drivers/scsi/sg.c

   1 /*
   2  *  History:
   3  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   4  *           to allow user process control of SCSI devices.
   5  *  Development Sponsored by Killy Corp. NY NY
   6  *
   7  * Original driver (sg.c):
   8  *        Copyright (C) 1992 Lawrence Foard
   9  * Version 2 and 3 extensions to driver:
  10  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License as published by
  14  * the Free Software Foundation; either version 2, or (at your option)
  15  * any later version.
  16  *
  17  */
  18
  19 static int sg_version_num = 30536;      /* 2 digits for each component */
  20 #define SG_VERSION_STR "3.5.36"
  21
  22 /*
  23  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  24  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  25  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  26  *        (otherwise the macros compile to empty statements).
  27  *
  28  */
  29 #include <linux/module.h>
  30
  31 #include <linux/fs.h>
  32 #include <linux/kernel.h>
  33 #include <linux/sched.h>
  34 #include <linux/string.h>
  35 #include <linux/mm.h>
  36 #include <linux/errno.h>
  37 #include <linux/mtio.h>
  38 #include <linux/ioctl.h>
  39 #include <linux/slab.h>
  40 #include <linux/fcntl.h>
  41 #include <linux/init.h>
  42 #include <linux/poll.h>
  43 #include <linux/moduleparam.h>
  44 #include <linux/cdev.h>
  45 #include <linux/idr.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/delay.h>
  49 #include <linux/blktrace_api.h>
  50 #include <linux/mutex.h>
  51 #include <linux/atomic.h>
  52 #include <linux/ratelimit.h>
  53 #include <linux/uio.h>
  54 #include <linux/cred.h> /* for sg_check_file_access() */
  55
  56 #include "scsi.h"
  57 #include <scsi/scsi_dbg.h>
  58 #include <scsi/scsi_host.h>
  59 #include <scsi/scsi_driver.h>
  60 #include <scsi/scsi_ioctl.h>
  61 #include <scsi/sg.h>
  62
  63 #include "scsi_logging.h"
  64
  65 #ifdef CONFIG_SCSI_PROC_FS
  66 #include <linux/proc_fs.h>
  67 static char *sg_version_date = "20140603";
  68
  69 static int sg_proc_init(void);
  70 static void sg_proc_cleanup(void);
  71 #endif
  72
  73 #define SG_ALLOW_DIO_DEF 0
  74
  75 #define SG_MAX_DEVS 32768
  76
  77 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  78  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  79  * than 16 bytes are "variable length" whose length is a multiple of 4
  80  */
  81 #define SG_MAX_CDB_SIZE 252
  82
  83 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  84
  85 int sg_big_buff = SG_DEF_RESERVED_SIZE;
  86 /* N.B. This variable is readable and writeable via
  87    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  88    of this size (or less if there is not enough memory) will be reserved
  89    for use by this file descriptor. [Deprecated usage: this variable is also
  90    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
  91    the kernel (i.e. it is not a module).] */
  92 static int def_reserved_size = -1;      /* picks up init parameter */
  93 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
  94
  95 static int scatter_elem_sz = SG_SCATTER_SZ;
  96 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
  97
  98 #define SG_SECTOR_SZ 512
  99
 100 static int sg_add_device(struct device *, struct class_interface *);
 101 static void sg_remove_device(struct device *, struct class_interface *);
 102
 103 static DEFINE_IDR(sg_index_idr);
 104 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
 105                                                            file descriptor list for device */
 106
 107 static struct class_interface sg_interface = {
 108         .add_dev        = sg_add_device,
 109         .remove_dev     = sg_remove_device,
 110 };
 111
 112 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 113         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 114         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 115         unsigned bufflen;       /* Size of (aggregate) data buffer */
 116         struct page **pages;
 117         int page_order;
 118         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 119         unsigned char cmd_opcode; /* first byte of command */
 120 } Sg_scatter_hold;
 121
 122 struct sg_device;               /* forward declarations */
 123 struct sg_fd;
 124
 125 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 126         struct list_head entry; /* list entry */
 127         struct sg_fd *parentfp; /* NULL -> not in use */
 128         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 129         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 130         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 131         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 132         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 133         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 134         /* done protected by rq_list_lock */
 135         char done;              /* 0->before bh, 1->before read, 2->read */
 136         struct request *rq;
 137         struct bio *bio;
 138         struct execute_work ew;
 139 } Sg_request;
 140
 141 typedef struct sg_fd {          /* holds the state of a file descriptor */
 142         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 143         struct sg_device *parentdp;     /* owning device */
 144         wait_queue_head_t read_wait;    /* queue read until command done */
 145         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 146         struct mutex f_mutex;   /* protect against changes in this fd */
 147         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 148         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 149         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 150         struct list_head rq_list; /* head of request list */
 151         struct fasync_struct *async_qp; /* used by asynchronous notification */
 152         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 153         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 154         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 155         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 156         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 157         char mmap_called;       /* 0 -> mmap() never called on this fd */
 158         char res_in_use;        /* 1 -> 'reserve' array in use */
 159         struct kref f_ref;
 160         struct execute_work ew;
 161 } Sg_fd;
 162
 163 typedef struct sg_device { /* holds the state of each scsi generic device */
 164         struct scsi_device *device;
 165         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 166         struct mutex open_rel_lock;     /* held when in open() or release() */
 167         int sg_tablesize;       /* adapter's max scatter-gather table size */
 168         u32 index;              /* device index number */
 169         struct list_head sfds;
 170         rwlock_t sfd_lock;      /* protect access to sfd list */
 171         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 172         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 173         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 174         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 175         struct gendisk *disk;
 176         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 177         struct kref d_ref;
 178 } Sg_device;
 179
 180 /* tasklet or soft irq callback */
 181 static void sg_rq_end_io(struct request *rq, blk_status_t status);
 182 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 183 static int sg_finish_rem_req(Sg_request * srp);
 184 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 185 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 186                            Sg_request * srp);
 187 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 188                         const char __user *buf, size_t count, int blocking,
 189                         int read_only, int sg_io_owned, Sg_request **o_srp);
 190 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 191                            unsigned char *cmnd, int timeout, int blocking);
 192 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 193 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 194 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 195 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 196 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 197 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 198 static void sg_remove_sfp(struct kref *);
 199 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 200 static Sg_request *sg_add_request(Sg_fd * sfp);
 201 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 202 static Sg_device *sg_get_dev(int dev);
 203 static void sg_device_destroy(struct kref *kref);
 204
 205 #define SZ_SG_HEADER sizeof(struct sg_header)
 206 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 207 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 208 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 209
 210 #define sg_printk(prefix, sdp, fmt, a...) \
 211         sdev_prefix_printk(prefix, (sdp)->device,               \
 212                            (sdp)->disk->disk_name, fmt, ##a)
 213
 214 /*
 215  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 216  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 217  * to trigger read() and write() calls from various contexts with elevated
 218  * privileges. This can lead to kernel memory corruption (e.g. if these
 219  * interfaces are called through splice()) and privilege escalation inside
 220  * userspace (e.g. if a process with access to such a device passes a file
 221  * descriptor to a SUID binary as stdin/stdout/stderr).
 222  *
 223  * This function provides protection for the legacy API by restricting the
 224  * calling context.
 225  */
 226 static int sg_check_file_access(struct file *filp, const char *caller)
 227 {
 228         if (filp->f_cred != current_real_cred()) {
 229                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 230                         caller, task_tgid_vnr(current), current->comm);
 231                 return -EPERM;
 232         }
 233         if (uaccess_kernel()) {
 234                 pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
 235                         caller, task_tgid_vnr(current), current->comm);
 236                 return -EACCES;
 237         }
 238         return 0;
 239 }
 240
 241 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 242 {
 243         struct sg_fd *sfp = filp->private_data;
 244
 245         if (sfp->parentdp->device->type == TYPE_SCANNER)
 246                 return 0;
 247
 248         return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
 249 }
 250
 251 static int
 252 open_wait(Sg_device *sdp, int flags)
 253 {
 254         int retval = 0;
 255
 256         if (flags & O_EXCL) {
 257                 while (sdp->open_cnt > 0) {
 258                         mutex_unlock(&sdp->open_rel_lock);
 259                         retval = wait_event_interruptible(sdp->open_wait,
 260                                         (atomic_read(&sdp->detaching) ||
 261                                          !sdp->open_cnt));
 262                         mutex_lock(&sdp->open_rel_lock);
 263
 264                         if (retval) /* -ERESTARTSYS */
 265                                 return retval;
 266                         if (atomic_read(&sdp->detaching))
 267                                 return -ENODEV;
 268                 }
 269         } else {
 270                 while (sdp->exclude) {
 271                         mutex_unlock(&sdp->open_rel_lock);
 272                         retval = wait_event_interruptible(sdp->open_wait,
 273                                         (atomic_read(&sdp->detaching) ||
 274                                          !sdp->exclude));
 275                         mutex_lock(&sdp->open_rel_lock);
 276
 277                         if (retval) /* -ERESTARTSYS */
 278                                 return retval;
 279                         if (atomic_read(&sdp->detaching))
 280                                 return -ENODEV;
 281                 }
 282         }
 283
 284         return retval;
 285 }
 286
 287 /* Returns 0 on success, else a negated errno value */
 288 static int
 289 sg_open(struct inode *inode, struct file *filp)
 290 {
 291         int dev = iminor(inode);
 292         int flags = filp->f_flags;
 293         struct request_queue *q;
 294         Sg_device *sdp;
 295         Sg_fd *sfp;
 296         int retval;
 297
 298         nonseekable_open(inode, filp);
 299         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 300                 return -EPERM; /* Can't lock it with read only access */
 301         sdp = sg_get_dev(dev);
 302         if (IS_ERR(sdp))
 303                 return PTR_ERR(sdp);
 304
 305         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 306                                       "sg_open: flags=0x%x\n", flags));
 307
 308         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 309         /* Prevent the device driver from vanishing while we sleep */
 310         retval = scsi_device_get(sdp->device);
 311         if (retval)
 312                 goto sg_put;
 313
 314         retval = scsi_autopm_get_device(sdp->device);
 315         if (retval)
 316                 goto sdp_put;
 317
 318         /* scsi_block_when_processing_errors() may block so bypass
 319          * check if O_NONBLOCK. Permits SCSI commands to be issued
 320          * during error recovery. Tread carefully. */
 321         if (!((flags & O_NONBLOCK) ||
 322               scsi_block_when_processing_errors(sdp->device))) {
 323                 retval = -ENXIO;
 324                 /* we are in error recovery for this device */
 325                 goto error_out;
 326         }
 327
 328         mutex_lock(&sdp->open_rel_lock);
 329         if (flags & O_NONBLOCK) {
 330                 if (flags & O_EXCL) {
 331                         if (sdp->open_cnt > 0) {
 332                                 retval = -EBUSY;
 333                                 goto error_mutex_locked;
 334                         }
 335                 } else {
 336                         if (sdp->exclude) {
 337                                 retval = -EBUSY;
 338                                 goto error_mutex_locked;
 339                         }
 340                 }
 341         } else {
 342                 retval = open_wait(sdp, flags);
 343                 if (retval) /* -ERESTARTSYS or -ENODEV */
 344                         goto error_mutex_locked;
 345         }
 346
 347         /* N.B. at this point we are holding the open_rel_lock */
 348         if (flags & O_EXCL)
 349                 sdp->exclude = true;
 350
