         write_unlock_irqrestore(&sg_index_lock, iflags);
1503         idr_preload_end();
1504
1505         if (error) {
1506                 kfree(sdp);
1507                 return ERR_PTR(error);
1508         }
1509         return sdp;
1510 }
1511
1512 static int
1513 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1514 {
1515         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1516         struct gendisk *disk;
1517         Sg_device *sdp = NULL;
1518         struct cdev * cdev = NULL;
1519         int error;
1520         unsigned long iflags;
1521
1522         disk = alloc_disk(1);
1523         if (!disk) {
1524                 pr_warn("%s: alloc_disk failed\n", __func__);
1525                 return -ENOMEM;
1526         }
1527         disk->major = SCSI_GENERIC_MAJOR;
1528
1529         error = -ENOMEM;
1530         cdev = cdev_alloc();
1531         if (!cdev) {
1532                 pr_warn("%s: cdev_alloc failed\n", __func__);
1533                 goto out;
1534         }
1535         cdev->owner = THIS_MODULE;
1536         cdev->ops = &sg_fops;
1537
1538         sdp = sg_alloc(disk, scsidp);
1539         if (IS_ERR(sdp)) {
1540                 pr_warn("%s: sg_alloc failed\n", __func__);
1541                 error = PTR_ERR(sdp);
1542                 goto out;
1543         }
1544
1545         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1546         if (error)
1547                 goto cdev_add_err;
1548
1549         sdp->cdev = cdev;
1550         if (sg_sysfs_valid) {
1551                 struct device *sg_class_member;
1552
1553                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1554                                                 MKDEV(SCSI_GENERIC_MAJOR,
1555                                                       sdp->index),
1556                                                 sdp, "%s", disk->disk_name);
1557                 if (IS_ERR(sg_class_member)) {
1558                         pr_err("%s: device_create failed\n", __func__);
1559                         error = PTR_ERR(sg_class_member);
1560                         goto cdev_add_err;
1561                 }
1562                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1563                                           &sg_class_member->kobj, "generic");
1564                 if (error)
1565                         pr_err("%s: unable to make symlink 'generic' back "
1566                                "to sg%d\n", __func__, sdp->index);
1567         } else
1568                 pr_warn("%s: sg_sys Invalid\n", __func__);
1569
1570         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1571                     "type %d\n", sdp->index, scsidp->type);
1572
1573         dev_set_drvdata(cl_dev, sdp);
1574
1575         return 0;
1576
1577 cdev_add_err:
1578         write_lock_irqsave(&sg_index_lock, iflags);
1579         idr_remove(&sg_index_idr, sdp->index);
1580         write_unlock_irqrestore(&sg_index_lock, iflags);
1581         kfree(sdp);
1582
1583 out:
1584         put_disk(disk);
1585         if (cdev)
1586                 cdev_del(cdev);
1587         return error;
1588 }
1589
1590 static void
1591 sg_device_destroy(struct kref *kref)
1592 {
1593         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1594         unsigned long flags;
1595
1596         /* CAUTION!  Note that the device can still be found via idr_find()
1597          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1598          * any other cleanup.
1599          */
1600
1601         write_lock_irqsave(&sg_index_lock, flags);
1602         idr_remove(&sg_index_idr, sdp->index);
1603         write_unlock_irqrestore(&sg_index_lock, flags);
1604
1605         SCSI_LOG_TIMEOUT(3,
1606                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1607
1608         put_disk(sdp->disk);
1609         kfree(sdp);
1610 }
1611
1612 static void
1613 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1614 {
1615         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1616         Sg_device *sdp = dev_get_drvdata(cl_dev);
1617         unsigned long iflags;
1618         Sg_fd *sfp;
1619         int val;
1620
1621         if (!sdp)
1622                 return;
1623         /* want sdp->detaching non-zero as soon as possible */
1624         val = atomic_inc_return(&sdp->detaching);
1625         if (val > 1)
1626                 return; /* only want to do following once per device */
1627
1628         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1629                                       "%s\n", __func__));
1630
1631         read_lock_irqsave(&sdp->sfd_lock, iflags);
1632         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1633                 wake_up_interruptible_all(&sfp->read_wait);
1634                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1635         }
