drivers/scsi/sg.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  History:
   4  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   5  *           to allow user process control of SCSI devices.
   6  *  Development Sponsored by Killy Corp. NY NY
   7  *
   8  * Original driver (sg.c):
   9  *        Copyright (C) 1992 Lawrence Foard
  10  * Version 2 and 3 extensions to driver:
  11  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  12  */
  13
  14 static int sg_version_num = 30536;      /* 2 digits for each component */
  15 #define SG_VERSION_STR "3.5.36"
  16
  17 /*
  18  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  19  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  20  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  21  *        (otherwise the macros compile to empty statements).
  22  *
  23  */
  24 #include <linux/module.h>
  25
  26 #include <linux/fs.h>
  27 #include <linux/kernel.h>
  28 #include <linux/sched.h>
  29 #include <linux/string.h>
  30 #include <linux/mm.h>
  31 #include <linux/errno.h>
  32 #include <linux/mtio.h>
  33 #include <linux/ioctl.h>
  34 #include <linux/slab.h>
  35 #include <linux/fcntl.h>
  36 #include <linux/init.h>
  37 #include <linux/poll.h>
  38 #include <linux/moduleparam.h>
  39 #include <linux/cdev.h>
  40 #include <linux/idr.h>
  41 #include <linux/seq_file.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/delay.h>
  44 #include <linux/blktrace_api.h>
  45 #include <linux/mutex.h>
  46 #include <linux/atomic.h>
  47 #include <linux/ratelimit.h>
  48 #include <linux/uio.h>
  49 #include <linux/cred.h> /* for sg_check_file_access() */
  50
  51 #include "scsi.h"
  52 #include <scsi/scsi_dbg.h>
  53 #include <scsi/scsi_host.h>
  54 #include <scsi/scsi_driver.h>
  55 #include <scsi/scsi_ioctl.h>
  56 #include <scsi/sg.h>
  57
  58 #include "scsi_logging.h"
  59
  60 #ifdef CONFIG_SCSI_PROC_FS
  61 #include <linux/proc_fs.h>
  62 static char *sg_version_date = "20140603";
  63
  64 static int sg_proc_init(void);
  65 #endif
  66
  67 #define SG_ALLOW_DIO_DEF 0
  68
  69 #define SG_MAX_DEVS 32768
  70
  71 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  72  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  73  * than 16 bytes are "variable length" whose length is a multiple of 4
  74  */
  75 #define SG_MAX_CDB_SIZE 252
  76
  77 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  78
  79 int sg_big_buff = SG_DEF_RESERVED_SIZE;
  80 /* N.B. This variable is readable and writeable via
  81    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  82    of this size (or less if there is not enough memory) will be reserved
  83    for use by this file descriptor. [Deprecated usage: this variable is also
  84    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
  85    the kernel (i.e. it is not a module).] */
  86 static int def_reserved_size = -1;      /* picks up init parameter */
  87 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
  88
  89 static int scatter_elem_sz = SG_SCATTER_SZ;
  90 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
  91
  92 #define SG_SECTOR_SZ 512
  93
  94 static int sg_add_device(struct device *, struct class_interface *);
  95 static void sg_remove_device(struct device *, struct class_interface *);
  96
  97 static DEFINE_IDR(sg_index_idr);
  98 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
  99                                                            file descriptor list for device */
 100
 101 static struct class_interface sg_interface = {
 102         .add_dev        = sg_add_device,
 103         .remove_dev     = sg_remove_device,
 104 };
 105
 106 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 107         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 108         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 109         unsigned bufflen;       /* Size of (aggregate) data buffer */
 110         struct page **pages;
 111         int page_order;
 112         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 113         unsigned char cmd_opcode; /* first byte of command */
 114 } Sg_scatter_hold;
 115
 116 struct sg_device;               /* forward declarations */
 117 struct sg_fd;
 118
 119 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 120         struct list_head entry; /* list entry */
 121         struct sg_fd *parentfp; /* NULL -> not in use */
 122         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 123         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 124         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 125         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 126         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 127         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 128         /* done protected by rq_list_lock */
 129         char done;              /* 0->before bh, 1->before read, 2->read */
 130         struct request *rq;
 131         struct bio *bio;
 132         struct execute_work ew;
 133 } Sg_request;
 134
 135 typedef struct sg_fd {          /* holds the state of a file descriptor */
 136         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 137         struct sg_device *parentdp;     /* owning device */
 138         wait_queue_head_t read_wait;    /* queue read until command done */
 139         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 140         struct mutex f_mutex;   /* protect against changes in this fd */
 141         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 142         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 143         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 144         struct list_head rq_list; /* head of request list */
 145         struct fasync_struct *async_qp; /* used by asynchronous notification */
 146         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 147         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 148         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 149         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 150         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 151         char mmap_called;       /* 0 -> mmap() never called on this fd */
 152         char res_in_use;        /* 1 -> 'reserve' array in use */
 153         struct kref f_ref;
 154         struct execute_work ew;
 155 } Sg_fd;
 156
 157 typedef struct sg_device { /* holds the state of each scsi generic device */
 158         struct scsi_device *device;
 159         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 160         struct mutex open_rel_lock;     /* held when in open() or release() */
 161         int sg_tablesize;       /* adapter's max scatter-gather table size */
 162         u32 index;              /* device index number */
 163         struct list_head sfds;
 164         rwlock_t sfd_lock;      /* protect access to sfd list */
 165         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 166         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 167         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 168         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 169         struct gendisk *disk;
 170         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 171         struct kref d_ref;
 172 } Sg_device;
 173
 174 /* tasklet or soft irq callback */
 175 static void sg_rq_end_io(struct request *rq, blk_status_t status);
 176 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 177 static int sg_finish_rem_req(Sg_request * srp);
 178 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 179 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 180                            Sg_request * srp);
 181 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 182                         const char __user *buf, size_t count, int blocking,
 183                         int read_only, int sg_io_owned, Sg_request **o_srp);
 184 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 185                            unsigned char *cmnd, int timeout, int blocking);
 186 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 187 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 188 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 189 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 190 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 191 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 192 static void sg_remove_sfp(struct kref *);
 193 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 194 static Sg_request *sg_add_request(Sg_fd * sfp);
 195 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 196 static Sg_device *sg_get_dev(int dev);
 197 static void sg_device_destroy(struct kref *kref);
 198
 199 #define SZ_SG_HEADER sizeof(struct sg_header)
 200 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 201 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 202 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 203
 204 #define sg_printk(prefix, sdp, fmt, a...) \
 205         sdev_prefix_printk(prefix, (sdp)->device,               \
 206                            (sdp)->disk->disk_name, fmt, ##a)
 207
 208 /*
 209  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 210  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 211  * to trigger read() and write() calls from various contexts with elevated
 212  * privileges. This can lead to kernel memory corruption (e.g. if these
 213  * interfaces are called through splice()) and privilege escalation inside
 214  * userspace (e.g. if a process with access to such a device passes a file
 215  * descriptor to a SUID binary as stdin/stdout/stderr).
 216  *
 217  * This function provides protection for the legacy API by restricting the
 218  * calling context.
 219  */
 220 static int sg_check_file_access(struct file *filp, const char *caller)
 221 {
 222         if (filp->f_cred != current_real_cred()) {
 223                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 224                         caller, task_tgid_vnr(current), current->comm);
 225                 return -EPERM;
 226         }
 227         if (uaccess_kernel()) {
 228                 pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
 229                         caller, task_tgid_vnr(current), current->comm);
 230                 return -EACCES;
 231         }
 232         return 0;
 233 }
 234
 235 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 236 {
 237         struct sg_fd *sfp = filp->private_data;
 238
 239         if (sfp->parentdp->device->type == TYPE_SCANNER)
 240                 return 0;
 241
 242         return blk_verify_command(cmd, filp->f_mode);
 243 }
 244
 245 static int
 246 open_wait(Sg_device *sdp, int flags)
 247 {
 248         int retval = 0;
 249
 250         if (flags & O_EXCL) {
 251                 while (sdp->open_cnt > 0) {
 252                         mutex_unlock(&sdp->open_rel_lock);
 253                         retval = wait_event_interruptible(sdp->open_wait,
 254                                         (atomic_read(&sdp->detaching) ||
 255                                          !sdp->open_cnt));
 256                         mutex_lock(&sdp->open_rel_lock);
 257
 258                         if (retval) /* -ERESTARTSYS */
 259                                 return retval;
 260                         if (atomic_read(&sdp->detaching))
 261                                 return -ENODEV;
 262                 }
 263         } else {
 264                 while (sdp->exclude) {
 265                         mutex_unlock(&sdp->open_rel_lock);
 266                         retval = wait_event_interruptible(sdp->open_wait,
 267                                         (atomic_read(&sdp->detaching) ||
 268                                          !sdp->exclude));
 269                         mutex_lock(&sdp->open_rel_lock);
 270
 271                         if (retval) /* -ERESTARTSYS */
 272                                 return retval;
 273                         if (atomic_read(&sdp->detaching))
 274                                 return -ENODEV;
 275                 }
 276         }
 277
 278         return retval;
 279 }
 280
 281 /* Returns 0 on success, else a negated errno value */
 282 static int
 283 sg_open(struct inode *inode, struct file *filp)
 284 {
 285         int dev = iminor(inode);
 286         int flags = filp->f_flags;
 287         struct request_queue *q;
 288         Sg_device *sdp;
 289         Sg_fd *sfp;
 290         int retval;
 291
 292         nonseekable_open(inode, filp);
 293         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 294                 return -EPERM; /* Can't lock it with read only access */
 295         sdp = sg_get_dev(dev);
 296         if (IS_ERR(sdp))
 297                 return PTR_ERR(sdp);
 298
 299         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 300                                       "sg_open: flags=0x%x\n", flags));
 301
 302         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 303         /* Prevent the device driver from vanishing while we sleep */
 304         retval = scsi_device_get(sdp->device);
 305         if (retval)
 306                 goto sg_put;
 307
 308         retval = scsi_autopm_get_device(sdp->device);
 309         if (retval)
 310                 goto sdp_put;
 311
 312         /* scsi_block_when_processing_errors() may block so bypass
 313          * check if O_NONBLOCK. Permits SCSI commands to be issued
 314          * during error recovery. Tread carefully. */
 315         if (!((flags & O_NONBLOCK) ||
 316               scsi_block_when_processing_errors(sdp->device))) {
 317                 retval = -ENXIO;
 318                 /* we are in error recovery for this device */
 319                 goto error_out;
 320         }
 321
 322         mutex_lock(&sdp->open_rel_lock);
 323         if (flags & O_NONBLOCK) {
 324                 if (flags & O_EXCL) {
 325                         if (sdp->open_cnt > 0) {
 326                                 retval = -EBUSY;
 327                                 goto error_mutex_locked;
 328                         }
 329                 } else {
 330                         if (sdp->exclude) {
 331                                 retval = -EBUSY;
 332                                 goto error_mutex_locked;
 333                         }
 334                 }
 335         } else {