 351         if (sdp->open_cnt < 1) {  /* no existing opens */
 352                 sdp->sgdebug = 0;
 353                 q = sdp->device->request_queue;
 354                 sdp->sg_tablesize = queue_max_segments(q);
 355         }
 356         sfp = sg_add_sfp(sdp);
 357         if (IS_ERR(sfp)) {
 358                 retval = PTR_ERR(sfp);
 359                 goto out_undo;
 360         }
 361
 362         filp->private_data = sfp;
 363         sdp->open_cnt++;
 364         mutex_unlock(&sdp->open_rel_lock);
 365
 366         retval = 0;
 367 sg_put:
 368         kref_put(&sdp->d_ref, sg_device_destroy);
 369         return retval;
 370
 371 out_undo:
 372         if (flags & O_EXCL) {
 373                 sdp->exclude = false;   /* undo if error */
 374                 wake_up_interruptible(&sdp->open_wait);
 375         }
 376 error_mutex_locked:
 377         mutex_unlock(&sdp->open_rel_lock);
 378 error_out:
 379         scsi_autopm_put_device(sdp->device);
 380 sdp_put:
 381         scsi_device_put(sdp->device);
 382         goto sg_put;
 383 }
 384
 385 /* Release resources associated with a successful sg_open()
 386  * Returns 0 on success, else a negated errno value */
 387 static int
 388 sg_release(struct inode *inode, struct file *filp)
 389 {
 390         Sg_device *sdp;
 391         Sg_fd *sfp;
 392
 393         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 394                 return -ENXIO;
 395         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 396
 397         mutex_lock(&sdp->open_rel_lock);
 398         scsi_autopm_put_device(sdp->device);
 399         kref_put(&sfp->f_ref, sg_remove_sfp);
 400         sdp->open_cnt--;
 401
 402         /* possibly many open()s waiting on exlude clearing, start many;
 403          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 404         if (sdp->exclude) {
 405                 sdp->exclude = false;
 406                 wake_up_interruptible_all(&sdp->open_wait);
 407         } else if (0 == sdp->open_cnt) {
 408                 wake_up_interruptible(&sdp->open_wait);
 409         }
 410         mutex_unlock(&sdp->open_rel_lock);
 411         return 0;
 412 }
 413
 414 static ssize_t
 415 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 416 {
 417         Sg_device *sdp;
 418         Sg_fd *sfp;
 419         Sg_request *srp;
 420         int req_pack_id = -1;
 421         sg_io_hdr_t *hp;
 422         struct sg_header *old_hdr = NULL;
 423         int retval = 0;
 424
 425         /*
 426          * This could cause a response to be stranded. Close the associated
 427          * file descriptor to free up any resources being held.
 428          */
 429         retval = sg_check_file_access(filp, __func__);
 430         if (retval)
 431                 return retval;
 432
 433         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 434                 return -ENXIO;
 435         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 436                                       "sg_read: count=%d\n", (int) count));
 437
 438         if (!access_ok(VERIFY_WRITE, buf, count))
 439                 return -EFAULT;
 440         if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
 441                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 442                 if (!old_hdr)
 443                         return -ENOMEM;
 444                 if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
 445                         retval = -EFAULT;
 446                         goto free_old_hdr;
 447                 }
 448                 if (old_hdr->reply_len < 0) {
 449                         if (count >= SZ_SG_IO_HDR) {
 450                                 sg_io_hdr_t *new_hdr;
 451                                 new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
 452                                 if (!new_hdr) {
 453                                         retval = -ENOMEM;
 454                                         goto free_old_hdr;
 455                                 }
 456                                 retval =__copy_from_user
 457                                     (new_hdr, buf, SZ_SG_IO_HDR);
 458                                 req_pack_id = new_hdr->pack_id;
 459                                 kfree(new_hdr);
 460                                 if (retval) {
 461                                         retval = -EFAULT;
 462                                         goto free_old_hdr;
 463                                 }
 464                         }
 465                 } else
 466                         req_pack_id = old_hdr->pack_id;
 467         }
 468         srp = sg_get_rq_mark(sfp, req_pack_id);
 469         if (!srp) {             /* now wait on packet to arrive */
 470                 if (atomic_read(&sdp->detaching)) {
 471                         retval = -ENODEV;
 472                         goto free_old_hdr;
 473                 }
 474                 if (filp->f_flags & O_NONBLOCK) {
 475                         retval = -EAGAIN;
 476                         goto free_old_hdr;
 477                 }
 478                 retval = wait_event_interruptible(sfp->read_wait,
 479                         (atomic_read(&sdp->detaching) ||
 480                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 481                 if (atomic_read(&sdp->detaching)) {
 482                         retval = -ENODEV;
 483                         goto free_old_hdr;
 484                 }
 485                 if (retval) {
 486                         /* -ERESTARTSYS as signal hit process */
 487                         goto free_old_hdr;
 488                 }
 489         }
 490         if (srp->header.interface_id != '\0') {
 491                 retval = sg_new_read(sfp, buf, count, srp);
 492                 goto free_old_hdr;
 493         }
 494
 495         hp = &srp->header;
 496         if (old_hdr == NULL) {
 497                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 498                 if (! old_hdr) {
 499                         retval = -ENOMEM;
 500                         goto free_old_hdr;
 501                 }
 502         }
 503         memset(old_hdr, 0, SZ_SG_HEADER);
 504         old_hdr->reply_len = (int) hp->timeout;
 505         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 506         old_hdr->pack_id = hp->pack_id;
 507         old_hdr->twelve_byte =
 508             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 509         old_hdr->target_status = hp->masked_status;
 510         old_hdr->host_status = hp->host_status;
 511         old_hdr->driver_status = hp->driver_status;
 512         if ((CHECK_CONDITION & hp->masked_status) ||
 513             (DRIVER_SENSE & hp->driver_status))
 514                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 515                        sizeof (old_hdr->sense_buffer));
 516         switch (hp->host_status) {
 517         /* This setup of 'result' is for backward compatibility and is best
 518            ignored by the user who should use target, host + driver status */
 519         case DID_OK:
 520         case DID_PASSTHROUGH:
 521         case DID_SOFT_ERROR:
 522                 old_hdr->result = 0;
 523                 break;
 524         case DID_NO_CONNECT:
 525         case DID_BUS_BUSY:
 526         case DID_TIME_OUT:
 527                 old_hdr->result = EBUSY;
 528                 break;
 529         case DID_BAD_TARGET:
 530         case DID_ABORT:
 531         case DID_PARITY:
 532         case DID_RESET:
 533         case DID_BAD_INTR:
 534                 old_hdr->result = EIO;
 535                 break;
 536         case DID_ERROR:
 537                 old_hdr->result = (srp->sense_b[0] == 0 &&
 538                                   hp->masked_status == GOOD) ? 0 : EIO;
 539                 break;
 540         default:
 541                 old_hdr->result = EIO;
 542                 break;
 543         }
 544
 545         /* Now copy the result back to the user buffer.  */
 546         if (count >= SZ_SG_HEADER) {
 547                 if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 548                         retval = -EFAULT;
 549                         goto free_old_hdr;
 550                 }
 551                 buf += SZ_SG_HEADER;
 552                 if (count > old_hdr->reply_len)
 553                         count = old_hdr->reply_len;
 554                 if (count > SZ_SG_HEADER) {
 555                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 556                                 retval = -EFAULT;
 557                                 goto free_old_hdr;
 558                         }
 559                 }
 560         } else
 561                 count = (old_hdr->result == 0) ? 0 : -EIO;
 562         sg_finish_rem_req(srp);
 563         sg_remove_request(sfp, srp);
 564         retval = count;
 565 free_old_hdr:
 566         kfree(old_hdr);
 567         return retval;
 568 }
 569
 570 static ssize_t
 571 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 572 {
 573         sg_io_hdr_t *hp = &srp->header;
 574         int err = 0, err2;
 575         int len;
 576
 577         if (count < SZ_SG_IO_HDR) {
 578                 err = -EINVAL;
 579                 goto err_out;
 580         }
 581         hp->sb_len_wr = 0;
 582         if ((hp->mx_sb_len > 0) && hp->sbp) {
 583                 if ((CHECK_CONDITION & hp->masked_status) ||
 584                     (DRIVER_SENSE & hp->driver_status)) {
 585                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 586                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 587                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 588                         len = (len > sb_len) ? sb_len : len;
 589                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 590                                 err = -EFAULT;
 591                                 goto err_out;
 592                         }
 593                         hp->sb_len_wr = len;
 594                 }
 595         }
 596         if (hp->masked_status || hp->host_status || hp->driver_status)
 597                 hp->info |= SG_INFO_CHECK;
 598         if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
 599                 err = -EFAULT;
 600                 goto err_out;
 601         }
 602 err_out:
 603         err2 = sg_finish_rem_req(srp);
 604         sg_remove_request(sfp, srp);
 605         return err ? : err2 ? : count;
 606 }
 607
 608 static ssize_t
 609 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 610 {
 611         int mxsize, cmd_size, k;
 612         int input_size, blocking;
 613         unsigned char opcode;
 614         Sg_device *sdp;
 615         Sg_fd *sfp;
 616         Sg_request *srp;
 617         struct sg_header old_hdr;
 618         sg_io_hdr_t *hp;
 619         unsigned char cmnd[SG_MAX_CDB_SIZE];
 620         int retval;
 621
 622         retval = sg_check_file_access(filp, __func__);
 623         if (retval)
 624                 return retval;
 625
 626         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 627                 return -ENXIO;
 628         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 629                                       "sg_write: count=%d\n", (int) count));
 630         if (atomic_read(&sdp->detaching))
 631                 return -ENODEV;
 632         if (!((filp->f_flags & O_NONBLOCK) ||
 633               scsi_block_when_processing_errors(sdp->device)))
 634                 return -ENXIO;
 635
 636         if (!access_ok(VERIFY_READ, buf, count))
 637                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 638         if (count < SZ_SG_HEADER)
 639                 return -EIO;
 640         if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 641                 return -EFAULT;
 642         blocking = !(filp->f_flags & O_NONBLOCK);
 643         if (old_hdr.reply_len < 0)
 644                 return sg_new_write(sfp, filp, buf, count,
 645                                     blocking, 0, 0, NULL);
 646         if (count < (SZ_SG_HEADER + 6))
 647                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 648
 649         if (!(srp = sg_add_request(sfp))) {
 650                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 651                                               "sg_write: queue full\n"));
 652                 return -EDOM;
 653         }
 654         buf += SZ_SG_HEADER;
 655         __get_user(opcode, buf);
 656         mutex_lock(&sfp->f_mutex);
 657         if (sfp->next_cmd_len > 0) {
 658                 cmd_size = sfp->next_cmd_len;
 659                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 660         } else {
 661                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 662                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 663                         cmd_size = 12;
 664         }
 665         mutex_unlock(&sfp->f_mutex);
 666         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 667                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 668 /* Determine buffer size.  */
 669         input_size = count - cmd_size;
 670         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 671         mxsize -= SZ_SG_HEADER;
 672         input_size -= SZ_SG_HEADER;
 673         if (input_size < 0) {
 674                 sg_remove_request(sfp, srp);
 675                 return -EIO;    /* User did not pass enough bytes for this command. */
 676         }
 677         hp = &srp->header;
 678         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 679         hp->cmd_len = (unsigned char) cmd_size;
 680         hp->iovec_count = 0;
 681         hp->mx_sb_len = 0;
 682         if (input_size > 0)
 683                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 684                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 685         else
 686                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 687         hp->dxfer_len = mxsize;
 688         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 689             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 690                 hp->dxferp = (char __user *)buf + cmd_size;
 691         else
 692                 hp->dxferp = NULL;
 693         hp->sbp = NULL;
 694         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 695         hp->flags = input_size; /* structure abuse ... */
 696         hp->pack_id = old_hdr.pack_id;
 697         hp->usr_ptr = NULL;
 698         if (__copy_from_user(cmnd, buf, cmd_size)) {
 699                 sg_remove_request(sfp, srp);
 700                 return -EFAULT;
 701         }
 702         /*
 703          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 704          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 705          * is a non-zero input_size, so emit a warning.