1636         wake_up_interruptible_all(&sdp->open_wait);
1637         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1638
1639         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1640         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1641         cdev_del(sdp->cdev);
1642         sdp->cdev = NULL;
1643
1644         kref_put(&sdp->d_ref, sg_device_destroy);
1645 }
1646
1647 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1648 module_param_named(def_reserved_size, def_reserved_size, int,
1649                    S_IRUGO | S_IWUSR);
1650 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1651
1652 MODULE_AUTHOR("Douglas Gilbert");
1653 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1654 MODULE_LICENSE("GPL");
1655 MODULE_VERSION(SG_VERSION_STR);
1656 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1657
1658 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1659                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1660 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1661 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1662
1663 static int __init
1664 init_sg(void)
1665 {
1666         int rc;
1667
1668         if (scatter_elem_sz < PAGE_SIZE) {
1669                 scatter_elem_sz = PAGE_SIZE;
1670                 scatter_elem_sz_prev = scatter_elem_sz;
1671         }
1672         if (def_reserved_size >= 0)
1673                 sg_big_buff = def_reserved_size;
1674         else
1675                 def_reserved_size = sg_big_buff;
1676
1677         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1678                                     SG_MAX_DEVS, "sg");
1679         if (rc)
1680                 return rc;
1681         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1682         if ( IS_ERR(sg_sysfs_class) ) {
1683                 rc = PTR_ERR(sg_sysfs_class);
1684                 goto err_out;
1685         }
1686         sg_sysfs_valid = 1;
1687         rc = scsi_register_interface(&sg_interface);
1688         if (0 == rc) {
1689 #ifdef CONFIG_SCSI_PROC_FS
1690                 sg_proc_init();
1691 #endif                          /* CONFIG_SCSI_PROC_FS */
1692                 return 0;
1693         }
1694         class_destroy(sg_sysfs_class);
1695 err_out:
1696         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1697         return rc;
1698 }
1699
1700 static void __exit
1701 exit_sg(void)
1702 {
1703 #ifdef CONFIG_SCSI_PROC_FS
1704         sg_proc_cleanup();
1705 #endif                          /* CONFIG_SCSI_PROC_FS */
1706         scsi_unregister_interface(&sg_interface);
1707         class_destroy(sg_sysfs_class);
1708         sg_sysfs_valid = 0;
1709         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1710                                  SG_MAX_DEVS);
1711         idr_destroy(&sg_index_idr);
1712 }
1713
1714 static int
1715 sg_start_req(Sg_request *srp, unsigned char *cmd)
1716 {
1717         int res;
1718         struct request *rq;
1719         Sg_fd *sfp = srp->parentfp;
1720         sg_io_hdr_t *hp = &srp->header;
1721         int dxfer_len = (int) hp->dxfer_len;
1722         int dxfer_dir = hp->dxfer_direction;
1723         unsigned int iov_count = hp->iovec_count;
1724         Sg_scatter_hold *req_schp = &srp->data;
1725         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1726         struct request_queue *q = sfp->parentdp->device->request_queue;
1727         struct rq_map_data *md, map_data;
1728         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1729         unsigned char *long_cmdp = NULL;
1730
1731         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1732                                       "sg_start_req: dxfer_len=%d\n",
1733                                       dxfer_len));
1734
1735         if (hp->cmd_len > BLK_MAX_CDB) {
1736                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1737                 if (!long_cmdp)
1738                         return -ENOMEM;
1739         }
1740
1741         /*
1742          * NOTE
1743          *
1744          * With scsi-mq enabled, there are a fixed number of preallocated
1745          * requests equal in number to shost->can_queue.  If all of the
1746          * preallocated requests are already in use, then using GFP_ATOMIC with
1747          * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1748          * will cause blk_get_request() to sleep until an active command
1749          * completes, freeing up a request.  Neither option is ideal, but
1750          * GFP_KERNEL is the better choice to prevent userspace from getting an
1751          * unexpected EWOULDBLOCK.
1752          *
1753          * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1754          * does not sleep except under memory pressure.