 336                 retval = open_wait(sdp, flags);
 337                 if (retval) /* -ERESTARTSYS or -ENODEV */
 338                         goto error_mutex_locked;
 339         }
 340
 341         /* N.B. at this point we are holding the open_rel_lock */
 342         if (flags & O_EXCL)
 343                 sdp->exclude = true;
 344
 345         if (sdp->open_cnt < 1) {  /* no existing opens */
 346                 sdp->sgdebug = 0;
 347                 q = sdp->device->request_queue;
 348                 sdp->sg_tablesize = queue_max_segments(q);
 349         }
 350         sfp = sg_add_sfp(sdp);
 351         if (IS_ERR(sfp)) {
 352                 retval = PTR_ERR(sfp);
 353                 goto out_undo;
 354         }
 355
 356         filp->private_data = sfp;
 357         sdp->open_cnt++;
 358         mutex_unlock(&sdp->open_rel_lock);
 359
 360         retval = 0;
 361 sg_put:
 362         kref_put(&sdp->d_ref, sg_device_destroy);
 363         return retval;
 364
 365 out_undo:
 366         if (flags & O_EXCL) {
 367                 sdp->exclude = false;   /* undo if error */
 368                 wake_up_interruptible(&sdp->open_wait);
 369         }
 370 error_mutex_locked:
 371         mutex_unlock(&sdp->open_rel_lock);
 372 error_out:
 373         scsi_autopm_put_device(sdp->device);
 374 sdp_put:
 375         scsi_device_put(sdp->device);
 376         goto sg_put;
 377 }
 378
 379 /* Release resources associated with a successful sg_open()
 380  * Returns 0 on success, else a negated errno value */
 381 static int
 382 sg_release(struct inode *inode, struct file *filp)
 383 {
 384         Sg_device *sdp;
 385         Sg_fd *sfp;
 386
 387         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 388                 return -ENXIO;
 389         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 390
 391         mutex_lock(&sdp->open_rel_lock);
 392         scsi_autopm_put_device(sdp->device);
 393         kref_put(&sfp->f_ref, sg_remove_sfp);
 394         sdp->open_cnt--;
 395
 396         /* possibly many open()s waiting on exlude clearing, start many;
 397          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 398         if (sdp->exclude) {
 399                 sdp->exclude = false;
 400                 wake_up_interruptible_all(&sdp->open_wait);
 401         } else if (0 == sdp->open_cnt) {
 402                 wake_up_interruptible(&sdp->open_wait);
 403         }
 404         mutex_unlock(&sdp->open_rel_lock);
 405         return 0;
 406 }
 407
 408 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 409 {
 410         struct sg_header __user *old_hdr = buf;
 411         int reply_len;
 412
 413         if (count >= SZ_SG_HEADER) {
 414                 /* negative reply_len means v3 format, otherwise v1/v2 */
 415                 if (get_user(reply_len, &old_hdr->reply_len))
 416                         return -EFAULT;
 417
 418                 if (reply_len >= 0)
 419                         return get_user(*pack_id, &old_hdr->pack_id);
 420
 421                 if (in_compat_syscall() &&
 422                     count >= sizeof(struct compat_sg_io_hdr)) {
 423                         struct compat_sg_io_hdr __user *hp = buf;
 424
 425                         return get_user(*pack_id, &hp->pack_id);
 426                 }
 427
 428                 if (count >= sizeof(struct sg_io_hdr)) {
 429                         struct sg_io_hdr __user *hp = buf;
 430
 431                         return get_user(*pack_id, &hp->pack_id);
 432                 }
 433         }
 434
 435         /* no valid header was passed, so ignore the pack_id */
 436         *pack_id = -1;
 437         return 0;
 438 }
 439
 440 static ssize_t
 441 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 442 {
 443         Sg_device *sdp;
 444         Sg_fd *sfp;
 445         Sg_request *srp;
 446         int req_pack_id = -1;
 447         sg_io_hdr_t *hp;
 448         struct sg_header *old_hdr;
 449         int retval;
 450
 451         /*
 452          * This could cause a response to be stranded. Close the associated
 453          * file descriptor to free up any resources being held.
 454          */
 455         retval = sg_check_file_access(filp, __func__);
 456         if (retval)
 457                 return retval;
 458
 459         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 460                 return -ENXIO;
 461         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 462                                       "sg_read: count=%d\n", (int) count));
 463
 464         if (sfp->force_packid)
 465                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 466         if (retval)
 467                 return retval;
 468
 469         srp = sg_get_rq_mark(sfp, req_pack_id);
 470         if (!srp) {             /* now wait on packet to arrive */
 471                 if (atomic_read(&sdp->detaching))
 472                         return -ENODEV;
 473                 if (filp->f_flags & O_NONBLOCK)
 474                         return -EAGAIN;
 475                 retval = wait_event_interruptible(sfp->read_wait,
 476                         (atomic_read(&sdp->detaching) ||
 477                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 478                 if (atomic_read(&sdp->detaching))
 479                         return -ENODEV;
 480                 if (retval)
 481                         /* -ERESTARTSYS as signal hit process */
 482                         return retval;
 483         }
 484         if (srp->header.interface_id != '\0')
 485                 return sg_new_read(sfp, buf, count, srp);
 486
 487         hp = &srp->header;
 488         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 489         if (!old_hdr)
 490                 return -ENOMEM;
 491
 492         old_hdr->reply_len = (int) hp->timeout;
 493         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 494         old_hdr->pack_id = hp->pack_id;
 495         old_hdr->twelve_byte =
 496             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 497         old_hdr->target_status = hp->masked_status;
 498         old_hdr->host_status = hp->host_status;
 499         old_hdr->driver_status = hp->driver_status;
 500         if ((CHECK_CONDITION & hp->masked_status) ||
 501             (DRIVER_SENSE & hp->driver_status))
 502                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 503                        sizeof (old_hdr->sense_buffer));
 504         switch (hp->host_status) {
 505         /* This setup of 'result' is for backward compatibility and is best
 506            ignored by the user who should use target, host + driver status */
 507         case DID_OK:
 508         case DID_PASSTHROUGH:
 509         case DID_SOFT_ERROR:
 510                 old_hdr->result = 0;
 511                 break;
 512         case DID_NO_CONNECT:
 513         case DID_BUS_BUSY:
 514         case DID_TIME_OUT:
 515                 old_hdr->result = EBUSY;
 516                 break;
 517         case DID_BAD_TARGET:
 518         case DID_ABORT:
 519         case DID_PARITY:
 520         case DID_RESET:
 521         case DID_BAD_INTR:
 522                 old_hdr->result = EIO;
 523                 break;
 524         case DID_ERROR:
 525                 old_hdr->result = (srp->sense_b[0] == 0 &&
 526                                   hp->masked_status == GOOD) ? 0 : EIO;
 527                 break;
 528         default:
 529                 old_hdr->result = EIO;
 530                 break;
 531         }
 532
 533         /* Now copy the result back to the user buffer.  */
 534         if (count >= SZ_SG_HEADER) {
 535                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 536                         retval = -EFAULT;
 537                         goto free_old_hdr;
 538                 }
 539                 buf += SZ_SG_HEADER;
 540                 if (count > old_hdr->reply_len)
 541                         count = old_hdr->reply_len;
 542                 if (count > SZ_SG_HEADER) {
 543                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 544                                 retval = -EFAULT;
 545                                 goto free_old_hdr;
 546                         }
 547                 }
 548         } else
 549                 count = (old_hdr->result == 0) ? 0 : -EIO;
 550         sg_finish_rem_req(srp);
 551         sg_remove_request(sfp, srp);
 552         retval = count;
 553 free_old_hdr:
 554         kfree(old_hdr);
 555         return retval;
 556 }
 557
 558 static ssize_t
 559 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 560 {
 561         sg_io_hdr_t *hp = &srp->header;
 562         int err = 0, err2;
 563         int len;
 564
 565         if (in_compat_syscall()) {
 566                 if (count < sizeof(struct compat_sg_io_hdr)) {
 567                         err = -EINVAL;
 568                         goto err_out;
 569                 }
 570         } else if (count < SZ_SG_IO_HDR) {
 571                 err = -EINVAL;
 572                 goto err_out;
 573         }
 574         hp->sb_len_wr = 0;
 575         if ((hp->mx_sb_len > 0) && hp->sbp) {
 576                 if ((CHECK_CONDITION & hp->masked_status) ||
 577                     (DRIVER_SENSE & hp->driver_status)) {
 578                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 579                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 580                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 581                         len = (len > sb_len) ? sb_len : len;
 582                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 583                                 err = -EFAULT;
 584                                 goto err_out;
 585                         }
 586                         hp->sb_len_wr = len;
 587                 }
 588         }
 589         if (hp->masked_status || hp->host_status || hp->driver_status)
 590                 hp->info |= SG_INFO_CHECK;
 591         err = put_sg_io_hdr(hp, buf);
 592 err_out:
 593         err2 = sg_finish_rem_req(srp);
 594         sg_remove_request(sfp, srp);
 595         return err ? : err2 ? : count;
 596 }
 597
 598 static ssize_t
 599 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 600 {
 601         int mxsize, cmd_size, k;
 602         int input_size, blocking;
 603         unsigned char opcode;
 604         Sg_device *sdp;
 605         Sg_fd *sfp;
 606         Sg_request *srp;
 607         struct sg_header old_hdr;
 608         sg_io_hdr_t *hp;
 609         unsigned char cmnd[SG_MAX_CDB_SIZE];
 610         int retval;
 611
 612         retval = sg_check_file_access(filp, __func__);
 613         if (retval)
 614                 return retval;
 615
 616         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 617                 return -ENXIO;
 618         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 619                                       "sg_write: count=%d\n", (int) count));
 620         if (atomic_read(&sdp->detaching))
 621                 return -ENODEV;
 622         if (!((filp->f_flags & O_NONBLOCK) ||
 623               scsi_block_when_processing_errors(sdp->device)))
 624                 return -ENXIO;
 625
 626         if (count < SZ_SG_HEADER)
 627                 return -EIO;
 628         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 629                 return -EFAULT;
 630         blocking = !(filp->f_flags & O_NONBLOCK);
 631         if (old_hdr.reply_len < 0)
 632                 return sg_new_write(sfp, filp, buf, count,
 633                                     blocking, 0, 0, NULL);
 634         if (count < (SZ_SG_HEADER + 6))
 635                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 636
 637         buf += SZ_SG_HEADER;
 638         if (get_user(opcode, buf))
 639                 return -EFAULT;
 640
 641         if (!(srp = sg_add_request(sfp))) {
 642                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 643                                               "sg_write: queue full\n"));
 644                 return -EDOM;
 645         }
 646         mutex_lock(&sfp->f_mutex);
 647         if (sfp->next_cmd_len > 0) {
 648                 cmd_size = sfp->next_cmd_len;
 649                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 650         } else {
 651                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 652                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 653                         cmd_size = 12;
 654         }
 655         mutex_unlock(&sfp->f_mutex);
 656         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 657                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 658 /* Determine buffer size.  */
 659         input_size = count - cmd_size;
 660         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 661         mxsize -= SZ_SG_HEADER;
 662         input_size -= SZ_SG_HEADER;
 663         if (input_size < 0) {
 664                 sg_remove_request(sfp, srp);
 665                 return -EIO;    /* User did not pass enough bytes for this command. */
 666         }
 667         hp = &srp->header;
 668         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 669         hp->cmd_len = (unsigned char) cmd_size;
 670         hp->iovec_count = 0;
 671         hp->mx_sb_len = 0;
 672         if (input_size > 0)
 673                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 674                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 675         else
 676                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 677         hp->dxfer_len = mxsize;
 678         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 679             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 680                 hp->dxferp = (char __user *)buf + cmd_size;
 681         else
 682                 hp->dxferp = NULL;
 683         hp->sbp = NULL;
 684         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 685         hp->flags = input_size; /* structure abuse ... */
 686         hp->pack_id = old_hdr.pack_id;
 687         hp->usr_ptr = NULL;
 688         if (copy_from_user(cmnd, buf, cmd_size))
 689                 return -EFAULT;
 690         /*
 691          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 692          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 693          * is a non-zero input_size, so emit a warning.