 706          */
 707         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 708                 printk_ratelimited(KERN_WARNING
 709                                    "sg_write: data in/out %d/%d bytes "
 710                                    "for SCSI command 0x%x-- guessing "
 711                                    "data in;\n   program %s not setting "
 712                                    "count and/or reply_len properly\n",
 713                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 714                                    input_size, (unsigned int) cmnd[0],
 715                                    current->comm);
 716         }
 717         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 718         return (k < 0) ? k : count;
 719 }
 720
 721 static ssize_t
 722 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 723                  size_t count, int blocking, int read_only, int sg_io_owned,
 724                  Sg_request **o_srp)
 725 {
 726         int k;
 727         Sg_request *srp;
 728         sg_io_hdr_t *hp;
 729         unsigned char cmnd[SG_MAX_CDB_SIZE];
 730         int timeout;
 731         unsigned long ul_timeout;
 732
 733         if (count < SZ_SG_IO_HDR)
 734                 return -EINVAL;
 735         if (!access_ok(VERIFY_READ, buf, count))
 736                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 737
 738         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 739         if (!(srp = sg_add_request(sfp))) {
 740                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 741                                               "sg_new_write: queue full\n"));
 742                 return -EDOM;
 743         }
 744         srp->sg_io_owned = sg_io_owned;
 745         hp = &srp->header;
 746         if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
 747                 sg_remove_request(sfp, srp);
 748                 return -EFAULT;
 749         }
 750         if (hp->interface_id != 'S') {
 751                 sg_remove_request(sfp, srp);
 752                 return -ENOSYS;
 753         }
 754         if (hp->flags & SG_FLAG_MMAP_IO) {
 755                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 756                         sg_remove_request(sfp, srp);
 757                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 758                 }
 759                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 760                         sg_remove_request(sfp, srp);
 761                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 762                 }
 763                 if (sfp->res_in_use) {
 764                         sg_remove_request(sfp, srp);
 765                         return -EBUSY;  /* reserve buffer already being used */
 766                 }
 767         }
 768         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 769         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 770         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 771                 sg_remove_request(sfp, srp);
 772                 return -EMSGSIZE;
 773         }
 774         if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
 775                 sg_remove_request(sfp, srp);
 776                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 777         }
 778         if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 779                 sg_remove_request(sfp, srp);
 780                 return -EFAULT;
 781         }
 782         if (read_only && sg_allow_access(file, cmnd)) {
 783                 sg_remove_request(sfp, srp);
 784                 return -EPERM;
 785         }
 786         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 787         if (k < 0)
 788                 return k;
 789         if (o_srp)
 790                 *o_srp = srp;
 791         return count;
 792 }
 793
 794 static int
 795 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 796                 unsigned char *cmnd, int timeout, int blocking)
 797 {
 798         int k, at_head;
 799         Sg_device *sdp = sfp->parentdp;
 800         sg_io_hdr_t *hp = &srp->header;
 801
 802         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 803         hp->status = 0;
 804         hp->masked_status = 0;
 805         hp->msg_status = 0;
 806         hp->info = 0;
 807         hp->host_status = 0;
 808         hp->driver_status = 0;
 809         hp->resid = 0;
 810         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 811                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 812                         (int) cmnd[0], (int) hp->cmd_len));
 813
 814         if (hp->dxfer_len >= SZ_256M) {
 815                 sg_remove_request(sfp, srp);
 816                 return -EINVAL;
 817         }
 818
 819         k = sg_start_req(srp, cmnd);
 820         if (k) {
 821                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 822                         "sg_common_write: start_req err=%d\n", k));
 823                 sg_finish_rem_req(srp);
 824                 sg_remove_request(sfp, srp);
 825                 return k;       /* probably out of space --> ENOMEM */
 826         }
 827         if (atomic_read(&sdp->detaching)) {
 828                 if (srp->bio) {
 829                         scsi_req_free_cmd(scsi_req(srp->rq));
 830                         blk_end_request_all(srp->rq, BLK_STS_IOERR);
 831                         srp->rq = NULL;
 832                 }
 833
 834                 sg_finish_rem_req(srp);
 835                 sg_remove_request(sfp, srp);
 836                 return -ENODEV;
 837         }
 838
 839         hp->duration = jiffies_to_msecs(jiffies);
 840         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 841             (SG_FLAG_Q_AT_TAIL & hp->flags))
 842                 at_head = 0;
 843         else
 844                 at_head = 1;
 845
 846         srp->rq->timeout = timeout;
 847         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 848         blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
 849                               srp->rq, at_head, sg_rq_end_io);
 850         return 0;
 851 }
 852
 853 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 854 {
 855         unsigned long flags;
 856         int ret;
 857
 858         read_lock_irqsave(&sfp->rq_list_lock, flags);
 859         ret = srp->done;
 860         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 861         return ret;
 862 }
 863
 864 static int max_sectors_bytes(struct request_queue *q)
 865 {
 866         unsigned int max_sectors = queue_max_sectors(q);
 867
 868         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 869
 870         return max_sectors << 9;
 871 }
 872
 873 static void
 874 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 875 {
 876         Sg_request *srp;
 877         int val;
 878         unsigned int ms;
 879
 880         val = 0;
 881         list_for_each_entry(srp, &sfp->rq_list, entry) {
 882                 if (val >= SG_MAX_QUEUE)
 883                         break;
 884                 rinfo[val].req_state = srp->done + 1;
 885                 rinfo[val].problem =
 886                         srp->header.masked_status &
 887                         srp->header.host_status &
 888                         srp->header.driver_status;
 889                 if (srp->done)
 890                         rinfo[val].duration =
 891                                 srp->header.duration;
 892                 else {
 893                         ms = jiffies_to_msecs(jiffies);
 894                         rinfo[val].duration =
 895                                 (ms > srp->header.duration) ?
 896                                 (ms - srp->header.duration) : 0;
 897                 }
 898                 rinfo[val].orphan = srp->orphan;
 899                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 900                 rinfo[val].pack_id = srp->header.pack_id;
 901                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 902                 val++;
 903         }
 904 }
 905
 906 static long
 907 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
 908 {
 909         void __user *p = (void __user *)arg;
 910         int __user *ip = p;
 911         int result, val, read_only;
 912         Sg_device *sdp;
 913         Sg_fd *sfp;
 914         Sg_request *srp;
 915         unsigned long iflags;
 916
 917         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 918                 return -ENXIO;
 919
 920         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 921                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 922         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 923
 924         switch (cmd_in) {
 925         case SG_IO:
 926                 if (atomic_read(&sdp->detaching))
 927                         return -ENODEV;
 928                 if (!scsi_block_when_processing_errors(sdp->device))
 929                         return -ENXIO;
 930                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
 931                         return -EFAULT;
 932                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 933                                  1, read_only, 1, &srp);
 934                 if (result < 0)
 935                         return result;
 936                 result = wait_event_interruptible(sfp->read_wait,
 937                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 938                 if (atomic_read(&sdp->detaching))
 939                         return -ENODEV;
 940                 write_lock_irq(&sfp->rq_list_lock);
 941                 if (srp->done) {
 942                         srp->done = 2;
 943                         write_unlock_irq(&sfp->rq_list_lock);
 944                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 945                         return (result < 0) ? result : 0;
 946                 }
 947                 srp->orphan = 1;
 948                 write_unlock_irq(&sfp->rq_list_lock);
 949                 return result;  /* -ERESTARTSYS because signal hit process */
 950         case SG_SET_TIMEOUT:
 951                 result = get_user(val, ip);
 952                 if (result)
 953                         return result;
 954                 if (val < 0)
 955                         return -EIO;
 956                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 957                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 958                                     INT_MAX);
 959                 sfp->timeout_user = val;
 960                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 961
 962                 return 0;
 963         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 964                                 /* strange ..., for backward compatibility */
 965                 return sfp->timeout_user;
 966         case SG_SET_FORCE_LOW_DMA:
 967                 /*
 968                  * N.B. This ioctl never worked properly, but failed to
 969                  * return an error value. So returning '0' to keep compability
 970                  * with legacy applications.