1755          */
1756         rq = blk_get_request(q, rw, GFP_KERNEL);
1757         if (IS_ERR(rq)) {
1758                 kfree(long_cmdp);
1759                 return PTR_ERR(rq);
1760         }
1761
1762         blk_rq_set_block_pc(rq);
1763
1764         if (hp->cmd_len > BLK_MAX_CDB)
1765                 rq->cmd = long_cmdp;
1766         memcpy(rq->cmd, cmd, hp->cmd_len);
1767         rq->cmd_len = hp->cmd_len;
1768
1769         srp->rq = rq;
1770         rq->end_io_data = srp;
1771         rq->sense = srp->sense_b;
1772         rq->retries = SG_DEFAULT_RETRIES;
1773
1774         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1775                 return 0;
1776
1777         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1778             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1779             !sfp->parentdp->device->host->unchecked_isa_dma &&
1780             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1781                 md = NULL;
1782         else
1783                 md = &map_data;
1784
1785         if (md) {
1786                 mutex_lock(&sfp->f_mutex);
1787                 if (dxfer_len <= rsv_schp->bufflen &&
1788                     !sfp->res_in_use) {
1789                         sfp->res_in_use = 1;
1790                         sg_link_reserve(sfp, srp, dxfer_len);
1791                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1792                         res = -EBUSY; /* sfp->res_in_use == 1 */
1793                         if (dxfer_len > rsv_schp->bufflen)
1794                                 res = -ENOMEM;
1795                         mutex_unlock(&sfp->f_mutex);
1796                         return res;
1797                 } else {
1798                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1799                         if (res) {
1800                                 mutex_unlock(&sfp->f_mutex);
1801                                 return res;
1802                         }
1803                 }
1804                 mutex_unlock(&sfp->f_mutex);
1805
1806                 md->pages = req_schp->pages;
1807                 md->page_order = req_schp->page_order;
1808                 md->nr_entries = req_schp->k_use_sg;
1809                 md->offset = 0;
1810                 md->null_mapped = hp->dxferp ? 0 : 1;
1811                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1812                         md->from_user = 1;
1813                 else
1814                         md->from_user = 0;
1815         }
1816
1817         if (iov_count) {
1818                 struct iovec *iov = NULL;
1819                 struct iov_iter i;
1820
1821                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1822                 if (res < 0)
1823                         return res;
1824
1825                 iov_iter_truncate(&i, hp->dxfer_len);
1826                 if (!iov_iter_count(&i)) {
1827                         kfree(iov);
1828                         return -EINVAL;
1829                 }
1830
1831                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1832                 kfree(iov);
1833         } else
1834                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1835                                       hp->dxfer_len, GFP_ATOMIC);
1836
1837         if (!res) {
1838                 srp->bio = rq->bio;
1839
1840                 if (!md) {
1841                         req_schp->dio_in_use = 1;
1842                         hp->info |= SG_INFO_DIRECT_IO;
1843                 }
1844         }
1845         return res;
1846 }
1847
1848 static int
1849 sg_finish_rem_req(Sg_request *srp)
1850 {
1851         int ret = 0;
1852
1853         Sg_fd *sfp = srp->parentfp;
1854         Sg_scatter_hold *req_schp = &srp->data;
1855
1856         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1857                                       "sg_finish_rem_req: res_used=%d\n",
1858                                       (int) srp->res_used));
1859         if (srp->bio)
1860                 ret = blk_rq_unmap_user(srp->bio);
1861
1862         if (srp->rq) {
1863                 if (srp->rq->cmd != srp->rq->__cmd)
1864                         kfree(srp->rq->cmd);
1865                 blk_put_request(srp->rq);
1866         }
1867
1868         if (srp->res_used)
1869                 sg_unlink_reserve(sfp, srp);
1870         else
1871                 sg_remove_scat(sfp, req_schp);
1872
1873         return ret;
1874 }
1875
1876 static int
1877 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1878 {
1879         int sg_bufflen = tablesize * sizeof(struct page *);
1880         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1881
1882         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1883         if (!schp->pages)
1884                 return -ENOMEM;
1885         schp->sglist_len = sg_bufflen;
1886         return tablesize;       /* number of scat_gath elements allocated */
1887 }
1888
1889 static int
1890 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1891 {
1892         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1893         int sg_tablesize = sfp->parentdp->sg_tablesize;
1894         int blk_size = buff_size, order;
1895         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1896         struct sg_device *sdp = sfp->parentdp;
1897
1898         if (blk_size < 0)
1899                 return -EFAULT;
1900         if (0 == blk_size)
1901                 ++blk_size;     /* don't know why */
1902         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1903         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1904         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1905                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1906                 buff_size, blk_size));
1907
1908         /* N.B. ret_sz carried into this block ... */
1909         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1910         if (mx_sc_elems < 0)
1911                 return mx_sc_elems;     /* most likely -ENOMEM */
1912
1913         num = scatter_elem_sz;
1914         if (unlikely(num != scatter_elem_sz_prev)) {
1915                 if (num < PAGE_SIZE) {
1916                         scatter_elem_sz = PAGE_SIZE;
1917                         scatter_elem_sz_prev = PAGE_SIZE;
1918                 } else
1919                         scatter_elem_sz_prev = num;
1920         }
1921
1922         if (sdp->device->host->unchecked_isa_dma)
1923                 gfp_mask |= GFP_DMA;
1924
1925         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1926                 gfp_mask |= __GFP_ZERO;
1927
1928         order = get_order(num);
1929 retry:
1930         ret_sz = 1 << (PAGE_SHIFT + order);
1931
1932         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1933              k++, rem_sz -= ret_sz) {
1934
1935                 num = (rem_sz > scatter_elem_sz_prev) ?