 694          */
 695         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 696                 printk_ratelimited(KERN_WARNING
 697                                    "sg_write: data in/out %d/%d bytes "
 698                                    "for SCSI command 0x%x-- guessing "
 699                                    "data in;\n   program %s not setting "
 700                                    "count and/or reply_len properly\n",
 701                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 702                                    input_size, (unsigned int) cmnd[0],
 703                                    current->comm);
 704         }
 705         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 706         return (k < 0) ? k : count;
 707 }
 708
 709 static ssize_t
 710 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 711                  size_t count, int blocking, int read_only, int sg_io_owned,
 712                  Sg_request **o_srp)
 713 {
 714         int k;
 715         Sg_request *srp;
 716         sg_io_hdr_t *hp;
 717         unsigned char cmnd[SG_MAX_CDB_SIZE];
 718         int timeout;
 719         unsigned long ul_timeout;
 720
 721         if (count < SZ_SG_IO_HDR)
 722                 return -EINVAL;
 723
 724         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 725         if (!(srp = sg_add_request(sfp))) {
 726                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 727                                               "sg_new_write: queue full\n"));
 728                 return -EDOM;
 729         }
 730         srp->sg_io_owned = sg_io_owned;
 731         hp = &srp->header;
 732         if (get_sg_io_hdr(hp, buf)) {
 733                 sg_remove_request(sfp, srp);
 734                 return -EFAULT;
 735         }
 736         if (hp->interface_id != 'S') {
 737                 sg_remove_request(sfp, srp);
 738                 return -ENOSYS;
 739         }
 740         if (hp->flags & SG_FLAG_MMAP_IO) {
 741                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 742                         sg_remove_request(sfp, srp);
 743                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 744                 }
 745                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 746                         sg_remove_request(sfp, srp);
 747                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 748                 }
 749                 if (sfp->res_in_use) {
 750                         sg_remove_request(sfp, srp);
 751                         return -EBUSY;  /* reserve buffer already being used */
 752                 }
 753         }
 754         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 755         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 756         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 757                 sg_remove_request(sfp, srp);
 758                 return -EMSGSIZE;
 759         }
 760         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 761                 sg_remove_request(sfp, srp);
 762                 return -EFAULT;
 763         }
 764         if (read_only && sg_allow_access(file, cmnd)) {
 765                 sg_remove_request(sfp, srp);
 766                 return -EPERM;
 767         }
 768         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 769         if (k < 0)
 770                 return k;
 771         if (o_srp)
 772                 *o_srp = srp;
 773         return count;
 774 }
 775
 776 static int
 777 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 778                 unsigned char *cmnd, int timeout, int blocking)
 779 {
 780         int k, at_head;
 781         Sg_device *sdp = sfp->parentdp;
 782         sg_io_hdr_t *hp = &srp->header;
 783
 784         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 785         hp->status = 0;
 786         hp->masked_status = 0;
 787         hp->msg_status = 0;
 788         hp->info = 0;
 789         hp->host_status = 0;
 790         hp->driver_status = 0;
 791         hp->resid = 0;
 792         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 793                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 794                         (int) cmnd[0], (int) hp->cmd_len));
 795
 796         if (hp->dxfer_len >= SZ_256M) {
 797                 sg_remove_request(sfp, srp);
 798                 return -EINVAL;
 799         }
 800
 801         k = sg_start_req(srp, cmnd);
 802         if (k) {
 803                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 804                         "sg_common_write: start_req err=%d\n", k));
 805                 sg_finish_rem_req(srp);
 806                 sg_remove_request(sfp, srp);
 807                 return k;       /* probably out of space --> ENOMEM */
 808         }
 809         if (atomic_read(&sdp->detaching)) {
 810                 if (srp->bio) {
 811                         scsi_req_free_cmd(scsi_req(srp->rq));
 812                         blk_put_request(srp->rq);
 813                         srp->rq = NULL;
 814                 }
 815
 816                 sg_finish_rem_req(srp);
 817                 sg_remove_request(sfp, srp);
 818                 return -ENODEV;
 819         }
 820
 821         hp->duration = jiffies_to_msecs(jiffies);
 822         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 823             (SG_FLAG_Q_AT_TAIL & hp->flags))
 824                 at_head = 0;
 825         else
 826                 at_head = 1;
 827
 828         srp->rq->timeout = timeout;
 829         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 830         blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
 831                               srp->rq, at_head, sg_rq_end_io);
 832         return 0;
 833 }
 834
 835 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 836 {
 837         unsigned long flags;
 838         int ret;
 839
 840         read_lock_irqsave(&sfp->rq_list_lock, flags);
 841         ret = srp->done;
 842         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 843         return ret;
 844 }
 845
 846 static int max_sectors_bytes(struct request_queue *q)
 847 {
 848         unsigned int max_sectors = queue_max_sectors(q);
 849
 850         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 851
 852         return max_sectors << 9;
 853 }
 854
 855 static void
 856 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 857 {
 858         Sg_request *srp;
 859         int val;
 860         unsigned int ms;
 861
 862         val = 0;
 863         list_for_each_entry(srp, &sfp->rq_list, entry) {
 864                 if (val >= SG_MAX_QUEUE)
 865                         break;
 866                 rinfo[val].req_state = srp->done + 1;
 867                 rinfo[val].problem =
 868                         srp->header.masked_status &
 869                         srp->header.host_status &
 870                         srp->header.driver_status;
 871                 if (srp->done)
 872                         rinfo[val].duration =
 873                                 srp->header.duration;
 874                 else {
 875                         ms = jiffies_to_msecs(jiffies);
 876                         rinfo[val].duration =
 877                                 (ms > srp->header.duration) ?
 878                                 (ms - srp->header.duration) : 0;
 879                 }
 880                 rinfo[val].orphan = srp->orphan;
 881                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 882                 rinfo[val].pack_id = srp->header.pack_id;
 883                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 884                 val++;
 885         }
 886 }
 887
 888 #ifdef CONFIG_COMPAT
 889 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 890         char req_state;
 891         char orphan;
 892         char sg_io_owned;
 893         char problem;
 894         int pack_id;
 895         compat_uptr_t usr_ptr;
 896         unsigned int duration;
 897         int unused;
 898 };
 899
 900 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 901                                     struct sg_req_info *rinfo)
 902 {
 903         int i;
 904         for (i = 0; i < SG_MAX_QUEUE; i++) {
 905                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 906                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 907                     put_user(rinfo[i].duration, &o[i].duration) ||
 908                     put_user(rinfo[i].unused, &o[i].unused))
 909                         return -EFAULT;
 910         }
 911         return 0;
 912 }
 913 #endif
 914
 915 static long
 916 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 917                 unsigned int cmd_in, void __user *p)
 918 {
 919         int __user *ip = p;
 920         int result, val, read_only;
 921         Sg_request *srp;
 922         unsigned long iflags;
 923
 924         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 925                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 926         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 927
 928         switch (cmd_in) {
 929         case SG_IO:
 930                 if (atomic_read(&sdp->detaching))
 931                         return -ENODEV;
 932                 if (!scsi_block_when_processing_errors(sdp->device))
 933                         return -ENXIO;
 934                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 935                                  1, read_only, 1, &srp);
 936                 if (result < 0)
 937                         return result;
 938                 result = wait_event_interruptible(sfp->read_wait,
 939                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 940                 if (atomic_read(&sdp->detaching))
 941                         return -ENODEV;
 942                 write_lock_irq(&sfp->rq_list_lock);
 943                 if (srp->done) {
 944                         srp->done = 2;
 945                         write_unlock_irq(&sfp->rq_list_lock);
 946                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 947                         return (result < 0) ? result : 0;
 948                 }
 949                 srp->orphan = 1;
 950                 write_unlock_irq(&sfp->rq_list_lock);
 951                 return result;  /* -ERESTARTSYS because signal hit process */
 952         case SG_SET_TIMEOUT:
 953                 result = get_user(val, ip);
 954                 if (result)
 955                         return result;
 956                 if (val < 0)
 957                         return -EIO;
 958                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 959                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 960                                     INT_MAX);
 961                 sfp->timeout_user = val;
 962                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 963
 964                 return 0;
 965         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 966                                 /* strange ..., for backward compatibility */
 967                 return sfp->timeout_user;
 968         case SG_SET_FORCE_LOW_DMA:
 969                 /*
 970                  * N.B. This ioctl never worked properly, but failed to
 971                  * return an error value. So returning '0' to keep compability
 972                  * with legacy applications.