 971                  */
 972                 return 0;
 973         case SG_GET_LOW_DMA:
 974                 return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
 975         case SG_GET_SCSI_ID:
 976                 if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
 977                         return -EFAULT;
 978                 else {
 979                         sg_scsi_id_t __user *sg_idp = p;
 980
 981                         if (atomic_read(&sdp->detaching))
 982                                 return -ENODEV;
 983                         __put_user((int) sdp->device->host->host_no,
 984                                    &sg_idp->host_no);
 985                         __put_user((int) sdp->device->channel,
 986                                    &sg_idp->channel);
 987                         __put_user((int) sdp->device->id, &sg_idp->scsi_id);
 988                         __put_user((int) sdp->device->lun, &sg_idp->lun);
 989                         __put_user((int) sdp->device->type, &sg_idp->scsi_type);
 990                         __put_user((short) sdp->device->host->cmd_per_lun,
 991                                    &sg_idp->h_cmd_per_lun);
 992                         __put_user((short) sdp->device->queue_depth,
 993                                    &sg_idp->d_queue_depth);
 994                         __put_user(0, &sg_idp->unused[0]);
 995                         __put_user(0, &sg_idp->unused[1]);
 996                         return 0;
 997                 }
 998         case SG_SET_FORCE_PACK_ID:
 999                 result = get_user(val, ip);
1000                 if (result)
1001                         return result;
1002                 sfp->force_packid = val ? 1 : 0;
1003                 return 0;
1004         case SG_GET_PACK_ID:
1005                 if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
1006                         return -EFAULT;
1007                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1008                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1009                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1010                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1011                                                        iflags);
1012                                 __put_user(srp->header.pack_id, ip);
1013                                 return 0;
1014                         }
1015                 }
1016                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1017                 __put_user(-1, ip);
1018                 return 0;
1019         case SG_GET_NUM_WAITING:
1020                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1021                 val = 0;
1022                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1023                         if ((1 == srp->done) && (!srp->sg_io_owned))
1024                                 ++val;
1025                 }
1026                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1027                 return put_user(val, ip);
1028         case SG_GET_SG_TABLESIZE:
1029                 return put_user(sdp->sg_tablesize, ip);
1030         case SG_SET_RESERVED_SIZE:
1031                 result = get_user(val, ip);
1032                 if (result)
1033                         return result;
1034                 if (val < 0)
1035                         return -EINVAL;
1036                 val = min_t(int, val,
1037                             max_sectors_bytes(sdp->device->request_queue));
1038                 mutex_lock(&sfp->f_mutex);
1039                 if (val != sfp->reserve.bufflen) {
1040                         if (sfp->mmap_called ||
1041                             sfp->res_in_use) {
1042                                 mutex_unlock(&sfp->f_mutex);
1043                                 return -EBUSY;
1044                         }
1045
1046                         sg_remove_scat(sfp, &sfp->reserve);
1047                         sg_build_reserve(sfp, val);
1048                 }
1049                 mutex_unlock(&sfp->f_mutex);
1050                 return 0;
1051         case SG_GET_RESERVED_SIZE:
1052                 val = min_t(int, sfp->reserve.bufflen,
1053                             max_sectors_bytes(sdp->device->request_queue));
1054                 return put_user(val, ip);
1055         case SG_SET_COMMAND_Q:
1056                 result = get_user(val, ip);
1057                 if (result)
1058                         return result;
1059                 sfp->cmd_q = val ? 1 : 0;
1060                 return 0;
1061         case SG_GET_COMMAND_Q:
1062                 return put_user((int) sfp->cmd_q, ip);
1063         case SG_SET_KEEP_ORPHAN:
1064                 result = get_user(val, ip);
1065                 if (result)
1066                         return result;
1067                 sfp->keep_orphan = val;
1068                 return 0;
1069         case SG_GET_KEEP_ORPHAN:
1070                 return put_user((int) sfp->keep_orphan, ip);
1071         case SG_NEXT_CMD_LEN:
1072                 result = get_user(val, ip);
1073                 if (result)
1074                         return result;
1075                 if (val > SG_MAX_CDB_SIZE)
1076                         return -ENOMEM;
1077                 sfp->next_cmd_len = (val > 0) ? val : 0;
1078                 return 0;
1079         case SG_GET_VERSION_NUM:
1080                 return put_user(sg_version_num, ip);
1081         case SG_GET_ACCESS_COUNT:
1082                 /* faked - we don't have a real access count anymore */
1083                 val = (sdp->device ? 1 : 0);
1084                 return put_user(val, ip);
1085         case SG_GET_REQUEST_TABLE:
1086                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1087                         return -EFAULT;
1088                 else {
1089                         sg_req_info_t *rinfo;
1090
1091                         rinfo = kzalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
1092                                         GFP_KERNEL);
1093                         if (!rinfo)
1094                                 return -ENOMEM;
1095                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1096                         sg_fill_request_table(sfp, rinfo);
1097                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1098                         result = __copy_to_user(p, rinfo,
1099                                                 SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1100                         result = result ? -EFAULT : 0;
1101                         kfree(rinfo);
1102                         return result;
1103                 }
1104         case SG_EMULATED_HOST:
1105                 if (atomic_read(&sdp->detaching))
1106                         return -ENODEV;
1107                 return put_user(sdp->device->host->hostt->emulated, ip);
1108         case SCSI_IOCTL_SEND_COMMAND:
1109                 if (atomic_read(&sdp->detaching))
1110                         return -ENODEV;
1111                 if (read_only) {
1112                         unsigned char opcode = WRITE_6;
1113                         Scsi_Ioctl_Command __user *siocp = p;
1114
1115                         if (copy_from_user(&opcode, siocp->data, 1))
1116                                 return -EFAULT;
1117                         if (sg_allow_access(filp, &opcode))
1118                                 return -EPERM;
1119                 }
1120                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1121         case SG_SET_DEBUG:
1122                 result = get_user(val, ip);
1123                 if (result)
1124                         return result;
1125                 sdp->sgdebug = (char) val;
1126                 return 0;
1127         case BLKSECTGET:
1128                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1129                                 ip);
1130         case BLKTRACESETUP:
1131                 return blk_trace_setup(sdp->device->request_queue,
1132                                        sdp->disk->disk_name,
1133                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1134                                        NULL, p);
1135         case BLKTRACESTART:
1136                 return blk_trace_startstop(sdp->device->request_queue, 1);
1137         case BLKTRACESTOP:
1138                 return blk_trace_startstop(sdp->device->request_queue, 0);
1139         case BLKTRACETEARDOWN:
1140                 return blk_trace_remove(sdp->device->request_queue);
1141         case SCSI_IOCTL_GET_IDLUN:
1142         case SCSI_IOCTL_GET_BUS_NUMBER:
1143         case SCSI_IOCTL_PROBE_HOST:
1144         case SG_GET_TRANSFORM:
1145         case SG_SCSI_RESET:
1146                 if (atomic_read(&sdp->detaching))
1147                         return -ENODEV;
1148                 break;
1149         default:
1150                 if (read_only)
1151                         return -EPERM;  /* don't know so take safe approach */
1152                 break;
1153         }
1154
1155         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1156                         cmd_in, filp->f_flags & O_NDELAY);
1157         if (result)
1158                 return result;
1159         return scsi_ioctl(sdp->device, cmd_in, p);
1160 }
1161
1162 #ifdef CONFIG_COMPAT
1163 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1164 {
1165         Sg_device *sdp;
1166         Sg_fd *sfp;
1167         struct scsi_device *sdev;
1168
1169         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1170                 return -ENXIO;
1171
1172         sdev = sdp->device;
1173         if (sdev->host->hostt->compat_ioctl) {
1174                 int ret;
1175
1176                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1177
1178                 return ret;
1179         }
1180
1181         return -ENOIOCTLCMD;
1182 }
1183 #endif
1184
1185 static unsigned int
1186 sg_poll(struct file *filp, poll_table * wait)
1187 {
1188         unsigned int res = 0;
1189         Sg_device *sdp;
1190         Sg_fd *sfp;
1191         Sg_request *srp;
1192         int count = 0;
1193         unsigned long iflags;
1194
1195         sfp = filp->private_data;
1196         if (!sfp)
1197                 return POLLERR;
1198         sdp = sfp->parentdp;
1199         if (!sdp)
1200                 return POLLERR;
1201         poll_wait(filp, &sfp->read_wait, wait);
1202         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1203         list_for_each_entry(srp, &sfp->rq_list, entry) {
1204                 /* if any read waiting, flag it */
1205                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1206                         res = POLLIN | POLLRDNORM;
1207                 ++count;
1208         }
1209         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1210
1211         if (atomic_read(&sdp->detaching))
1212                 res |= POLLHUP;
1213         else if (!sfp->cmd_q) {
1214                 if (0 == count)
1215                         res |= POLLOUT | POLLWRNORM;
1216         } else if (count < SG_MAX_QUEUE)
1217                 res |= POLLOUT | POLLWRNORM;
1218         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1219                                       "sg_poll: res=0x%x\n", (int) res));
1220         return res;
1221 }
1222
1223 static int
1224 sg_fasync(int fd, struct file *filp, int mode)
1225 {
1226         Sg_device *sdp;
1227         Sg_fd *sfp;
1228
1229         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1230                 return -ENXIO;
1231         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1232                                       "sg_fasync: mode=%d\n", mode));
1233
1234         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1235 }
1236
1237 static int
1238 sg_vma_fault(struct vm_fault *vmf)
1239 {
1240         struct vm_area_struct *vma = vmf->vma;
1241         Sg_fd *sfp;
1242         unsigned long offset, len, sa;
1243         Sg_scatter_hold *rsv_schp;
1244         int k, length;
1245
1246         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1247                 return VM_FAULT_SIGBUS;
1248         rsv_schp = &sfp->reserve;
1249         offset = vmf->pgoff << PAGE_SHIFT;
1250         if (offset >= rsv_schp->bufflen)
1251                 return VM_FAULT_SIGBUS;
1252         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1253                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1254                                       offset, rsv_schp->k_use_sg));
1255         sa = vma->vm_start;
1256         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1257         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1258                 len = vma->vm_end - sa;
1259                 len = (len < length) ? len : length;
1260                 if (offset < len) {
1261                         struct page *page = nth_page(rsv_schp->pages[k],
1262                                                      offset >> PAGE_SHIFT);
1263                         get_page(page); /* increment page count */
1264                         vmf->page = page;
1265                         return 0; /* success */
1266                 }
1267                 sa += len;
1268                 offset -= len;
1269         }
1270
1271         return VM_FAULT_SIGBUS;
1272 }
1273
1274 static const struct vm_operations_struct sg_mmap_vm_ops = {
1275         .fault = sg_vma_fault,
1276 };
1277
1278 static int
1279 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1280 {
1281         Sg_fd *sfp;
1282         unsigned long req_sz, len, sa;
1283         Sg_scatter_hold *rsv_schp;
1284         int k, length;
1285         int ret = 0;
1286
1287         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1288                 return -ENXIO;
1289         req_sz = vma->vm_end - vma->vm_start;
1290         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1291                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1292                                       (void *) vma->vm_start, (int) req_sz));
1293         if (vma->vm_pgoff)
1294                 return -EINVAL; /* want no offset */
1295         rsv_schp = &sfp->reserve;
1296         mutex_lock(&sfp->f_mutex);
1297         if (req_sz > rsv_schp->bufflen) {
1298                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1299                 goto out;
1300         }
1301
1302         sa = vma->vm_start;
1303         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1304         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1305                 len = vma->vm_end - sa;
1306                 len = (len < length) ? len : length;
1307                 sa += len;
1308         }
1309
1310         sfp->mmap_called = 1;
1311         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1312         vma->vm_private_data = sfp;
1313         vma->vm_ops = &sg_mmap_vm_ops;
1314 out:
1315         mutex_unlock(&sfp->f_mutex);
1316         return ret;
1317 }
1318
1319 static void
1320 sg_rq_end_io_usercontext(struct work_struct *work)
1321 {
1322         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1323         struct sg_fd *sfp = srp->parentfp;
1324
1325         sg_finish_rem_req(srp);
1326         sg_remove_request(sfp, srp);
1327         kref_put(&sfp->f_ref, sg_remove_sfp);
1328 }
1329
1330 /*
1331  * This function is a "bottom half" handler that is called by the mid
1332  * level when a command is completed (or has failed).