1936                         scatter_elem_sz_prev : rem_sz;
1937
1938                 schp->pages[k] = alloc_pages(gfp_mask | __GFP_ZERO, order);
1939                 if (!schp->pages[k])
1940                         goto out;
1941
1942                 if (num == scatter_elem_sz_prev) {
1943                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1944                                 scatter_elem_sz = ret_sz;
1945                                 scatter_elem_sz_prev = ret_sz;
1946                         }
1947                 }
1948
1949                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1950                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1951                                  k, num, ret_sz));
1952         }               /* end of for loop */
1953
1954         schp->page_order = order;
1955         schp->k_use_sg = k;
1956         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1957                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1958                          k, rem_sz));
1959
1960         schp->bufflen = blk_size;
1961         if (rem_sz > 0) /* must have failed */
1962                 return -ENOMEM;
1963         return 0;
1964 out:
1965         for (i = 0; i < k; i++)
1966                 __free_pages(schp->pages[i], order);
1967
1968         if (--order >= 0)
1969                 goto retry;
1970
1971         return -ENOMEM;
1972 }
1973
1974 static void
1975 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1976 {
1977         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1978                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1979         if (schp->pages && schp->sglist_len > 0) {
1980                 if (!schp->dio_in_use) {
1981                         int k;
1982
1983                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1984                                 SCSI_LOG_TIMEOUT(5,
1985                                         sg_printk(KERN_INFO, sfp->parentdp,
1986                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1987                                         k, schp->pages[k]));
1988                                 __free_pages(schp->pages[k], schp->page_order);
1989                         }
1990
1991                         kfree(schp->pages);
1992                 }
1993         }
1994         memset(schp, 0, sizeof (*schp));
1995 }
1996
1997 static int
1998 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1999 {
2000         Sg_scatter_hold *schp = &srp->data;
2001         int k, num;
2002
2003         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2004                          "sg_read_oxfer: num_read_xfer=%d\n",
2005                          num_read_xfer));
2006         if ((!outp) || (num_read_xfer <= 0))
2007                 return 0;
2008
2009         num = 1 << (PAGE_SHIFT + schp->page_order);
2010         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2011                 if (num > num_read_xfer) {
2012                         if (__copy_to_user(outp, page_address(schp->pages[k]),
2013                                            num_read_xfer))
2014                                 return -EFAULT;
2015                         break;
2016                 } else {
2017                         if (__copy_to_user(outp, page_address(schp->pages[k]),
2018                                            num))
2019                                 return -EFAULT;
2020                         num_read_xfer -= num;
2021                         if (num_read_xfer <= 0)
2022                                 break;
2023                         outp += num;
2024                 }
2025         }
2026
2027         return 0;
2028 }
2029
2030 static void
2031 sg_build_reserve(Sg_fd * sfp, int req_size)
2032 {
2033         Sg_scatter_hold *schp = &sfp->reserve;
2034
2035         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2036                          "sg_build_reserve: req_size=%d\n", req_size));
2037         do {
2038                 if (req_size < PAGE_SIZE)
2039                         req_size = PAGE_SIZE;
2040                 if (0 == sg_build_indirect(schp, sfp, req_size))
2041                         return;
2042                 else
2043                         sg_remove_scat(sfp, schp);
2044                 req_size >>= 1; /* divide by 2 */
2045         } while (req_size > (PAGE_SIZE / 2));
2046 }
2047
2048 static void
2049 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2050 {
2051         Sg_scatter_hold *req_schp = &srp->data;
2052         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2053         int k, num, rem;
2054
2055         srp->res_used = 1;
2056         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2057                          "sg_link_reserve: size=%d\n", size));
2058         rem = size;
2059
2060         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2061         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2062                 if (rem <= num) {
2063                         req_schp->k_use_sg = k + 1;
2064                         req_schp->sglist_len = rsv_schp->sglist_len;
2065                         req_schp->pages = rsv_schp->pages;
2066
2067                         req_schp->bufflen = size;
2068                         req_schp->page_order = rsv_schp->page_order;
2069                         break;
2070                 } else
2071                         rem -= num;
2072         }
2073
2074         if (k >= rsv_schp->k_use_sg)
2075                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2076                                  "sg_link_reserve: BAD size\n"));
2077 }
2078
2079 static void
2080 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2081 {
2082         Sg_scatter_hold *req_schp = &srp->data;
2083
2084         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2085                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2086                                       (int) req_schp->k_use_sg));
2087         req_schp->k_use_sg = 0;