 973                  */
 974                 return 0;
 975         case SG_GET_LOW_DMA:
 976                 return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
 977         case SG_GET_SCSI_ID:
 978                 {
 979                         sg_scsi_id_t v;
 980
 981                         if (atomic_read(&sdp->detaching))
 982                                 return -ENODEV;
 983                         memset(&v, 0, sizeof(v));
 984                         v.host_no = sdp->device->host->host_no;
 985                         v.channel = sdp->device->channel;
 986                         v.scsi_id = sdp->device->id;
 987                         v.lun = sdp->device->lun;
 988                         v.scsi_type = sdp->device->type;
 989                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 990                         v.d_queue_depth = sdp->device->queue_depth;
 991                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 992                                 return -EFAULT;
 993                         return 0;
 994                 }
 995         case SG_SET_FORCE_PACK_ID:
 996                 result = get_user(val, ip);
 997                 if (result)
 998                         return result;
 999                 sfp->force_packid = val ? 1 : 0;
1000                 return 0;
1001         case SG_GET_PACK_ID:
1002                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1003                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1004                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1005                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1006                                                        iflags);
1007                                 return put_user(srp->header.pack_id, ip);
1008                         }
1009                 }
1010                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1011                 return put_user(-1, ip);
1012         case SG_GET_NUM_WAITING:
1013                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1014                 val = 0;
1015                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1016                         if ((1 == srp->done) && (!srp->sg_io_owned))
1017                                 ++val;
1018                 }
1019                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1020                 return put_user(val, ip);
1021         case SG_GET_SG_TABLESIZE:
1022                 return put_user(sdp->sg_tablesize, ip);
1023         case SG_SET_RESERVED_SIZE:
1024                 result = get_user(val, ip);
1025                 if (result)
1026                         return result;
1027                 if (val < 0)
1028                         return -EINVAL;
1029                 val = min_t(int, val,
1030                             max_sectors_bytes(sdp->device->request_queue));
1031                 mutex_lock(&sfp->f_mutex);
1032                 if (val != sfp->reserve.bufflen) {
1033                         if (sfp->mmap_called ||
1034                             sfp->res_in_use) {
1035                                 mutex_unlock(&sfp->f_mutex);
1036                                 return -EBUSY;
1037                         }
1038
1039                         sg_remove_scat(sfp, &sfp->reserve);
1040                         sg_build_reserve(sfp, val);
1041                 }
1042                 mutex_unlock(&sfp->f_mutex);
1043                 return 0;
1044         case SG_GET_RESERVED_SIZE:
1045                 val = min_t(int, sfp->reserve.bufflen,
1046                             max_sectors_bytes(sdp->device->request_queue));
1047                 return put_user(val, ip);
1048         case SG_SET_COMMAND_Q:
1049                 result = get_user(val, ip);
1050                 if (result)
1051                         return result;
1052                 sfp->cmd_q = val ? 1 : 0;
1053                 return 0;
1054         case SG_GET_COMMAND_Q:
1055                 return put_user((int) sfp->cmd_q, ip);
1056         case SG_SET_KEEP_ORPHAN:
1057                 result = get_user(val, ip);
1058                 if (result)
1059                         return result;
1060                 sfp->keep_orphan = val;
1061                 return 0;
1062         case SG_GET_KEEP_ORPHAN:
1063                 return put_user((int) sfp->keep_orphan, ip);
1064         case SG_NEXT_CMD_LEN:
1065                 result = get_user(val, ip);
1066                 if (result)
1067                         return result;
1068                 if (val > SG_MAX_CDB_SIZE)
1069                         return -ENOMEM;
1070                 sfp->next_cmd_len = (val > 0) ? val : 0;
1071                 return 0;
1072         case SG_GET_VERSION_NUM:
1073                 return put_user(sg_version_num, ip);
1074         case SG_GET_ACCESS_COUNT:
1075                 /* faked - we don't have a real access count anymore */
1076                 val = (sdp->device ? 1 : 0);
1077                 return put_user(val, ip);
1078         case SG_GET_REQUEST_TABLE:
1079                 {
1080                         sg_req_info_t *rinfo;
1081
1082                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1083                                         GFP_KERNEL);
1084                         if (!rinfo)
1085                                 return -ENOMEM;
1086                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1087                         sg_fill_request_table(sfp, rinfo);
1088                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1089         #ifdef CONFIG_COMPAT
1090                         if (in_compat_syscall())
1091                                 result = put_compat_request_table(p, rinfo);
1092                         else
1093         #endif
1094                                 result = copy_to_user(p, rinfo,
1095                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1096                         result = result ? -EFAULT : 0;
1097                         kfree(rinfo);
1098                         return result;
1099                 }
1100         case SG_EMULATED_HOST:
1101                 if (atomic_read(&sdp->detaching))
1102                         return -ENODEV;
1103                 return put_user(sdp->device->host->hostt->emulated, ip);
1104         case SCSI_IOCTL_SEND_COMMAND:
1105                 if (atomic_read(&sdp->detaching))
1106                         return -ENODEV;
1107                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1108         case SG_SET_DEBUG:
1109                 result = get_user(val, ip);
1110                 if (result)
1111                         return result;
1112                 sdp->sgdebug = (char) val;
1113                 return 0;
1114         case BLKSECTGET:
1115                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1116                                 ip);
1117         case BLKTRACESETUP:
1118                 return blk_trace_setup(sdp->device->request_queue,
1119                                        sdp->disk->disk_name,
1120                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1121                                        NULL, p);
1122         case BLKTRACESTART:
1123                 return blk_trace_startstop(sdp->device->request_queue, 1);
1124         case BLKTRACESTOP:
1125                 return blk_trace_startstop(sdp->device->request_queue, 0);
1126         case BLKTRACETEARDOWN:
1127                 return blk_trace_remove(sdp->device->request_queue);
1128         case SCSI_IOCTL_GET_IDLUN:
1129         case SCSI_IOCTL_GET_BUS_NUMBER:
1130         case SCSI_IOCTL_PROBE_HOST:
1131         case SG_GET_TRANSFORM:
1132         case SG_SCSI_RESET:
1133                 if (atomic_read(&sdp->detaching))
1134                         return -ENODEV;
1135                 break;
1136         default:
1137                 if (read_only)
1138                         return -EPERM;  /* don't know so take safe approach */
1139                 break;
1140         }
1141
1142         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1143                         cmd_in, filp->f_flags & O_NDELAY);
1144         if (result)
1145                 return result;
1146
1147         return -ENOIOCTLCMD;
1148 }
1149
1150 static long
1151 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1152 {
1153         void __user *p = (void __user *)arg;
1154         Sg_device *sdp;
1155         Sg_fd *sfp;
1156         int ret;
1157
1158         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1159                 return -ENXIO;
1160
1161         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1162         if (ret != -ENOIOCTLCMD)
1163                 return ret;
1164
1165         return scsi_ioctl(sdp->device, cmd_in, p);
1166 }
1167
1168 #ifdef CONFIG_COMPAT
1169 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1170 {
1171         void __user *p = compat_ptr(arg);
1172         Sg_device *sdp;
1173         Sg_fd *sfp;
1174         int ret;
1175
1176         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1177                 return -ENXIO;
1178
1179         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1180         if (ret != -ENOIOCTLCMD)
1181                 return ret;
1182
1183         return scsi_compat_ioctl(sdp->device, cmd_in, p);
1184 }
1185 #endif
1186
1187 static __poll_t
1188 sg_poll(struct file *filp, poll_table * wait)
1189 {
1190         __poll_t res = 0;
1191         Sg_device *sdp;
1192         Sg_fd *sfp;
1193         Sg_request *srp;
1194         int count = 0;
1195         unsigned long iflags;
1196
1197         sfp = filp->private_data;
1198         if (!sfp)
1199                 return EPOLLERR;
1200         sdp = sfp->parentdp;
1201         if (!sdp)
1202                 return EPOLLERR;
1203         poll_wait(filp, &sfp->read_wait, wait);
1204         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1205         list_for_each_entry(srp, &sfp->rq_list, entry) {
1206                 /* if any read waiting, flag it */
1207                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1208                         res = EPOLLIN | EPOLLRDNORM;
1209                 ++count;
1210         }
1211         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1212
1213         if (atomic_read(&sdp->detaching))
1214                 res |= EPOLLHUP;
1215         else if (!sfp->cmd_q) {
1216                 if (0 == count)
1217                         res |= EPOLLOUT | EPOLLWRNORM;
1218         } else if (count < SG_MAX_QUEUE)
1219                 res |= EPOLLOUT | EPOLLWRNORM;
1220         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1221                                       "sg_poll: res=0x%x\n", (__force u32) res));
1222         return res;
1223 }
1224
1225 static int
1226 sg_fasync(int fd, struct file *filp, int mode)
1227 {
1228         Sg_device *sdp;
1229         Sg_fd *sfp;
1230
1231         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1232                 return -ENXIO;
1233         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1234                                       "sg_fasync: mode=%d\n", mode));
1235
1236         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1237 }
1238
1239 static vm_fault_t
1240 sg_vma_fault(struct vm_fault *vmf)
1241 {
1242         struct vm_area_struct *vma = vmf->vma;
1243         Sg_fd *sfp;
1244         unsigned long offset, len, sa;
1245         Sg_scatter_hold *rsv_schp;
1246         int k, length;
1247
1248         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1249                 return VM_FAULT_SIGBUS;
1250         rsv_schp = &sfp->reserve;
1251         offset = vmf->pgoff << PAGE_SHIFT;
1252         if (offset >= rsv_schp->bufflen)
1253                 return VM_FAULT_SIGBUS;
1254         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1255                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1256                                       offset, rsv_schp->k_use_sg));
1257         sa = vma->vm_start;
1258         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1259         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1260                 len = vma->vm_end - sa;
1261                 len = (len < length) ? len : length;
1262                 if (offset < len) {
1263                         struct page *page = nth_page(rsv_schp->pages[k],
1264                                                      offset >> PAGE_SHIFT);
1265                         get_page(page); /* increment page count */
1266                         vmf->page = page;
1267                         return 0; /* success */
1268                 }
1269                 sa += len;
1270                 offset -= len;
1271         }
1272
1273         return VM_FAULT_SIGBUS;
1274 }
1275
1276 static const struct vm_operations_struct sg_mmap_vm_ops = {
1277         .fault = sg_vma_fault,
1278 };
1279
1280 static int
1281 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1282 {
1283         Sg_fd *sfp;
1284         unsigned long req_sz, len, sa;
1285         Sg_scatter_hold *rsv_schp;
1286         int k, length;
1287         int ret = 0;
1288
1289         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1290                 return -ENXIO;
1291         req_sz = vma->vm_end - vma->vm_start;
1292         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1293                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1294                                       (void *) vma->vm_start, (int) req_sz));
1295         if (vma->vm_pgoff)
1296                 return -EINVAL; /* want no offset */
1297         rsv_schp = &sfp->reserve;
1298         mutex_lock(&sfp->f_mutex);
1299         if (req_sz > rsv_schp->bufflen) {
1300                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1301                 goto out;
1302         }
1303
1304         sa = vma->vm_start;
1305         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1306         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1307                 len = vma->vm_end - sa;
1308                 len = (len < length) ? len : length;
1309                 sa += len;
1310         }
1311
1312         sfp->mmap_called = 1;
1313         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1314         vma->vm_private_data = sfp;
1315         vma->vm_ops = &sg_mmap_vm_ops;
1316 out:
1317         mutex_unlock(&sfp->f_mutex);
1318         return ret;
1319 }
1320
1321 static void
1322 sg_rq_end_io_usercontext(struct work_struct *work)
1323 {
1324         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1325         struct sg_fd *sfp = srp->parentfp;
1326
1327         sg_finish_rem_req(srp);
1328         sg_remove_request(sfp, srp);
1329         kref_put(&sfp->f_ref, sg_remove_sfp);
1330 }
1331
1332 /*
1333  * This function is a "bottom half" handler that is called by the mid
1334  * level when a command is completed (or has failed).