1333  */
1334 static void
1335 sg_rq_end_io(struct request *rq, blk_status_t status)
1336 {
1337         struct sg_request *srp = rq->end_io_data;
1338         struct scsi_request *req = scsi_req(rq);
1339         Sg_device *sdp;
1340         Sg_fd *sfp;
1341         unsigned long iflags;
1342         unsigned int ms;
1343         char *sense;
1344         int result, resid, done = 1;
1345
1346         if (WARN_ON(srp->done != 0))
1347                 return;
1348
1349         sfp = srp->parentfp;
1350         if (WARN_ON(sfp == NULL))
1351                 return;
1352
1353         sdp = sfp->parentdp;
1354         if (unlikely(atomic_read(&sdp->detaching)))
1355                 pr_info("%s: device detaching\n", __func__);
1356
1357         sense = req->sense;
1358         result = req->result;
1359         resid = req->resid_len;
1360
1361         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1362                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1363                                       srp->header.pack_id, result));
1364         srp->header.resid = resid;
1365         ms = jiffies_to_msecs(jiffies);
1366         srp->header.duration = (ms > srp->header.duration) ?
1367                                 (ms - srp->header.duration) : 0;
1368         if (0 != result) {
1369                 struct scsi_sense_hdr sshdr;
1370
1371                 srp->header.status = 0xff & result;
1372                 srp->header.masked_status = status_byte(result);
1373                 srp->header.msg_status = msg_byte(result);
1374                 srp->header.host_status = host_byte(result);
1375                 srp->header.driver_status = driver_byte(result);
1376                 if ((sdp->sgdebug > 0) &&
1377                     ((CHECK_CONDITION == srp->header.masked_status) ||
1378                      (COMMAND_TERMINATED == srp->header.masked_status)))
1379                         __scsi_print_sense(sdp->device, __func__, sense,
1380                                            SCSI_SENSE_BUFFERSIZE);
1381
1382                 /* Following if statement is a patch supplied by Eric Youngdale */
1383                 if (driver_byte(result) != 0
1384                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1385                     && !scsi_sense_is_deferred(&sshdr)
1386                     && sshdr.sense_key == UNIT_ATTENTION
1387                     && sdp->device->removable) {
1388                         /* Detected possible disc change. Set the bit - this */
1389                         /* may be used if there are filesystems using this device */
1390                         sdp->device->changed = 1;
1391                 }
1392         }
1393
1394         if (req->sense_len)
1395                 memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1396
1397         /* Rely on write phase to clean out srp status values, so no "else" */
1398
1399         /*
1400          * Free the request as soon as it is complete so that its resources
1401          * can be reused without waiting for userspace to read() the
1402          * result.  But keep the associated bio (if any) around until
1403          * blk_rq_unmap_user() can be called from user context.
1404          */
1405         srp->rq = NULL;
1406         scsi_req_free_cmd(scsi_req(rq));
1407         __blk_put_request(rq->q, rq);
1408
1409         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1410         if (unlikely(srp->orphan)) {
1411                 if (sfp->keep_orphan)
1412                         srp->sg_io_owned = 0;
1413                 else
1414                         done = 0;
1415         }
1416         srp->done = done;
1417         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1418
1419         if (likely(done)) {
1420                 /* Now wake up any sg_read() that is waiting for this
1421                  * packet.
1422                  */
1423                 wake_up_interruptible(&sfp->read_wait);
1424                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1425                 kref_put(&sfp->f_ref, sg_remove_sfp);
1426         } else {
1427                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1428                 schedule_work(&srp->ew.work);
1429         }
1430 }
1431
1432 static const struct file_operations sg_fops = {
1433         .owner = THIS_MODULE,
1434         .read = sg_read,
1435         .write = sg_write,
1436         .poll = sg_poll,
1437         .unlocked_ioctl = sg_ioctl,
1438 #ifdef CONFIG_COMPAT
1439         .compat_ioctl = sg_compat_ioctl,
1440 #endif
1441         .open = sg_open,
1442         .mmap = sg_mmap,
1443         .release = sg_release,
1444         .fasync = sg_fasync,
1445         .llseek = no_llseek,
1446 };
1447
1448 static struct class *sg_sysfs_class;
1449
1450 static int sg_sysfs_valid = 0;
1451
1452 static Sg_device *
1453 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1454 {
1455         struct request_queue *q = scsidp->request_queue;
1456         Sg_device *sdp;
1457         unsigned long iflags;
1458         int error;
1459         u32 k;
1460
1461         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1462         if (!sdp) {
1463                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1464                             "failure\n", __func__);
1465                 return ERR_PTR(-ENOMEM);
1466         }
1467
1468         idr_preload(GFP_KERNEL);
1469         write_lock_irqsave(&sg_index_lock, iflags);
1470
1471         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1472         if (error < 0) {
1473                 if (error == -ENOSPC) {
1474                         sdev_printk(KERN_WARNING, scsidp,
1475                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1476                                     scsidp->type, SG_MAX_DEVS - 1);
1477                         error = -ENODEV;
1478                 } else {
1479                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1480                                     "allocation Sg_device failure: %d\n",
1481                                     __func__, error);
1482                 }
1483                 goto out_unlock;
1484         }
1485         k = error;
1486
1487         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1488                                         "sg_alloc: dev=%d \n", k));
1489         sprintf(disk->disk_name, "sg%d", k);
1490         disk->first_minor = k;
1491         sdp->disk = disk;
1492         sdp->device = scsidp;
1493         mutex_init(&sdp->open_rel_lock);
1494         INIT_LIST_HEAD(&sdp->sfds);
1495         init_waitqueue_head(&sdp->open_wait);
1496         atomic_set(&sdp->detaching, 0);
1497         rwlock_init(&sdp->sfd_lock);
1498         sdp->sg_tablesize = queue_max_segments(q);
1499         sdp->index = k;
1500         kref_init(&sdp->d_ref);
1501         error = 0;
1502
1503 out_unlock:
1504         write_unlock_irqrestore(&sg_index_lock, iflags);
1505         idr_preload_end();
1506
1507         if (error) {
1508                 kfree(sdp);
1509                 return ERR_PTR(error);
1510         }
1511         return sdp;
1512 }
1513
1514 static int
1515 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1516 {
1517         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1518         struct gendisk *disk;
1519         Sg_device *sdp = NULL;
1520         struct cdev * cdev = NULL;
1521         int error;
1522         unsigned long iflags;
1523
1524         disk = alloc_disk(1);
1525         if (!disk) {
1526                 pr_warn("%s: alloc_disk failed\n", __func__);
1527                 return -ENOMEM;
1528         }
1529         disk->major = SCSI_GENERIC_MAJOR;
1530
1531         error = -ENOMEM;
1532         cdev = cdev_alloc();
1533         if (!cdev) {
1534                 pr_warn("%s: cdev_alloc failed\n", __func__);
1535                 goto out;
1536         }
1537         cdev->owner = THIS_MODULE;
1538         cdev->ops = &sg_fops;
1539
1540         sdp = sg_alloc(disk, scsidp);
1541         if (IS_ERR(sdp)) {
1542                 pr_warn("%s: sg_alloc failed\n", __func__);
1543                 error = PTR_ERR(sdp);
1544                 goto out;
1545         }
1546
1547         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1548         if (error)
1549                 goto cdev_add_err;
1550
1551         sdp->cdev = cdev;
1552         if (sg_sysfs_valid) {
1553                 struct device *sg_class_member;
1554
1555                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1556                                                 MKDEV(SCSI_GENERIC_MAJOR,
1557                                                       sdp->index),
1558                                                 sdp, "%s", disk->disk_name);
1559                 if (IS_ERR(sg_class_member)) {
1560                         pr_err("%s: device_create failed\n", __func__);
1561                         error = PTR_ERR(sg_class_member);
1562                         goto cdev_add_err;
1563                 }
1564                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1565                                           &sg_class_member->kobj, "generic");
1566                 if (error)
1567                         pr_err("%s: unable to make symlink 'generic' back "
1568                                "to sg%d\n", __func__, sdp->index);
1569         } else
1570                 pr_warn("%s: sg_sys Invalid\n", __func__);
1571
1572         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1573                     "type %d\n", sdp->index, scsidp->type);
1574
1575         dev_set_drvdata(cl_dev, sdp);
1576
1577         return 0;
1578
1579 cdev_add_err:
1580         write_lock_irqsave(&sg_index_lock, iflags);
1581         idr_remove(&sg_index_idr, sdp->index);
1582         write_unlock_irqrestore(&sg_index_lock, iflags);
1583         kfree(sdp);
1584
1585 out:
1586         put_disk(disk);
1587         if (cdev)
1588                 cdev_del(cdev);
1589         return error;
1590 }
1591
1592 static void
1593 sg_device_destroy(struct kref *kref)
1594 {
1595         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1596         unsigned long flags;
1597
1598         /* CAUTION!  Note that the device can still be found via idr_find()
1599          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1600          * any other cleanup.