2088         req_schp->bufflen = 0;
2089         req_schp->pages = NULL;
2090         req_schp->page_order = 0;
2091         req_schp->sglist_len = 0;
2092         srp->res_used = 0;
2093         /* Called without mutex lock to avoid deadlock */
2094         sfp->res_in_use = 0;
2095 }
2096
2097 static Sg_request *
2098 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2099 {
2100         Sg_request *resp;
2101         unsigned long iflags;
2102
2103         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2104         list_for_each_entry(resp, &sfp->rq_list, entry) {
2105                 /* look for requests that are ready + not SG_IO owned */
2106                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2107                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2108                         resp->done = 2; /* guard against other readers */
2109                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2110                         return resp;
2111                 }
2112         }
2113         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2114         return NULL;
2115 }
2116
2117 /* always adds to end of list */
2118 static Sg_request *
2119 sg_add_request(Sg_fd * sfp)
2120 {
2121         int k;
2122         unsigned long iflags;
2123         Sg_request *rp = sfp->req_arr;
2124
2125         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2126         if (!list_empty(&sfp->rq_list)) {
2127                 if (!sfp->cmd_q)
2128                         goto out_unlock;
2129
2130                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2131                         if (!rp->parentfp)
2132                                 break;
2133                 }
2134                 if (k >= SG_MAX_QUEUE)
2135                         goto out_unlock;
2136         }
2137         memset(rp, 0, sizeof (Sg_request));
2138         rp->parentfp = sfp;
2139         rp->header.duration = jiffies_to_msecs(jiffies);
2140         list_add_tail(&rp->entry, &sfp->rq_list);
2141         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2142         return rp;
2143 out_unlock:
2144         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2145         return NULL;
2146 }
2147
2148 /* Return of 1 for found; 0 for not found */
2149 static int
2150 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2151 {
2152         unsigned long iflags;
2153         int res = 0;
2154
2155         if (!sfp || !srp || list_empty(&sfp->rq_list))
2156                 return res;
2157         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2158         if (!list_empty(&srp->entry)) {
2159                 list_del(&srp->entry);
2160                 srp->parentfp = NULL;
2161                 res = 1;
2162         }
2163         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2164         return res;
2165 }
2166
2167 static Sg_fd *
2168 sg_add_sfp(Sg_device * sdp)
2169 {
2170         Sg_fd *sfp;
2171         unsigned long iflags;
2172         int bufflen;
2173
2174         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2175         if (!sfp)
2176                 return ERR_PTR(-ENOMEM);
2177
2178         init_waitqueue_head(&sfp->read_wait);
2179         rwlock_init(&sfp->rq_list_lock);
2180         INIT_LIST_HEAD(&sfp->rq_list);
2181         kref_init(&sfp->f_ref);
2182         mutex_init(&sfp->f_mutex);
2183         sfp->timeout = SG_DEFAULT_TIMEOUT;
2184         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2185         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2186         sfp->cmd_q = SG_DEF_COMMAND_Q;
2187         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2188         sfp->parentdp = sdp;
2189         write_lock_irqsave(&sdp->sfd_lock, iflags);
2190         if (atomic_read(&sdp->detaching)) {
2191                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2192                 kfree(sfp);
2193                 return ERR_PTR(-ENODEV);
2194         }
2195         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2196         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2197         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2198                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2199         if (unlikely(sg_big_buff != def_reserved_size))
2200                 sg_big_buff = def_reserved_size;
2201
2202         bufflen = min_t(int, sg_big_buff,
2203                         max_sectors_bytes(sdp->device->request_queue));
2204         sg_build_reserve(sfp, bufflen);
2205         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2206                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2207                                       sfp->reserve.bufflen,
2208                                       sfp->reserve.k_use_sg));
2209
2210         kref_get(&sdp->d_ref);
2211         __module_get(THIS_MODULE);
2212         return sfp;
2213 }
2214
2215 static void
2216 sg_remove_sfp_usercontext(struct work_struct *work)
2217 {
2218         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2219         struct sg_device *sdp = sfp->parentdp;
2220         Sg_request *srp;
2221         unsigned long iflags;
2222
2223         /* Cleanup any responses which were never read(). */
2224         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2225         while (!list_empty(&sfp->rq_list)) {
2226                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2227                 sg_finish_rem_req(srp);
2228                 list_del(&srp->entry);
2229                 srp->parentfp = NULL;
2230         }
2231         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2232
2233         if (sfp->reserve.bufflen > 0) {
2234                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2235                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2236                                 (int) sfp->reserve.bufflen,
2237                                 (int) sfp->reserve.k_use_sg));
2238                 sg_remove_scat(sfp, &sfp->reserve);
2239         }
2240