1335  */
1336 static void
1337 sg_rq_end_io(struct request *rq, blk_status_t status)
1338 {
1339         struct sg_request *srp = rq->end_io_data;
1340         struct scsi_request *req = scsi_req(rq);
1341         Sg_device *sdp;
1342         Sg_fd *sfp;
1343         unsigned long iflags;
1344         unsigned int ms;
1345         char *sense;
1346         int result, resid, done = 1;
1347
1348         if (WARN_ON(srp->done != 0))
1349                 return;
1350
1351         sfp = srp->parentfp;
1352         if (WARN_ON(sfp == NULL))
1353                 return;
1354
1355         sdp = sfp->parentdp;
1356         if (unlikely(atomic_read(&sdp->detaching)))
1357                 pr_info("%s: device detaching\n", __func__);
1358
1359         sense = req->sense;
1360         result = req->result;
1361         resid = req->resid_len;
1362
1363         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1364                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1365                                       srp->header.pack_id, result));
1366         srp->header.resid = resid;
1367         ms = jiffies_to_msecs(jiffies);
1368         srp->header.duration = (ms > srp->header.duration) ?
1369                                 (ms - srp->header.duration) : 0;
1370         if (0 != result) {
1371                 struct scsi_sense_hdr sshdr;
1372
1373                 srp->header.status = 0xff & result;
1374                 srp->header.masked_status = status_byte(result);
1375                 srp->header.msg_status = msg_byte(result);
1376                 srp->header.host_status = host_byte(result);
1377                 srp->header.driver_status = driver_byte(result);
1378                 if ((sdp->sgdebug > 0) &&
1379                     ((CHECK_CONDITION == srp->header.masked_status) ||
1380                      (COMMAND_TERMINATED == srp->header.masked_status)))
1381                         __scsi_print_sense(sdp->device, __func__, sense,
1382                                            SCSI_SENSE_BUFFERSIZE);
1383
1384                 /* Following if statement is a patch supplied by Eric Youngdale */
1385                 if (driver_byte(result) != 0
1386                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1387                     && !scsi_sense_is_deferred(&sshdr)
1388                     && sshdr.sense_key == UNIT_ATTENTION
1389                     && sdp->device->removable) {
1390                         /* Detected possible disc change. Set the bit - this */
1391                         /* may be used if there are filesystems using this device */
1392                         sdp->device->changed = 1;
1393                 }
1394         }
1395
1396         if (req->sense_len)
1397                 memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1398
1399         /* Rely on write phase to clean out srp status values, so no "else" */
1400
1401         /*
1402          * Free the request as soon as it is complete so that its resources
1403          * can be reused without waiting for userspace to read() the
1404          * result.  But keep the associated bio (if any) around until
1405          * blk_rq_unmap_user() can be called from user context.
1406          */
1407         srp->rq = NULL;
1408         scsi_req_free_cmd(scsi_req(rq));
1409         blk_put_request(rq);
1410
1411         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1412         if (unlikely(srp->orphan)) {
1413                 if (sfp->keep_orphan)
1414                         srp->sg_io_owned = 0;
1415                 else
1416                         done = 0;
1417         }
1418         srp->done = done;
1419         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1420
1421         if (likely(done)) {
1422                 /* Now wake up any sg_read() that is waiting for this
1423                  * packet.
1424                  */
1425                 wake_up_interruptible(&sfp->read_wait);
1426                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1427                 kref_put(&sfp->f_ref, sg_remove_sfp);
1428         } else {
1429                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1430                 schedule_work(&srp->ew.work);
1431         }
1432 }
1433
1434 static const struct file_operations sg_fops = {
1435         .owner = THIS_MODULE,
1436         .read = sg_read,
1437         .write = sg_write,
1438         .poll = sg_poll,
1439         .unlocked_ioctl = sg_ioctl,
1440 #ifdef CONFIG_COMPAT
1441         .compat_ioctl = sg_compat_ioctl,
1442 #endif
1443         .open = sg_open,
1444         .mmap = sg_mmap,
1445         .release = sg_release,
1446         .fasync = sg_fasync,
1447         .llseek = no_llseek,
1448 };
1449
1450 static struct class *sg_sysfs_class;
1451
1452 static int sg_sysfs_valid = 0;
1453
1454 static Sg_device *
1455 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1456 {
1457         struct request_queue *q = scsidp->request_queue;
1458         Sg_device *sdp;
1459         unsigned long iflags;
1460         int error;
1461         u32 k;
1462
1463         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1464         if (!sdp) {
1465                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1466                             "failure\n", __func__);
1467                 return ERR_PTR(-ENOMEM);
1468         }
1469
1470         idr_preload(GFP_KERNEL);
1471         write_lock_irqsave(&sg_index_lock, iflags);
1472
1473         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1474         if (error < 0) {
1475                 if (error == -ENOSPC) {
1476                         sdev_printk(KERN_WARNING, scsidp,
1477                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1478                                     scsidp->type, SG_MAX_DEVS - 1);
1479                         error = -ENODEV;
1480                 } else {
1481                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1482                                     "allocation Sg_device failure: %d\n",
1483                                     __func__, error);
1484                 }
1485                 goto out_unlock;
1486         }
1487         k = error;
1488
1489         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1490                                         "sg_alloc: dev=%d \n", k));
1491         sprintf(disk->disk_name, "sg%d", k);
1492         disk->first_minor = k;
1493         sdp->disk = disk;
1494         sdp->device = scsidp;
1495         mutex_init(&sdp->open_rel_lock);
1496         INIT_LIST_HEAD(&sdp->sfds);
1497         init_waitqueue_head(&sdp->open_wait);
1498         atomic_set(&sdp->detaching, 0);
1499         rwlock_init(&sdp->sfd_lock);
1500         sdp->sg_tablesize = queue_max_segments(q);
1501         sdp->index = k;
1502         kref_init(&sdp->d_ref);
1503         error = 0;
1504
1505 out_unlock:
1506         write_unlock_irqrestore(&sg_index_lock, iflags);
1507         idr_preload_end();
1508
1509         if (error) {
1510                 kfree(sdp);
1511                 return ERR_PTR(error);
1512         }
1513         return sdp;
1514 }
1515
1516 static int
1517 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1518 {
1519         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1520         struct gendisk *disk;
1521         Sg_device *sdp = NULL;
1522         struct cdev * cdev = NULL;
1523         int error;
1524         unsigned long iflags;
1525
1526         disk = alloc_disk(1);
1527         if (!disk) {
1528                 pr_warn("%s: alloc_disk failed\n", __func__);
1529                 return -ENOMEM;
1530         }
1531         disk->major = SCSI_GENERIC_MAJOR;
1532
1533         error = -ENOMEM;
1534         cdev = cdev_alloc();
1535         if (!cdev) {
1536                 pr_warn("%s: cdev_alloc failed\n", __func__);
1537                 goto out;
1538         }
1539         cdev->owner = THIS_MODULE;
1540         cdev->ops = &sg_fops;
1541
1542         sdp = sg_alloc(disk, scsidp);
1543         if (IS_ERR(sdp)) {
1544                 pr_warn("%s: sg_alloc failed\n", __func__);
1545                 error = PTR_ERR(sdp);
1546                 goto out;
1547         }
1548
1549         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1550         if (error)
1551                 goto cdev_add_err;
1552
1553         sdp->cdev = cdev;
1554         if (sg_sysfs_valid) {
1555                 struct device *sg_class_member;
1556
1557                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1558                                                 MKDEV(SCSI_GENERIC_MAJOR,
1559                                                       sdp->index),
1560                                                 sdp, "%s", disk->disk_name);
1561                 if (IS_ERR(sg_class_member)) {
1562                         pr_err("%s: device_create failed\n", __func__);
1563                         error = PTR_ERR(sg_class_member);
1564                         goto cdev_add_err;
1565                 }
1566                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1567                                           &sg_class_member->kobj, "generic");
1568                 if (error)
1569                         pr_err("%s: unable to make symlink 'generic' back "
1570                                "to sg%d\n", __func__, sdp->index);
1571         } else
1572                 pr_warn("%s: sg_sys Invalid\n", __func__);
1573
1574         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1575                     "type %d\n", sdp->index, scsidp->type);
1576
1577         dev_set_drvdata(cl_dev, sdp);
1578
1579         return 0;
1580
1581 cdev_add_err:
1582         write_lock_irqsave(&sg_index_lock, iflags);
1583         idr_remove(&sg_index_idr, sdp->index);
1584         write_unlock_irqrestore(&sg_index_lock, iflags);
1585         kfree(sdp);
1586
1587 out:
1588         put_disk(disk);
1589         if (cdev)
1590                 cdev_del(cdev);
1591         return error;
1592 }
1593
1594 static void
1595 sg_device_destroy(struct kref *kref)
1596 {
1597         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1598         unsigned long flags;
1599
1600         /* CAUTION!  Note that the device can still be found via idr_find()
1601          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1602          * any other cleanup.