1601          */
1602
1603         write_lock_irqsave(&sg_index_lock, flags);
1604         idr_remove(&sg_index_idr, sdp->index);
1605         write_unlock_irqrestore(&sg_index_lock, flags);
1606
1607         SCSI_LOG_TIMEOUT(3,
1608                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1609
1610         put_disk(sdp->disk);
1611         kfree(sdp);
1612 }
1613
1614 static void
1615 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1616 {
1617         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1618         Sg_device *sdp = dev_get_drvdata(cl_dev);
1619         unsigned long iflags;
1620         Sg_fd *sfp;
1621         int val;
1622
1623         if (!sdp)
1624                 return;
1625         /* want sdp->detaching non-zero as soon as possible */
1626         val = atomic_inc_return(&sdp->detaching);
1627         if (val > 1)
1628                 return; /* only want to do following once per device */
1629
1630         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1631                                       "%s\n", __func__));
1632
1633         read_lock_irqsave(&sdp->sfd_lock, iflags);
1634         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1635                 wake_up_interruptible_all(&sfp->read_wait);
1636                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1637         }
1638         wake_up_interruptible_all(&sdp->open_wait);
1639         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1640
1641         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1642         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1643         cdev_del(sdp->cdev);
1644         sdp->cdev = NULL;
1645
1646         kref_put(&sdp->d_ref, sg_device_destroy);
1647 }
1648
1649 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1650 module_param_named(def_reserved_size, def_reserved_size, int,
1651                    S_IRUGO | S_IWUSR);
1652 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1653
1654 MODULE_AUTHOR("Douglas Gilbert");
1655 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1656 MODULE_LICENSE("GPL");
1657 MODULE_VERSION(SG_VERSION_STR);
1658 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1659
1660 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1661                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1662 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1663 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1664
1665 static int __init
1666 init_sg(void)
1667 {
1668         int rc;
1669
1670         if (scatter_elem_sz < PAGE_SIZE) {
1671                 scatter_elem_sz = PAGE_SIZE;
1672                 scatter_elem_sz_prev = scatter_elem_sz;
1673         }
1674         if (def_reserved_size >= 0)
1675                 sg_big_buff = def_reserved_size;
1676         else
1677                 def_reserved_size = sg_big_buff;
1678
1679         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1680                                     SG_MAX_DEVS, "sg");
1681         if (rc)
1682                 return rc;
1683         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1684         if ( IS_ERR(sg_sysfs_class) ) {
1685                 rc = PTR_ERR(sg_sysfs_class);
1686                 goto err_out;
1687         }
1688         sg_sysfs_valid = 1;
1689         rc = scsi_register_interface(&sg_interface);
1690         if (0 == rc) {
1691 #ifdef CONFIG_SCSI_PROC_FS
1692                 sg_proc_init();
1693 #endif                          /* CONFIG_SCSI_PROC_FS */
1694                 return 0;
1695         }
1696         class_destroy(sg_sysfs_class);
1697 err_out:
1698         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1699         return rc;
1700 }
1701
1702 static void __exit
1703 exit_sg(void)
1704 {
1705 #ifdef CONFIG_SCSI_PROC_FS
1706         sg_proc_cleanup();
1707 #endif                          /* CONFIG_SCSI_PROC_FS */
1708         scsi_unregister_interface(&sg_interface);
1709         class_destroy(sg_sysfs_class);
1710         sg_sysfs_valid = 0;
1711         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1712                                  SG_MAX_DEVS);
1713         idr_destroy(&sg_index_idr);
1714 }
1715
1716 static int
1717 sg_start_req(Sg_request *srp, unsigned char *cmd)
1718 {
1719         int res;
1720         struct request *rq;
1721         struct scsi_request *req;
1722         Sg_fd *sfp = srp->parentfp;
1723         sg_io_hdr_t *hp = &srp->header;
1724         int dxfer_len = (int) hp->dxfer_len;
1725         int dxfer_dir = hp->dxfer_direction;
1726         unsigned int iov_count = hp->iovec_count;
1727         Sg_scatter_hold *req_schp = &srp->data;
1728         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1729         struct request_queue *q = sfp->parentdp->device->request_queue;
1730         struct rq_map_data *md, map_data;
1731         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1732         unsigned char *long_cmdp = NULL;
1733
1734         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1735                                       "sg_start_req: dxfer_len=%d\n",
1736                                       dxfer_len));
1737
1738         if (hp->cmd_len > BLK_MAX_CDB) {
1739                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1740                 if (!long_cmdp)
1741                         return -ENOMEM;
1742         }
1743
1744         /*
1745          * NOTE
1746          *
1747          * With scsi-mq enabled, there are a fixed number of preallocated
1748          * requests equal in number to shost->can_queue.  If all of the
1749          * preallocated requests are already in use, then using GFP_ATOMIC with
1750          * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1751          * will cause blk_get_request() to sleep until an active command
1752          * completes, freeing up a request.  Neither option is ideal, but
1753          * GFP_KERNEL is the better choice to prevent userspace from getting an
1754          * unexpected EWOULDBLOCK.
1755          *
1756          * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1757          * does not sleep except under memory pressure.
1758          */
1759         rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1760                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, GFP_KERNEL);
1761         if (IS_ERR(rq)) {
1762                 kfree(long_cmdp);
1763                 return PTR_ERR(rq);
1764         }
1765         req = scsi_req(rq);
1766
1767         if (hp->cmd_len > BLK_MAX_CDB)
1768                 req->cmd = long_cmdp;
1769         memcpy(req->cmd, cmd, hp->cmd_len);
1770         req->cmd_len = hp->cmd_len;
1771
1772         srp->rq = rq;
1773         rq->end_io_data = srp;
1774         req->retries = SG_DEFAULT_RETRIES;
1775
1776         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1777                 return 0;
1778
1779         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1780             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1781             !sfp->parentdp->device->host->unchecked_isa_dma &&
1782             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1783                 md = NULL;
1784         else
1785                 md = &map_data;
1786
1787         if (md) {
1788                 mutex_lock(&sfp->f_mutex);
1789                 if (dxfer_len <= rsv_schp->bufflen &&
1790                     !sfp->res_in_use) {
1791                         sfp->res_in_use = 1;
1792                         sg_link_reserve(sfp, srp, dxfer_len);
1793                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1794                         res = -EBUSY; /* sfp->res_in_use == 1 */
1795                         if (dxfer_len > rsv_schp->bufflen)
1796                                 res = -ENOMEM;
1797                         mutex_unlock(&sfp->f_mutex);
1798                         return res;
1799                 } else {
1800                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1801                         if (res) {
1802                                 mutex_unlock(&sfp->f_mutex);
1803                                 return res;
1804                         }
1805                 }
1806                 mutex_unlock(&sfp->f_mutex);
1807
1808                 md->pages = req_schp->pages;
1809                 md->page_order = req_schp->page_order;
1810                 md->nr_entries = req_schp->k_use_sg;
1811                 md->offset = 0;
1812                 md->null_mapped = hp->dxferp ? 0 : 1;
1813                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1814                         md->from_user = 1;
1815                 else
1816                         md->from_user = 0;
1817         }
1818
1819         if (iov_count) {
1820                 struct iovec *iov = NULL;
1821                 struct iov_iter i;
1822
1823                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1824                 if (res < 0)
1825                         return res;
1826
1827                 iov_iter_truncate(&i, hp->dxfer_len);
1828                 if (!iov_iter_count(&i)) {
1829                         kfree(iov);
1830                         return -EINVAL;
1831                 }
1832
1833                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1834                 kfree(iov);
1835         } else
1836                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1837                                       hp->dxfer_len, GFP_ATOMIC);
1838
1839         if (!res) {
1840                 srp->bio = rq->bio;
1841
1842                 if (!md) {
1843                         req_schp->dio_in_use = 1;
1844                         hp->info |= SG_INFO_DIRECT_IO;
1845                 }
1846         }
1847         return res;
1848 }
1849
1850 static int
1851 sg_finish_rem_req(Sg_request *srp)
1852 {
1853         int ret = 0;
1854
1855         Sg_fd *sfp = srp->parentfp;
1856         Sg_scatter_hold *req_schp = &srp->data;
1857
1858         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1859                                       "sg_finish_rem_req: res_used=%d\n",
1860                                       (int) srp->res_used));
1861         if (srp->bio)
1862                 ret = blk_rq_unmap_user(srp->bio);
1863
1864         if (srp->rq) {
1865                 scsi_req_free_cmd(scsi_req(srp->rq));
1866                 blk_put_request(srp->rq);
1867         }
1868
1869         if (srp->res_used)
1870                 sg_unlink_reserve(sfp, srp);
1871         else
1872                 sg_remove_scat(sfp, req_schp);
1873
1874         return ret;
1875 }
1876
1877 static int
1878 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1879 {
1880         int sg_bufflen = tablesize * sizeof(struct page *);
1881         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1882
1883         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1884         if (!schp->pages)
1885                 return -ENOMEM;
1886         schp->sglist_len = sg_bufflen;
1887         return tablesize;       /* number of scat_gath elements allocated */
1888 }
1889
1890 static int
1891 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1892 {
1893         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1894         int sg_tablesize = sfp->parentdp->sg_tablesize;
1895         int blk_size = buff_size, order;
1896         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1897         struct sg_device *sdp = sfp->parentdp;
1898
1899         if (blk_size < 0)
1900                 return -EFAULT;
1901         if (0 == blk_size)
1902                 ++blk_size;     /* don't know why */
1903         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1904         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1905         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1906                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1907                 buff_size, blk_size));
1908
1909         /* N.B. ret_sz carried into this block ... */
1910         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1911         if (mx_sc_elems < 0)
1912                 return mx_sc_elems;     /* most likely -ENOMEM */
1913
1914         num = scatter_elem_sz;
1915         if (unlikely(num != scatter_elem_sz_prev)) {
1916                 if (num < PAGE_SIZE) {
1917                         scatter_elem_sz = PAGE_SIZE;
1918                         scatter_elem_sz_prev = PAGE_SIZE;
1919                 } else
1920                         scatter_elem_sz_prev = num;
1921         }
1922
1923         if (sdp->device->host->unchecked_isa_dma)
1924                 gfp_mask |= GFP_DMA;
1925
1926         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1927                 gfp_mask |= __GFP_ZERO;
1928
1929         order = get_order(num);
1930 retry:
1931         ret_sz = 1 << (PAGE_SHIFT + order);
1932
1933         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1934              k++, rem_sz -= ret_sz) {
1935
1936                 num = (rem_sz > scatter_elem_sz_prev) ?