2241         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2242                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2243         kfree(sfp);
2244
2245         scsi_device_put(sdp->device);
2246         kref_put(&sdp->d_ref, sg_device_destroy);
2247         module_put(THIS_MODULE);
2248 }
2249
2250 static void
2251 sg_remove_sfp(struct kref *kref)
2252 {
2253         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2254         struct sg_device *sdp = sfp->parentdp;
2255         unsigned long iflags;
2256
2257         write_lock_irqsave(&sdp->sfd_lock, iflags);
2258         list_del(&sfp->sfd_siblings);
2259         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2260
2261         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2262         schedule_work(&sfp->ew.work);
2263 }
2264
2265 #ifdef CONFIG_SCSI_PROC_FS
2266 static int
2267 sg_idr_max_id(int id, void *p, void *data)
2268 {
2269         int *k = data;
2270
2271         if (*k < id)
2272                 *k = id;
2273
2274         return 0;
2275 }
2276
2277 static int
2278 sg_last_dev(void)
2279 {
2280         int k = -1;
2281         unsigned long iflags;
2282
2283         read_lock_irqsave(&sg_index_lock, iflags);
2284         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2285         read_unlock_irqrestore(&sg_index_lock, iflags);
2286         return k + 1;           /* origin 1 */
2287 }
2288 #endif
2289
2290 /* must be called with sg_index_lock held */
2291 static Sg_device *sg_lookup_dev(int dev)
2292 {
2293         return idr_find(&sg_index_idr, dev);
2294 }
2295
2296 static Sg_device *
2297 sg_get_dev(int dev)
2298 {
2299         struct sg_device *sdp;
2300         unsigned long flags;
2301
2302         read_lock_irqsave(&sg_index_lock, flags);
2303         sdp = sg_lookup_dev(dev);
2304         if (!sdp)
2305                 sdp = ERR_PTR(-ENXIO);
2306         else if (atomic_read(&sdp->detaching)) {
2307                 /* If sdp->detaching, then the refcount may already be 0, in
2308                  * which case it would be a bug to do kref_get().
2309                  */
2310                 sdp = ERR_PTR(-ENODEV);
2311         } else
2312                 kref_get(&sdp->d_ref);
2313         read_unlock_irqrestore(&sg_index_lock, flags);
2314
2315         return sdp;
2316 }
2317
2318 #ifdef CONFIG_SCSI_PROC_FS
2319
2320 static struct proc_dir_entry *sg_proc_sgp = NULL;
2321
2322 static char sg_proc_sg_dirname[] = "scsi/sg";
2323
2324 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2325
2326 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2327 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2328                                   size_t count, loff_t *off);
2329 static const struct file_operations adio_fops = {
2330         .owner = THIS_MODULE,
2331         .open = sg_proc_single_open_adio,
2332         .read = seq_read,
2333         .llseek = seq_lseek,
2334         .write = sg_proc_write_adio,
2335         .release = single_release,
2336 };
2337
2338 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2339 static ssize_t sg_proc_write_dressz(struct file *filp,
2340                 const char __user *buffer, size_t count, loff_t *off);
2341 static const struct file_operations dressz_fops = {
2342         .owner = THIS_MODULE,
2343         .open = sg_proc_single_open_dressz,
2344         .read = seq_read,
2345         .llseek = seq_lseek,
2346         .write = sg_proc_write_dressz,
2347         .release = single_release,
2348 };
2349
2350 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2351 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2352 static const struct file_operations version_fops = {
2353         .owner = THIS_MODULE,
2354         .open = sg_proc_single_open_version,
2355         .read = seq_read,
2356         .llseek = seq_lseek,
2357         .release = single_release,
2358 };
2359
2360 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2361 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2362 static const struct file_operations devhdr_fops = {
2363         .owner = THIS_MODULE,
2364         .open = sg_proc_single_open_devhdr,
2365         .read = seq_read,
2366         .llseek = seq_lseek,
2367         .release = single_release,
2368 };
2369
2370 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2371 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2372 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2373 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2374 static void dev_seq_stop(struct seq_file *s, void *v);
2375 static const struct file_operations dev_fops = {
2376         .owner = THIS_MODULE,
2377         .open = sg_proc_open_dev,
2378         .read = seq_read,
2379         .llseek = seq_lseek,
2380         .release = seq_release,
2381 };
2382 static const struct seq_operations dev_seq_ops = {
2383         .start = dev_seq_start,
2384         .next  = dev_seq_next,
2385         .stop  = dev_seq_stop,
2386         .show  = sg_proc_seq_show_dev,
2387 };
2388
2389 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2390 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2391 static const struct file_operations devstrs_fops = {
2392         .owner = THIS_MODULE,
2393         .open = sg_proc_open_devstrs,
2394         .read = seq_read,
2395         .llseek = seq_lseek,
2396         .release = seq_release,
2397 };
2398 static const struct seq_operations devstrs_seq_ops = {
2399         .start = dev_seq_start,
2400         .next  = dev_seq_next,
2401         .stop  = dev_seq_stop,
2402         .show  = sg_proc_seq_show_devstrs,
2403 };
2404
2405 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2406 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2407 static const struct file_operations debug_fops = {
2408         .owner = THIS_MODULE,
2409         .open = sg_proc_open_debug,
2410         .read = seq_read,
2411         .llseek = seq_lseek,
2412         .release = seq_release,
2413 };
2414 static const struct seq_operations debug_seq_ops = {
2415         .start = dev_seq_start,
2416         .next  = dev_seq_next,