1603          */
1604
1605         write_lock_irqsave(&sg_index_lock, flags);
1606         idr_remove(&sg_index_idr, sdp->index);
1607         write_unlock_irqrestore(&sg_index_lock, flags);
1608
1609         SCSI_LOG_TIMEOUT(3,
1610                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1611
1612         put_disk(sdp->disk);
1613         kfree(sdp);
1614 }
1615
1616 static void
1617 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1618 {
1619         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1620         Sg_device *sdp = dev_get_drvdata(cl_dev);
1621         unsigned long iflags;
1622         Sg_fd *sfp;
1623         int val;
1624
1625         if (!sdp)
1626                 return;
1627         /* want sdp->detaching non-zero as soon as possible */
1628         val = atomic_inc_return(&sdp->detaching);
1629         if (val > 1)
1630                 return; /* only want to do following once per device */
1631
1632         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1633                                       "%s\n", __func__));
1634
1635         read_lock_irqsave(&sdp->sfd_lock, iflags);
1636         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1637                 wake_up_interruptible_all(&sfp->read_wait);
1638                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1639         }
1640         wake_up_interruptible_all(&sdp->open_wait);
1641         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1642
1643         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1644         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1645         cdev_del(sdp->cdev);
1646         sdp->cdev = NULL;
1647
1648         kref_put(&sdp->d_ref, sg_device_destroy);
1649 }
1650
1651 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1652 module_param_named(def_reserved_size, def_reserved_size, int,
1653                    S_IRUGO | S_IWUSR);
1654 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1655
1656 MODULE_AUTHOR("Douglas Gilbert");
1657 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1658 MODULE_LICENSE("GPL");
1659 MODULE_VERSION(SG_VERSION_STR);
1660 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1661
1662 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1663                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1664 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1665 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1666
1667 static int __init
1668 init_sg(void)
1669 {
1670         int rc;
1671
1672         if (scatter_elem_sz < PAGE_SIZE) {
1673                 scatter_elem_sz = PAGE_SIZE;
1674                 scatter_elem_sz_prev = scatter_elem_sz;
1675         }
1676         if (def_reserved_size >= 0)
1677                 sg_big_buff = def_reserved_size;
1678         else
1679                 def_reserved_size = sg_big_buff;
1680
1681         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1682                                     SG_MAX_DEVS, "sg");
1683         if (rc)
1684                 return rc;
1685         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1686         if ( IS_ERR(sg_sysfs_class) ) {
1687                 rc = PTR_ERR(sg_sysfs_class);
1688                 goto err_out;
1689         }
1690         sg_sysfs_valid = 1;
1691         rc = scsi_register_interface(&sg_interface);
1692         if (0 == rc) {
1693 #ifdef CONFIG_SCSI_PROC_FS
1694                 sg_proc_init();
1695 #endif                          /* CONFIG_SCSI_PROC_FS */
1696                 return 0;
1697         }
1698         class_destroy(sg_sysfs_class);
1699 err_out:
1700         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1701         return rc;
1702 }
1703
1704 static void __exit
1705 exit_sg(void)
1706 {
1707 #ifdef CONFIG_SCSI_PROC_FS
1708         remove_proc_subtree("scsi/sg", NULL);
1709 #endif                          /* CONFIG_SCSI_PROC_FS */
1710         scsi_unregister_interface(&sg_interface);
1711         class_destroy(sg_sysfs_class);
1712         sg_sysfs_valid = 0;
1713         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1714                                  SG_MAX_DEVS);
1715         idr_destroy(&sg_index_idr);
1716 }
1717
1718 static int
1719 sg_start_req(Sg_request *srp, unsigned char *cmd)
1720 {
1721         int res;
1722         struct request *rq;
1723         struct scsi_request *req;
1724         Sg_fd *sfp = srp->parentfp;
1725         sg_io_hdr_t *hp = &srp->header;
1726         int dxfer_len = (int) hp->dxfer_len;
1727         int dxfer_dir = hp->dxfer_direction;
1728         unsigned int iov_count = hp->iovec_count;
1729         Sg_scatter_hold *req_schp = &srp->data;
1730         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1731         struct request_queue *q = sfp->parentdp->device->request_queue;
1732         struct rq_map_data *md, map_data;
1733         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1734         unsigned char *long_cmdp = NULL;
1735
1736         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1737                                       "sg_start_req: dxfer_len=%d\n",
1738                                       dxfer_len));
1739
1740         if (hp->cmd_len > BLK_MAX_CDB) {
1741                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1742                 if (!long_cmdp)
1743                         return -ENOMEM;
1744         }
1745
1746         /*
1747          * NOTE
1748          *
1749          * With scsi-mq enabled, there are a fixed number of preallocated
1750          * requests equal in number to shost->can_queue.  If all of the
1751          * preallocated requests are already in use, then blk_get_request()
1752          * will sleep until an active command completes, freeing up a request.
1753          * Although waiting in an asynchronous interface is less than ideal, we
1754          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1755          * not expect an EWOULDBLOCK from this condition.
1756          */
1757         rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1758                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, 0);
1759         if (IS_ERR(rq)) {
1760                 kfree(long_cmdp);
1761                 return PTR_ERR(rq);
1762         }
1763         req = scsi_req(rq);
1764
1765         if (hp->cmd_len > BLK_MAX_CDB)
1766                 req->cmd = long_cmdp;
1767         memcpy(req->cmd, cmd, hp->cmd_len);
1768         req->cmd_len = hp->cmd_len;
1769
1770         srp->rq = rq;
1771         rq->end_io_data = srp;
1772         req->retries = SG_DEFAULT_RETRIES;
1773
1774         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1775                 return 0;
1776
1777         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1778             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1779             !sfp->parentdp->device->host->unchecked_isa_dma &&
1780             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1781                 md = NULL;
1782         else
1783                 md = &map_data;
1784
1785         if (md) {
1786                 mutex_lock(&sfp->f_mutex);
1787                 if (dxfer_len <= rsv_schp->bufflen &&
1788                     !sfp->res_in_use) {
1789                         sfp->res_in_use = 1;
1790                         sg_link_reserve(sfp, srp, dxfer_len);
1791                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1792                         res = -EBUSY; /* sfp->res_in_use == 1 */
1793                         if (dxfer_len > rsv_schp->bufflen)
1794                                 res = -ENOMEM;
1795                         mutex_unlock(&sfp->f_mutex);
1796                         return res;
1797                 } else {
1798                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1799                         if (res) {
1800                                 mutex_unlock(&sfp->f_mutex);
1801                                 return res;
1802                         }
1803                 }
1804                 mutex_unlock(&sfp->f_mutex);
1805
1806                 md->pages = req_schp->pages;
1807                 md->page_order = req_schp->page_order;
1808                 md->nr_entries = req_schp->k_use_sg;
1809                 md->offset = 0;
1810                 md->null_mapped = hp->dxferp ? 0 : 1;
1811                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1812                         md->from_user = 1;
1813                 else
1814                         md->from_user = 0;
1815         }
1816
1817         if (iov_count) {
1818                 struct iovec *iov = NULL;
1819                 struct iov_iter i;
1820
1821 #ifdef CONFIG_COMPAT
1822                 if (in_compat_syscall())
1823                         res = compat_import_iovec(rw, hp->dxferp, iov_count,
1824                                                   0, &iov, &i);
1825                 else
1826 #endif
1827                         res = import_iovec(rw, hp->dxferp, iov_count,
1828                                            0, &iov, &i);
1829                 if (res < 0)
1830                         return res;
1831
1832                 iov_iter_truncate(&i, hp->dxfer_len);
1833                 if (!iov_iter_count(&i)) {
1834                         kfree(iov);
1835                         return -EINVAL;
1836                 }
1837
1838                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1839                 kfree(iov);
1840         } else
1841                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1842                                       hp->dxfer_len, GFP_ATOMIC);
1843
1844         if (!res) {
1845                 srp->bio = rq->bio;
1846
1847                 if (!md) {
1848                         req_schp->dio_in_use = 1;
1849                         hp->info |= SG_INFO_DIRECT_IO;
1850                 }
1851         }
1852         return res;
1853 }
1854
1855 static int
1856 sg_finish_rem_req(Sg_request *srp)
1857 {
1858         int ret = 0;
1859
1860         Sg_fd *sfp = srp->parentfp;
1861         Sg_scatter_hold *req_schp = &srp->data;
1862
1863         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1864                                       "sg_finish_rem_req: res_used=%d\n",
1865                                       (int) srp->res_used));
1866         if (srp->bio)
1867                 ret = blk_rq_unmap_user(srp->bio);
1868
1869         if (srp->rq) {
1870                 scsi_req_free_cmd(scsi_req(srp->rq));
1871                 blk_put_request(srp->rq);
1872         }
1873
1874         if (srp->res_used)
1875                 sg_unlink_reserve(sfp, srp);
1876         else
1877                 sg_remove_scat(sfp, req_schp);
1878
1879         return ret;
1880 }
1881
1882 static int
1883 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1884 {
1885         int sg_bufflen = tablesize * sizeof(struct page *);
1886         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1887
1888         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1889         if (!schp->pages)
1890                 return -ENOMEM;
1891         schp->sglist_len = sg_bufflen;
1892         return tablesize;       /* number of scat_gath elements allocated */
1893 }
1894
1895 static int
1896 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1897 {
1898         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1899         int sg_tablesize = sfp->parentdp->sg_tablesize;
1900         int blk_size = buff_size, order;
1901         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1902         struct sg_device *sdp = sfp->parentdp;
1903
1904         if (blk_size < 0)
1905                 return -EFAULT;
1906         if (0 == blk_size)
1907                 ++blk_size;     /* don't know why */
1908         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1909         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1910         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1911                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1912                 buff_size, blk_size));
1913
1914         /* N.B. ret_sz carried into this block ... */
1915         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1916         if (mx_sc_elems < 0)
1917                 return mx_sc_elems;     /* most likely -ENOMEM */
1918
1919         num = scatter_elem_sz;
1920         if (unlikely(num != scatter_elem_sz_prev)) {
1921                 if (num < PAGE_SIZE) {
1922                         scatter_elem_sz = PAGE_SIZE;
1923                         scatter_elem_sz_prev = PAGE_SIZE;
1924                 } else
1925                         scatter_elem_sz_prev = num;
1926         }
1927
1928         if (sdp->device->host->unchecked_isa_dma)
1929                 gfp_mask |= GFP_DMA;
1930
1931         order = get_order(num);
1932 retry:
1933         ret_sz = 1 << (PAGE_SHIFT + order);
1934
1935         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1936              k++, rem_sz -= ret_sz) {
1937
1938                 num = (rem_sz > scatter_elem_sz_prev) ?