1937                         scatter_elem_sz_prev : rem_sz;
1938
1939                 schp->pages[k] = alloc_pages(gfp_mask | __GFP_ZERO, order);
1940                 if (!schp->pages[k])
1941                         goto out;
1942
1943                 if (num == scatter_elem_sz_prev) {
1944                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1945                                 scatter_elem_sz = ret_sz;
1946                                 scatter_elem_sz_prev = ret_sz;
1947                         }
1948                 }
1949
1950                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1951                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1952                                  k, num, ret_sz));
1953         }               /* end of for loop */
1954
1955         schp->page_order = order;
1956         schp->k_use_sg = k;
1957         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1958                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1959                          k, rem_sz));
1960
1961         schp->bufflen = blk_size;
1962         if (rem_sz > 0) /* must have failed */
1963                 return -ENOMEM;
1964         return 0;
1965 out:
1966         for (i = 0; i < k; i++)
1967                 __free_pages(schp->pages[i], order);
1968
1969         if (--order >= 0)
1970                 goto retry;
1971
1972         return -ENOMEM;
1973 }
1974
1975 static void
1976 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1977 {
1978         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1979                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1980         if (schp->pages && schp->sglist_len > 0) {
1981                 if (!schp->dio_in_use) {
1982                         int k;
1983
1984                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1985                                 SCSI_LOG_TIMEOUT(5,
1986                                         sg_printk(KERN_INFO, sfp->parentdp,
1987                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1988                                         k, schp->pages[k]));
1989                                 __free_pages(schp->pages[k], schp->page_order);
1990                         }
1991
1992                         kfree(schp->pages);
1993                 }
1994         }
1995         memset(schp, 0, sizeof (*schp));
1996 }
1997
1998 static int
1999 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
2000 {
2001         Sg_scatter_hold *schp = &srp->data;
2002         int k, num;
2003
2004         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2005                          "sg_read_oxfer: num_read_xfer=%d\n",
2006                          num_read_xfer));
2007         if ((!outp) || (num_read_xfer <= 0))
2008                 return 0;
2009
2010         num = 1 << (PAGE_SHIFT + schp->page_order);
2011         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2012                 if (num > num_read_xfer) {
2013                         if (__copy_to_user(outp, page_address(schp->pages[k]),
2014                                            num_read_xfer))
2015                                 return -EFAULT;
2016                         break;
2017                 } else {
2018                         if (__copy_to_user(outp, page_address(schp->pages[k]),
2019                                            num))
2020                                 return -EFAULT;
2021                         num_read_xfer -= num;
2022                         if (num_read_xfer <= 0)
2023                                 break;
2024                         outp += num;
2025                 }
2026         }
2027
2028         return 0;
2029 }
2030
2031 static void
2032 sg_build_reserve(Sg_fd * sfp, int req_size)
2033 {
2034         Sg_scatter_hold *schp = &sfp->reserve;
2035
2036         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2037                          "sg_build_reserve: req_size=%d\n", req_size));
2038         do {
2039                 if (req_size < PAGE_SIZE)
2040                         req_size = PAGE_SIZE;
2041                 if (0 == sg_build_indirect(schp, sfp, req_size))
2042                         return;
2043                 else
2044                         sg_remove_scat(sfp, schp);
2045                 req_size >>= 1; /* divide by 2 */
2046         } while (req_size > (PAGE_SIZE / 2));
2047 }
2048
2049 static void
2050 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2051 {
2052         Sg_scatter_hold *req_schp = &srp->data;
2053         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2054         int k, num, rem;
2055
2056         srp->res_used = 1;
2057         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2058                          "sg_link_reserve: size=%d\n", size));
2059         rem = size;
2060
2061         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2062         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2063                 if (rem <= num) {
2064                         req_schp->k_use_sg = k + 1;
2065                         req_schp->sglist_len = rsv_schp->sglist_len;
2066                         req_schp->pages = rsv_schp->pages;
2067
2068                         req_schp->bufflen = size;
2069                         req_schp->page_order = rsv_schp->page_order;
2070                         break;
2071                 } else
2072                         rem -= num;
2073         }
2074
2075         if (k >= rsv_schp->k_use_sg)
2076                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2077                                  "sg_link_reserve: BAD size\n"));
2078 }
2079
2080 static void
2081 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2082 {
2083         Sg_scatter_hold *req_schp = &srp->data;
2084
2085         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2086                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2087                                       (int) req_schp->k_use_sg));
2088         req_schp->k_use_sg = 0;
2089         req_schp->bufflen = 0;
2090         req_schp->pages = NULL;
2091         req_schp->page_order = 0;
2092         req_schp->sglist_len = 0;
2093         srp->res_used = 0;
2094         /* Called without mutex lock to avoid deadlock */
2095         sfp->res_in_use = 0;
2096 }
2097
2098 static Sg_request *
2099 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2100 {
2101         Sg_request *resp;
2102         unsigned long iflags;
2103
2104         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2105         list_for_each_entry(resp, &sfp->rq_list, entry) {
2106                 /* look for requests that are ready + not SG_IO owned */
2107                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2108                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2109                         resp->done = 2; /* guard against other readers */
2110                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2111                         return resp;
2112                 }
2113         }
2114         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2115         return NULL;
2116 }
2117
2118 /* always adds to end of list */
2119 static Sg_request *
2120 sg_add_request(Sg_fd * sfp)
2121 {
2122         int k;
2123         unsigned long iflags;
2124         Sg_request *rp = sfp->req_arr;
2125
2126         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2127         if (!list_empty(&sfp->rq_list)) {
2128                 if (!sfp->cmd_q)
2129                         goto out_unlock;
2130
2131                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2132                         if (!rp->parentfp)
2133                                 break;
2134                 }
2135                 if (k >= SG_MAX_QUEUE)
2136                         goto out_unlock;
2137         }
2138         memset(rp, 0, sizeof (Sg_request));
2139         rp->parentfp = sfp;
2140         rp->header.duration = jiffies_to_msecs(jiffies);
2141         list_add_tail(&rp->entry, &sfp->rq_list);
2142         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2143         return rp;
2144 out_unlock:
2145         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2146         return NULL;
2147 }
2148
2149 /* Return of 1 for found; 0 for not found */
2150 static int
2151 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2152 {
2153         unsigned long iflags;
2154         int res = 0;
2155
2156         if (!sfp || !srp || list_empty(&sfp->rq_list))
2157                 return res;
2158         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2159         if (!list_empty(&srp->entry)) {
2160                 list_del(&srp->entry);
2161                 srp->parentfp = NULL;
2162                 res = 1;
2163         }
2164         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2165         return res;
2166 }
2167
2168 static Sg_fd *
2169 sg_add_sfp(Sg_device * sdp)
2170 {
2171         Sg_fd *sfp;
2172         unsigned long iflags;
2173         int bufflen;
2174
2175         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2176         if (!sfp)
2177                 return ERR_PTR(-ENOMEM);
2178
2179         init_waitqueue_head(&sfp->read_wait);
2180         rwlock_init(&sfp->rq_list_lock);
2181         INIT_LIST_HEAD(&sfp->rq_list);
2182         kref_init(&sfp->f_ref);
2183         mutex_init(&sfp->f_mutex);
2184         sfp->timeout = SG_DEFAULT_TIMEOUT;
2185         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2186         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2187         sfp->cmd_q = SG_DEF_COMMAND_Q;
2188         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2189         sfp->parentdp = sdp;
2190         write_lock_irqsave(&sdp->sfd_lock, iflags);
2191         if (atomic_read(&sdp->detaching)) {
2192                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2193                 kfree(sfp);
2194                 return ERR_PTR(-ENODEV);
2195         }
2196         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2197         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2198         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2199                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2200         if (unlikely(sg_big_buff != def_reserved_size))
2201                 sg_big_buff = def_reserved_size;
2202
2203         bufflen = min_t(int, sg_big_buff,
2204                         max_sectors_bytes(sdp->device->request_queue));
2205         sg_build_reserve(sfp, bufflen);
2206         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2207                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2208                                       sfp->reserve.bufflen,
2209                                       sfp->reserve.k_use_sg));
2210
2211         kref_get(&sdp->d_ref);
2212         __module_get(THIS_MODULE);
2213         return sfp;
2214 }
2215
2216 static void
2217 sg_remove_sfp_usercontext(struct work_struct *work)
2218 {
2219         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2220         struct sg_device *sdp = sfp->parentdp;
2221         Sg_request *srp;
2222         unsigned long iflags;
2223
2224         /* Cleanup any responses which were never read(). */
2225         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2226         while (!list_empty(&sfp->rq_list)) {
2227                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2228                 sg_finish_rem_req(srp);
2229                 list_del(&srp->entry);
2230                 srp->parentfp = NULL;
2231         }
2232         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2233
2234         if (sfp->reserve.bufflen > 0) {
2235                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2236                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2237                                 (int) sfp->reserve.bufflen,
2238                                 (int) sfp->reserve.k_use_sg));
2239                 sg_remove_scat(sfp, &sfp->reserve);
2240         }
2241
2242         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2243                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2244         kfree(sfp);
2245
2246         scsi_device_put(sdp->device);
2247         kref_put(&sdp->d_ref, sg_device_destroy);
2248         module_put(THIS_MODULE);
2249 }
2250
2251 static void
2252 sg_remove_sfp(struct kref *kref)
2253 {
2254         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2255         struct sg_device *sdp = sfp->parentdp;
2256         unsigned long iflags;
2257
2258         write_lock_irqsave(&sdp->sfd_lock, iflags);
2259         list_del(&sfp->sfd_siblings);
2260         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2261
2262         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2263         schedule_work(&sfp->ew.work);
2264 }
2265
2266 #ifdef CONFIG_SCSI_PROC_FS
2267 static int
2268 sg_idr_max_id(int id, void *p, void *data)
2269 {
2270         int *k = data;
2271
2272         if (*k < id)
2273                 *k = id;
2274
2275         return 0;
2276 }
2277
2278 static int
2279 sg_last_dev(void)
2280 {
2281         int k = -1;
2282         unsigned long iflags;
2283
2284         read_lock_irqsave(&sg_index_lock, iflags);
2285         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2286         read_unlock_irqrestore(&sg_index_lock, iflags);
2287         return k + 1;           /* origin 1 */
2288 }
2289 #endif
2290
2291 /* must be called with sg_index_lock held */
2292 static Sg_device *sg_lookup_dev(int dev)
2293 {
2294         return idr_find(&sg_index_idr, dev);
2295 }
2296
2297 static Sg_device *
2298 sg_get_dev(int dev)
2299 {
2300         struct sg_device *sdp;
2301         unsigned long flags;
2302
2303         read_lock_irqsave(&sg_index_lock, flags);
2304         sdp = sg_lookup_dev(dev);
2305         if (!sdp)
2306                 sdp = ERR_PTR(-ENXIO);
2307         else if (atomic_read(&sdp->detaching)) {
2308                 /* If sdp->detaching, then the refcount may already be 0, in
2309                  * which case it would be a bug to do kref_get().