2417         .stop  = dev_seq_stop,
2418         .show  = sg_proc_seq_show_debug,
2419 };
2420
2421
2422 struct sg_proc_leaf {
2423         const char * name;
2424         const struct file_operations * fops;
2425 };
2426
2427 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2428         {"allow_dio", &adio_fops},
2429         {"debug", &debug_fops},
2430         {"def_reserved_size", &dressz_fops},
2431         {"device_hdr", &devhdr_fops},
2432         {"devices", &dev_fops},
2433         {"device_strs", &devstrs_fops},
2434         {"version", &version_fops}
2435 };
2436
2437 static int
2438 sg_proc_init(void)
2439 {
2440         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2441         int k;
2442
2443         sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2444         if (!sg_proc_sgp)
2445                 return 1;
2446         for (k = 0; k < num_leaves; ++k) {
2447                 const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2448                 umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2449                 proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2450         }
2451         return 0;
2452 }
2453
2454 static void
2455 sg_proc_cleanup(void)
2456 {
2457         int k;
2458         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2459
2460         if (!sg_proc_sgp)
2461                 return;
2462         for (k = 0; k < num_leaves; ++k)
2463                 remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2464         remove_proc_entry(sg_proc_sg_dirname, NULL);
2465 }
2466
2467
2468 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2469 {
2470         seq_printf(s, "%d\n", *((int *)s->private));
2471         return 0;
2472 }
2473
2474 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2475 {
2476         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2477 }
2478
2479 static ssize_t
2480 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2481                    size_t count, loff_t *off)
2482 {
2483         int err;
2484         unsigned long num;
2485
2486         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2487                 return -EACCES;
2488         err = kstrtoul_from_user(buffer, count, 0, &num);
2489         if (err)
2490                 return err;
2491         sg_allow_dio = num ? 1 : 0;
2492         return count;
2493 }
2494
2495 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2496 {
2497         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2498 }
2499
2500 static ssize_t
2501 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2502                      size_t count, loff_t *off)
2503 {
2504         int err;
2505         unsigned long k = ULONG_MAX;
2506
2507         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2508                 return -EACCES;
2509
2510         err = kstrtoul_from_user(buffer, count, 0, &k);
2511         if (err)
2512                 return err;
2513         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2514                 sg_big_buff = k;
2515                 return count;
2516         }
2517         return -ERANGE;
2518 }
2519
2520 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2521 {
2522         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2523                    sg_version_date);
2524         return 0;
2525 }
2526
2527 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2528 {
2529         return single_open(file, sg_proc_seq_show_version, NULL);
2530 }
2531
2532 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2533 {
2534         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2535         return 0;
2536 }
2537
2538 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2539 {
2540         return single_open(file, sg_proc_seq_show_devhdr, NULL);
2541 }
2542
2543 struct sg_proc_deviter {
2544         loff_t  index;
2545         size_t  max;
2546 };
2547
2548 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2549 {
2550         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2551
2552         s->private = it;
2553         if (! it)
2554                 return NULL;
2555
2556         it->index = *pos;
2557         it->max = sg_last_dev();
2558         if (it->index >= it->max)
2559                 return NULL;
2560         return it;
2561 }
2562
2563 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2564 {
2565         struct sg_proc_deviter * it = s->private;
2566
2567         *pos = ++it->index;
2568         return (it->index < it->max) ? it : NULL;
2569 }
2570
2571 static void dev_seq_stop(struct seq_file *s, void *v)
2572 {
2573         kfree(s->private);
2574 }
2575
2576 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2577 {
2578         return seq_open(file, &dev_seq_ops);
2579 }
2580
2581 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2582 {
2583         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2584         Sg_device *sdp;
2585         struct scsi_device *scsidp;
2586         unsigned long iflags;
2587
2588         read_lock_irqsave(&sg_index_lock, iflags);
2589         sdp = it ? sg_lookup_dev(it->index) : NULL;
2590         if ((NULL == sdp) || (NULL == sdp->device) ||
2591             (atomic_read(&sdp->detaching)))
2592                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2593         else {
2594                 scsidp = sdp->device;
2595                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2596                               scsidp->host->host_no, scsidp->channel,
2597                               scsidp->id, scsidp->lun, (int) scsidp->type,
2598                               1,
2599                               (int) scsidp->queue_depth,
2600                               (int) atomic_read(&scsidp->device_busy),
2601                               (int) scsi_device_online(scsidp));
2602         }
2603         read_unlock_irqrestore(&sg_index_lock, iflags);
2604         return 0;
2605 }
2606
2607 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2608 {
2609         return seq_open(file, &devstrs_seq_ops);