1939                         scatter_elem_sz_prev : rem_sz;
1940
1941                 schp->pages[k] = alloc_pages(gfp_mask, order);
1942                 if (!schp->pages[k])
1943                         goto out;
1944
1945                 if (num == scatter_elem_sz_prev) {
1946                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1947                                 scatter_elem_sz = ret_sz;
1948                                 scatter_elem_sz_prev = ret_sz;
1949                         }
1950                 }
1951
1952                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1953                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1954                                  k, num, ret_sz));
1955         }               /* end of for loop */
1956
1957         schp->page_order = order;
1958         schp->k_use_sg = k;
1959         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1960                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1961                          k, rem_sz));
1962
1963         schp->bufflen = blk_size;
1964         if (rem_sz > 0) /* must have failed */
1965                 return -ENOMEM;
1966         return 0;
1967 out:
1968         for (i = 0; i < k; i++)
1969                 __free_pages(schp->pages[i], order);
1970
1971         if (--order >= 0)
1972                 goto retry;
1973
1974         return -ENOMEM;
1975 }
1976
1977 static void
1978 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1979 {
1980         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1981                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1982         if (schp->pages && schp->sglist_len > 0) {
1983                 if (!schp->dio_in_use) {
1984                         int k;
1985
1986                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1987                                 SCSI_LOG_TIMEOUT(5,
1988                                         sg_printk(KERN_INFO, sfp->parentdp,
1989                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1990                                         k, schp->pages[k]));
1991                                 __free_pages(schp->pages[k], schp->page_order);
1992                         }
1993
1994                         kfree(schp->pages);
1995                 }
1996         }
1997         memset(schp, 0, sizeof (*schp));
1998 }
1999
2000 static int
2001 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
2002 {
2003         Sg_scatter_hold *schp = &srp->data;
2004         int k, num;
2005
2006         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2007                          "sg_read_oxfer: num_read_xfer=%d\n",
2008                          num_read_xfer));
2009         if ((!outp) || (num_read_xfer <= 0))
2010                 return 0;
2011
2012         num = 1 << (PAGE_SHIFT + schp->page_order);
2013         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2014                 if (num > num_read_xfer) {
2015                         if (copy_to_user(outp, page_address(schp->pages[k]),
2016                                            num_read_xfer))
2017                                 return -EFAULT;
2018                         break;
2019                 } else {
2020                         if (copy_to_user(outp, page_address(schp->pages[k]),
2021                                            num))
2022                                 return -EFAULT;
2023                         num_read_xfer -= num;
2024                         if (num_read_xfer <= 0)
2025                                 break;
2026                         outp += num;
2027                 }
2028         }
2029
2030         return 0;
2031 }
2032
2033 static void
2034 sg_build_reserve(Sg_fd * sfp, int req_size)
2035 {
2036         Sg_scatter_hold *schp = &sfp->reserve;
2037
2038         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2039                          "sg_build_reserve: req_size=%d\n", req_size));
2040         do {
2041                 if (req_size < PAGE_SIZE)
2042                         req_size = PAGE_SIZE;
2043                 if (0 == sg_build_indirect(schp, sfp, req_size))
2044                         return;
2045                 else
2046                         sg_remove_scat(sfp, schp);
2047                 req_size >>= 1; /* divide by 2 */
2048         } while (req_size > (PAGE_SIZE / 2));
2049 }
2050
2051 static void
2052 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2053 {
2054         Sg_scatter_hold *req_schp = &srp->data;
2055         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2056         int k, num, rem;
2057
2058         srp->res_used = 1;
2059         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2060                          "sg_link_reserve: size=%d\n", size));
2061         rem = size;
2062
2063         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2064         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2065                 if (rem <= num) {
2066                         req_schp->k_use_sg = k + 1;
2067                         req_schp->sglist_len = rsv_schp->sglist_len;
2068                         req_schp->pages = rsv_schp->pages;
2069
2070                         req_schp->bufflen = size;
2071                         req_schp->page_order = rsv_schp->page_order;
2072                         break;
2073                 } else
2074                         rem -= num;
2075         }
2076
2077         if (k >= rsv_schp->k_use_sg)
2078                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2079                                  "sg_link_reserve: BAD size\n"));
2080 }
2081
2082 static void
2083 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2084 {
2085         Sg_scatter_hold *req_schp = &srp->data;
2086
2087         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2088                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2089                                       (int) req_schp->k_use_sg));
2090         req_schp->k_use_sg = 0;
2091         req_schp->bufflen = 0;
2092         req_schp->pages = NULL;
2093         req_schp->page_order = 0;
2094         req_schp->sglist_len = 0;
2095         srp->res_used = 0;
2096         /* Called without mutex lock to avoid deadlock */
2097         sfp->res_in_use = 0;
2098 }
2099
2100 static Sg_request *
2101 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2102 {
2103         Sg_request *resp;
2104         unsigned long iflags;
2105
2106         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2107         list_for_each_entry(resp, &sfp->rq_list, entry) {
2108                 /* look for requests that are ready + not SG_IO owned */
2109                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2110                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2111                         resp->done = 2; /* guard against other readers */
2112                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2113                         return resp;
2114                 }
2115         }
2116         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2117         return NULL;
2118 }
2119
2120 /* always adds to end of list */
2121 static Sg_request *
2122 sg_add_request(Sg_fd * sfp)
2123 {
2124         int k;
2125         unsigned long iflags;
2126         Sg_request *rp = sfp->req_arr;
2127
2128         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2129         if (!list_empty(&sfp->rq_list)) {
2130                 if (!sfp->cmd_q)
2131                         goto out_unlock;
2132
2133                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2134                         if (!rp->parentfp)
2135                                 break;
2136                 }
2137                 if (k >= SG_MAX_QUEUE)
2138                         goto out_unlock;
2139         }
2140         memset(rp, 0, sizeof (Sg_request));
2141         rp->parentfp = sfp;
2142         rp->header.duration = jiffies_to_msecs(jiffies);
2143         list_add_tail(&rp->entry, &sfp->rq_list);
2144         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2145         return rp;
2146 out_unlock:
2147         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2148         return NULL;
2149 }
2150
2151 /* Return of 1 for found; 0 for not found */
2152 static int
2153 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2154 {
2155         unsigned long iflags;
2156         int res = 0;
2157
2158         if (!sfp || !srp || list_empty(&sfp->rq_list))
2159                 return res;
2160         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2161         if (!list_empty(&srp->entry)) {
2162                 list_del(&srp->entry);
2163                 srp->parentfp = NULL;
2164                 res = 1;
2165         }
2166         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2167         return res;
2168 }
2169
2170 static Sg_fd *
2171 sg_add_sfp(Sg_device * sdp)
2172 {
2173         Sg_fd *sfp;
2174         unsigned long iflags;
2175         int bufflen;
2176
2177         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2178         if (!sfp)
2179                 return ERR_PTR(-ENOMEM);
2180
2181         init_waitqueue_head(&sfp->read_wait);
2182         rwlock_init(&sfp->rq_list_lock);
2183         INIT_LIST_HEAD(&sfp->rq_list);
2184         kref_init(&sfp->f_ref);
2185         mutex_init(&sfp->f_mutex);
2186         sfp->timeout = SG_DEFAULT_TIMEOUT;
2187         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2188         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2189         sfp->cmd_q = SG_DEF_COMMAND_Q;
2190         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2191         sfp->parentdp = sdp;
2192         write_lock_irqsave(&sdp->sfd_lock, iflags);
2193         if (atomic_read(&sdp->detaching)) {
2194                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2195                 kfree(sfp);
2196                 return ERR_PTR(-ENODEV);
2197         }
2198         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2199         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2200         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2201                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2202         if (unlikely(sg_big_buff != def_reserved_size))
2203                 sg_big_buff = def_reserved_size;
2204
2205         bufflen = min_t(int, sg_big_buff,
2206                         max_sectors_bytes(sdp->device->request_queue));
2207         sg_build_reserve(sfp, bufflen);
2208         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2209                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2210                                       sfp->reserve.bufflen,
2211                                       sfp->reserve.k_use_sg));
2212
2213         kref_get(&sdp->d_ref);
2214         __module_get(THIS_MODULE);
2215         return sfp;
2216 }
2217
2218 static void
2219 sg_remove_sfp_usercontext(struct work_struct *work)
2220 {
2221         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2222         struct sg_device *sdp = sfp->parentdp;
2223         Sg_request *srp;
2224         unsigned long iflags;
2225
2226         /* Cleanup any responses which were never read(). */
2227         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2228         while (!list_empty(&sfp->rq_list)) {
2229                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2230                 sg_finish_rem_req(srp);
2231                 list_del(&srp->entry);
2232                 srp->parentfp = NULL;
2233         }
2234         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2235
2236         if (sfp->reserve.bufflen > 0) {
2237                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2238                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2239                                 (int) sfp->reserve.bufflen,
2240                                 (int) sfp->reserve.k_use_sg));
2241                 sg_remove_scat(sfp, &sfp->reserve);
2242         }
2243
2244         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2245                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2246         kfree(sfp);
2247
2248         scsi_device_put(sdp->device);
2249         kref_put(&sdp->d_ref, sg_device_destroy);
2250         module_put(THIS_MODULE);
2251 }
2252
2253 static void
2254 sg_remove_sfp(struct kref *kref)
2255 {
2256         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2257         struct sg_device *sdp = sfp->parentdp;
2258         unsigned long iflags;
2259
2260         write_lock_irqsave(&sdp->sfd_lock, iflags);
2261         list_del(&sfp->sfd_siblings);
2262         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2263
2264         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2265         schedule_work(&sfp->ew.work);
2266 }
2267
2268 #ifdef CONFIG_SCSI_PROC_FS
2269 static int
2270 sg_idr_max_id(int id, void *p, void *data)
2271 {
2272         int *k = data;
2273
2274         if (*k < id)
2275                 *k = id;
2276
2277         return 0;
2278 }
2279
2280 static int
2281 sg_last_dev(void)
2282 {
2283         int k = -1;
2284         unsigned long iflags;
2285
2286         read_lock_irqsave(&sg_index_lock, iflags);
2287         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2288         read_unlock_irqrestore(&sg_index_lock, iflags);
2289         return k + 1;           /* origin 1 */
2290 }
2291 #endif
2292
2293 /* must be called with sg_index_lock held */
2294 static Sg_device *sg_lookup_dev(int dev)
2295 {
2296         return idr_find(&sg_index_idr, dev);
2297 }
2298
2299 static Sg_device *
2300 sg_get_dev(int dev)
2301 {
2302         struct sg_device *sdp;
2303         unsigned long flags;
2304
2305         read_lock_irqsave(&sg_index_lock, flags);
2306         sdp = sg_lookup_dev(dev);
2307         if (!sdp)
2308                 sdp = ERR_PTR(-ENXIO);
2309         else if (atomic_read(&sdp->detaching)) {
2310                 /* If sdp->detaching, then the refcount may already be 0, in
2311                  * which case it would be a bug to do kref_get().
