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