2610 }
2611
2612 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2613 {
2614         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2615         Sg_device *sdp;
2616         struct scsi_device *scsidp;
2617         unsigned long iflags;
2618
2619         read_lock_irqsave(&sg_index_lock, iflags);
2620         sdp = it ? sg_lookup_dev(it->index) : NULL;
2621         scsidp = sdp ? sdp->device : NULL;
2622         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2623                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2624                            scsidp->vendor, scsidp->model, scsidp->rev);
2625         else
2626                 seq_puts(s, "<no active device>\n");
2627         read_unlock_irqrestore(&sg_index_lock, iflags);
2628         return 0;
2629 }
2630
2631 /* must be called while holding sg_index_lock */
2632 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2633 {
2634         int k, new_interface, blen, usg;
2635         Sg_request *srp;
2636         Sg_fd *fp;
2637         const sg_io_hdr_t *hp;
2638         const char * cp;
2639         unsigned int ms;
2640
2641         k = 0;
2642         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2643                 k++;
2644                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2645                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2646                            "(res)sgat=%d low_dma=%d\n", k,
2647                            jiffies_to_msecs(fp->timeout),
2648                            fp->reserve.bufflen,
2649                            (int) fp->reserve.k_use_sg,
2650                            (int) sdp->device->host->unchecked_isa_dma);
2651                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2652                            (int) fp->cmd_q, (int) fp->force_packid,
2653                            (int) fp->keep_orphan);
2654                 list_for_each_entry(srp, &fp->rq_list, entry) {
2655                         hp = &srp->header;
2656                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2657                         if (srp->res_used) {
2658                                 if (new_interface &&
2659                                     (SG_FLAG_MMAP_IO & hp->flags))
2660                                         cp = "     mmap>> ";
2661                                 else
2662                                         cp = "     rb>> ";
2663                         } else {
2664                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2665                                         cp = "     dio>> ";
2666                                 else
2667                                         cp = "     ";
2668                         }
2669                         seq_puts(s, cp);
2670                         blen = srp->data.bufflen;
2671                         usg = srp->data.k_use_sg;
2672                         seq_puts(s, srp->done ?
2673                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2674                                   : "act:");
2675                         seq_printf(s, " id=%d blen=%d",
2676                                    srp->header.pack_id, blen);
2677                         if (srp->done)
2678                                 seq_printf(s, " dur=%d", hp->duration);
2679                         else {
2680                                 ms = jiffies_to_msecs(jiffies);
2681                                 seq_printf(s, " t_o/elap=%d/%d",
2682                                         (new_interface ? hp->timeout :
2683                                                   jiffies_to_msecs(fp->timeout)),
2684                                         (ms > hp->duration ? ms - hp->duration : 0));
2685                         }
2686                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2687                                    (int) srp->data.cmd_opcode);
2688                 }
2689                 if (list_empty(&fp->rq_list))
2690                         seq_puts(s, "     No requests active\n");
2691                 read_unlock(&fp->rq_list_lock);
2692         }
2693 }
2694
2695 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2696 {
2697         return seq_open(file, &debug_seq_ops);
2698 }
2699
2700 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2701 {
2702         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2703         Sg_device *sdp;
2704         unsigned long iflags;
2705
2706         if (it && (0 == it->index))
2707                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2708                            (int)it->max, sg_big_buff);
2709
2710         read_lock_irqsave(&sg_index_lock, iflags);
2711         sdp = it ? sg_lookup_dev(it->index) : NULL;
2712         if (NULL == sdp)
2713                 goto skip;
2714         read_lock(&sdp->sfd_lock);
2715         if (!list_empty(&sdp->sfds)) {
2716                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2717                 if (atomic_read(&sdp->detaching))
2718                         seq_puts(s, "detaching pending close ");
2719                 else if (sdp->device) {
2720                         struct scsi_device *scsidp = sdp->device;
2721
2722                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2723                                    scsidp->host->host_no,
2724                                    scsidp->channel, scsidp->id,
2725                                    scsidp->lun,
2726                                    scsidp->host->hostt->emulated);
2727                 }
2728                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2729                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2730                 sg_proc_debug_helper(s, sdp);
2731         }
2732         read_unlock(&sdp->sfd_lock);
2733 skip:
2734         read_unlock_irqrestore(&sg_index_lock, iflags);
2735         return 0;
2736 }
2737
2738 #endif                          /* CONFIG_SCSI_PROC_FS */
2739
2740 module_init(init_sg);
2741 module_exit(exit_sg);