2312                  */
2313                 sdp = ERR_PTR(-ENODEV);
2314         } else
2315                 kref_get(&sdp->d_ref);
2316         read_unlock_irqrestore(&sg_index_lock, flags);
2317
2318         return sdp;
2319 }
2320
2321 #ifdef CONFIG_SCSI_PROC_FS
2322 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2323
2324 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2325 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2326                                   size_t count, loff_t *off);
2327 static const struct proc_ops adio_proc_ops = {
2328         .proc_open      = sg_proc_single_open_adio,
2329         .proc_read      = seq_read,
2330         .proc_lseek     = seq_lseek,
2331         .proc_write     = sg_proc_write_adio,
2332         .proc_release   = single_release,
2333 };
2334
2335 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2336 static ssize_t sg_proc_write_dressz(struct file *filp,
2337                 const char __user *buffer, size_t count, loff_t *off);
2338 static const struct proc_ops dressz_proc_ops = {
2339         .proc_open      = sg_proc_single_open_dressz,
2340         .proc_read      = seq_read,
2341         .proc_lseek     = seq_lseek,
2342         .proc_write     = sg_proc_write_dressz,
2343         .proc_release   = single_release,
2344 };
2345
2346 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2347 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2348 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2349 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2350 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2351 static void dev_seq_stop(struct seq_file *s, void *v);
2352 static const struct seq_operations dev_seq_ops = {
2353         .start = dev_seq_start,
2354         .next  = dev_seq_next,
2355         .stop  = dev_seq_stop,
2356         .show  = sg_proc_seq_show_dev,
2357 };
2358
2359 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2360 static const struct seq_operations devstrs_seq_ops = {
2361         .start = dev_seq_start,
2362         .next  = dev_seq_next,
2363         .stop  = dev_seq_stop,
2364         .show  = sg_proc_seq_show_devstrs,
2365 };
2366
2367 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2368 static const struct seq_operations debug_seq_ops = {
2369         .start = dev_seq_start,
2370         .next  = dev_seq_next,
2371         .stop  = dev_seq_stop,
2372         .show  = sg_proc_seq_show_debug,
2373 };
2374
2375 static int
2376 sg_proc_init(void)
2377 {
2378         struct proc_dir_entry *p;
2379
2380         p = proc_mkdir("scsi/sg", NULL);
2381         if (!p)
2382                 return 1;
2383
2384         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2385         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2386         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2387         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2388         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2389         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2390         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2391         return 0;
2392 }
2393
2394
2395 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2396 {
2397         seq_printf(s, "%d\n", *((int *)s->private));
2398         return 0;
2399 }
2400
2401 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2402 {
2403         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2404 }
2405
2406 static ssize_t
2407 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2408                    size_t count, loff_t *off)
2409 {
2410         int err;
2411         unsigned long num;
2412
2413         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2414                 return -EACCES;
2415         err = kstrtoul_from_user(buffer, count, 0, &num);
2416         if (err)
2417                 return err;
2418         sg_allow_dio = num ? 1 : 0;
2419         return count;
2420 }
2421
2422 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2423 {
2424         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2425 }
2426
2427 static ssize_t
2428 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2429                      size_t count, loff_t *off)
2430 {
2431         int err;
2432         unsigned long k = ULONG_MAX;
2433
2434         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2435                 return -EACCES;
2436
2437         err = kstrtoul_from_user(buffer, count, 0, &k);
2438         if (err)
2439                 return err;
2440         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2441                 sg_big_buff = k;
2442                 return count;
2443         }
2444         return -ERANGE;
2445 }
2446
2447 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2448 {
2449         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2450                    sg_version_date);
2451         return 0;
2452 }
2453
2454 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2455 {
2456         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2457         return 0;
2458 }
2459
2460 struct sg_proc_deviter {
2461         loff_t  index;
2462         size_t  max;
2463 };
2464
2465 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2466 {
2467         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2468
2469         s->private = it;
2470         if (! it)
2471                 return NULL;
2472
2473         it->index = *pos;
2474         it->max = sg_last_dev();
2475         if (it->index >= it->max)
2476                 return NULL;
2477         return it;
2478 }
2479
2480 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2481 {
2482         struct sg_proc_deviter * it = s->private;
2483
2484         *pos = ++it->index;
2485         return (it->index < it->max) ? it : NULL;
2486 }
2487
2488 static void dev_seq_stop(struct seq_file *s, void *v)
2489 {
2490         kfree(s->private);
2491 }
2492
2493 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2494 {
2495         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2496         Sg_device *sdp;
2497         struct scsi_device *scsidp;
2498         unsigned long iflags;
2499
2500         read_lock_irqsave(&sg_index_lock, iflags);
2501         sdp = it ? sg_lookup_dev(it->index) : NULL;
2502         if ((NULL == sdp) || (NULL == sdp->device) ||
2503             (atomic_read(&sdp->detaching)))
2504                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2505         else {
2506                 scsidp = sdp->device;
2507                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2508                               scsidp->host->host_no, scsidp->channel,
2509                               scsidp->id, scsidp->lun, (int) scsidp->type,
2510                               1,
2511                               (int) scsidp->queue_depth,
2512                               (int) atomic_read(&scsidp->device_busy),
2513                               (int) scsi_device_online(scsidp));
2514         }
2515         read_unlock_irqrestore(&sg_index_lock, iflags);
2516         return 0;
2517 }
2518
2519 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2520 {
2521         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2522         Sg_device *sdp;
2523         struct scsi_device *scsidp;
2524         unsigned long iflags;
2525
2526         read_lock_irqsave(&sg_index_lock, iflags);
2527         sdp = it ? sg_lookup_dev(it->index) : NULL;
2528         scsidp = sdp ? sdp->device : NULL;
2529         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2530                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2531                            scsidp->vendor, scsidp->model, scsidp->rev);
2532         else
2533                 seq_puts(s, "<no active device>\n");
2534         read_unlock_irqrestore(&sg_index_lock, iflags);
2535         return 0;
2536 }
2537
2538 /* must be called while holding sg_index_lock */
2539 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2540 {
2541         int k, new_interface, blen, usg;
2542         Sg_request *srp;
2543         Sg_fd *fp;
2544         const sg_io_hdr_t *hp;
2545         const char * cp;
2546         unsigned int ms;
2547
2548         k = 0;
2549         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2550                 k++;
2551                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2552                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2553                            "(res)sgat=%d low_dma=%d\n", k,
2554                            jiffies_to_msecs(fp->timeout),
2555                            fp->reserve.bufflen,
2556                            (int) fp->reserve.k_use_sg,
2557                            (int) sdp->device->host->unchecked_isa_dma);
2558                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2559                            (int) fp->cmd_q, (int) fp->force_packid,
2560                            (int) fp->keep_orphan);
2561                 list_for_each_entry(srp, &fp->rq_list, entry) {
2562                         hp = &srp->header;
2563                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2564                         if (srp->res_used) {
2565                                 if (new_interface &&
2566                                     (SG_FLAG_MMAP_IO & hp->flags))
2567                                         cp = "     mmap>> ";
2568                                 else
2569                                         cp = "     rb>> ";
2570                         } else {
2571                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2572                                         cp = "     dio>> ";
2573                                 else
2574                                         cp = "     ";
2575                         }
2576                         seq_puts(s, cp);
2577                         blen = srp->data.bufflen;
2578                         usg = srp->data.k_use_sg;
2579                         seq_puts(s, srp->done ?
2580                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2581                                   : "act:");
2582                         seq_printf(s, " id=%d blen=%d",
2583                                    srp->header.pack_id, blen);
2584                         if (srp->done)
2585                                 seq_printf(s, " dur=%d", hp->duration);
2586                         else {
2587                                 ms = jiffies_to_msecs(jiffies);
2588                                 seq_printf(s, " t_o/elap=%d/%d",
2589                                         (new_interface ? hp->timeout :
2590                                                   jiffies_to_msecs(fp->timeout)),
2591                                         (ms > hp->duration ? ms - hp->duration : 0));
2592                         }
2593                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2594                                    (int) srp->data.cmd_opcode);
2595                 }
2596                 if (list_empty(&fp->rq_list))
2597                         seq_puts(s, "     No requests active\n");
2598                 read_unlock(&fp->rq_list_lock);
2599         }
2600 }
2601
2602 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2603 {
2604         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2605         Sg_device *sdp;
2606         unsigned long iflags;
2607
2608         if (it && (0 == it->index))
2609                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2610                            (int)it->max, sg_big_buff);
2611
2612         read_lock_irqsave(&sg_index_lock, iflags);
2613         sdp = it ? sg_lookup_dev(it->index) : NULL;
2614         if (NULL == sdp)
2615                 goto skip;
2616         read_lock(&sdp->sfd_lock);
2617         if (!list_empty(&sdp->sfds)) {
2618                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2619                 if (atomic_read(&sdp->detaching))
2620                         seq_puts(s, "detaching pending close ");
2621                 else if (sdp->device) {
2622                         struct scsi_device *scsidp = sdp->device;
2623
2624                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2625                                    scsidp->host->host_no,
2626                                    scsidp->channel, scsidp->id,
2627                                    scsidp->lun,
2628                                    scsidp->host->hostt->emulated);
2629                 }
2630                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2631                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2632                 sg_proc_debug_helper(s, sdp);
2633         }
2634         read_unlock(&sdp->sfd_lock);
2635 skip:
2636         read_unlock_irqrestore(&sg_index_lock, iflags);
2637         return 0;
2638 }
2639
2640 #endif                          /* CONFIG_SCSI_PROC_FS */
2641
2642 module_init(init_sg);
2643 module_exit(exit_sg);