2310                  */
2311                 sdp = ERR_PTR(-ENODEV);
2312         } else
2313                 kref_get(&sdp->d_ref);
2314         read_unlock_irqrestore(&sg_index_lock, flags);
2315
2316         return sdp;
2317 }
2318
2319 #ifdef CONFIG_SCSI_PROC_FS
2320
2321 static struct proc_dir_entry *sg_proc_sgp = NULL;
2322
2323 static char sg_proc_sg_dirname[] = "scsi/sg";
2324
2325 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2326
2327 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2328 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2329                                   size_t count, loff_t *off);
2330 static const struct file_operations adio_fops = {
2331         .owner = THIS_MODULE,
2332         .open = sg_proc_single_open_adio,
2333         .read = seq_read,
2334         .llseek = seq_lseek,
2335         .write = sg_proc_write_adio,
2336         .release = single_release,
2337 };
2338
2339 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2340 static ssize_t sg_proc_write_dressz(struct file *filp,
2341                 const char __user *buffer, size_t count, loff_t *off);
2342 static const struct file_operations dressz_fops = {
2343         .owner = THIS_MODULE,
2344         .open = sg_proc_single_open_dressz,
2345         .read = seq_read,
2346         .llseek = seq_lseek,
2347         .write = sg_proc_write_dressz,
2348         .release = single_release,
2349 };
2350
2351 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2352 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2353 static const struct file_operations version_fops = {
2354         .owner = THIS_MODULE,
2355         .open = sg_proc_single_open_version,
2356         .read = seq_read,
2357         .llseek = seq_lseek,
2358         .release = single_release,
2359 };
2360
2361 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2362 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2363 static const struct file_operations devhdr_fops = {
2364         .owner = THIS_MODULE,
2365         .open = sg_proc_single_open_devhdr,
2366         .read = seq_read,
2367         .llseek = seq_lseek,
2368         .release = single_release,
2369 };
2370
2371 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2372 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2373 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2374 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2375 static void dev_seq_stop(struct seq_file *s, void *v);
2376 static const struct file_operations dev_fops = {
2377         .owner = THIS_MODULE,
2378         .open = sg_proc_open_dev,
2379         .read = seq_read,
2380         .llseek = seq_lseek,
2381         .release = seq_release,
2382 };
2383 static const struct seq_operations dev_seq_ops = {
2384         .start = dev_seq_start,
2385         .next  = dev_seq_next,
2386         .stop  = dev_seq_stop,
2387         .show  = sg_proc_seq_show_dev,
2388 };
2389
2390 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2391 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2392 static const struct file_operations devstrs_fops = {
2393         .owner = THIS_MODULE,
2394         .open = sg_proc_open_devstrs,
2395         .read = seq_read,
2396         .llseek = seq_lseek,
2397         .release = seq_release,
2398 };
2399 static const struct seq_operations devstrs_seq_ops = {
2400         .start = dev_seq_start,
2401         .next  = dev_seq_next,
2402         .stop  = dev_seq_stop,
2403         .show  = sg_proc_seq_show_devstrs,
2404 };
2405
2406 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2407 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2408 static const struct file_operations debug_fops = {
2409         .owner = THIS_MODULE,
2410         .open = sg_proc_open_debug,
2411         .read = seq_read,
2412         .llseek = seq_lseek,
2413         .release = seq_release,
2414 };
2415 static const struct seq_operations debug_seq_ops = {
2416         .start = dev_seq_start,
2417         .next  = dev_seq_next,
2418         .stop  = dev_seq_stop,
2419         .show  = sg_proc_seq_show_debug,
2420 };
2421
2422
2423 struct sg_proc_leaf {
2424         const char * name;
2425         const struct file_operations * fops;
2426 };
2427
2428 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2429         {"allow_dio", &adio_fops},
2430         {"debug", &debug_fops},
2431         {"def_reserved_size", &dressz_fops},
2432         {"device_hdr", &devhdr_fops},
2433         {"devices", &dev_fops},
2434         {"device_strs", &devstrs_fops},
2435         {"version", &version_fops}
2436 };
2437
2438 static int
2439 sg_proc_init(void)
2440 {
2441         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2442         int k;
2443
2444         sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2445         if (!sg_proc_sgp)
2446                 return 1;
2447         for (k = 0; k < num_leaves; ++k) {
2448                 const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2449                 umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2450                 proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2451         }
2452         return 0;
2453 }
2454
2455 static void
2456 sg_proc_cleanup(void)
2457 {
2458         int k;
2459         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2460
2461         if (!sg_proc_sgp)
2462                 return;
2463         for (k = 0; k < num_leaves; ++k)
2464                 remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2465         remove_proc_entry(sg_proc_sg_dirname, NULL);
2466 }
2467
2468
2469 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2470 {
2471         seq_printf(s, "%d\n", *((int *)s->private));
2472         return 0;
2473 }
2474
2475 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2476 {
2477         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2478 }
2479
2480 static ssize_t
2481 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2482                    size_t count, loff_t *off)
2483 {
2484         int err;
2485         unsigned long num;
2486
2487         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2488                 return -EACCES;
2489         err = kstrtoul_from_user(buffer, count, 0, &num);
2490         if (err)
2491                 return err;
2492         sg_allow_dio = num ? 1 : 0;
2493         return count;
2494 }
2495
2496 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2497 {
2498         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2499 }
2500
2501 static ssize_t
2502 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2503                      size_t count, loff_t *off)
2504 {
2505         int err;
2506         unsigned long k = ULONG_MAX;
2507
2508         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2509                 return -EACCES;
2510
2511         err = kstrtoul_from_user(buffer, count, 0, &k);
2512         if (err)
2513                 return err;
2514         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2515                 sg_big_buff = k;
2516                 return count;
2517         }
2518         return -ERANGE;
2519 }
2520
2521 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2522 {
2523         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2524                    sg_version_date);
2525         return 0;
2526 }
2527
2528 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2529 {
2530         return single_open(file, sg_proc_seq_show_version, NULL);
2531 }
2532
2533 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2534 {
2535         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2536         return 0;
2537 }
2538
2539 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2540 {
2541         return single_open(file, sg_proc_seq_show_devhdr, NULL);
2542 }
2543
2544 struct sg_proc_deviter {
2545         loff_t  index;
2546         size_t  max;
2547 };
2548
2549 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2550 {
2551         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2552
2553         s->private = it;
2554         if (! it)
2555                 return NULL;
2556
2557         it->index = *pos;
2558         it->max = sg_last_dev();
2559         if (it->index >= it->max)
2560                 return NULL;
2561         return it;
2562 }
2563
2564 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2565 {
2566         struct sg_proc_deviter * it = s->private;
2567
2568         *pos = ++it->index;
2569         return (it->index < it->max) ? it : NULL;
2570 }
2571
2572 static void dev_seq_stop(struct seq_file *s, void *v)
2573 {
2574         kfree(s->private);
2575 }
2576
2577 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2578 {
2579         return seq_open(file, &dev_seq_ops);
2580 }
2581
2582 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2583 {
2584         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2585         Sg_device *sdp;
2586         struct scsi_device *scsidp;
2587         unsigned long iflags;
2588
2589         read_lock_irqsave(&sg_index_lock, iflags);
2590         sdp = it ? sg_lookup_dev(it->index) : NULL;
2591         if ((NULL == sdp) || (NULL == sdp->device) ||
2592             (atomic_read(&sdp->detaching)))
2593                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2594         else {
2595                 scsidp = sdp->device;
2596                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2597                               scsidp->host->host_no, scsidp->channel,
2598                               scsidp->id, scsidp->lun, (int) scsidp->type,
2599                               1,
2600                               (int) scsidp->queue_depth,
2601                               (int) atomic_read(&scsidp->device_busy),
2602                               (int) scsi_device_online(scsidp));
2603         }
2604         read_unlock_irqrestore(&sg_index_lock, iflags);
2605         return 0;
2606 }
2607
2608 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2609 {
2610         return seq_open(file, &devstrs_seq_ops);
2611 }
2612
2613 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2614 {
2615         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2616         Sg_device *sdp;
2617         struct scsi_device *scsidp;
2618         unsigned long iflags;
2619
2620         read_lock_irqsave(&sg_index_lock, iflags);
2621         sdp = it ? sg_lookup_dev(it->index) : NULL;
2622         scsidp = sdp ? sdp->device : NULL;
2623         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2624                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2625                            scsidp->vendor, scsidp->model, scsidp->rev);
2626         else
2627                 seq_puts(s, "<no active device>\n");
2628         read_unlock_irqrestore(&sg_index_lock, iflags);
2629         return 0;
2630 }
2631
2632 /* must be called while holding sg_index_lock */
2633 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2634 {
2635         int k, new_interface, blen, usg;
2636         Sg_request *srp;
2637         Sg_fd *fp;
2638         const sg_io_hdr_t *hp;
2639         const char * cp;
2640         unsigned int ms;
2641
2642         k = 0;
2643         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2644                 k++;
2645                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2646                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2647                            "(res)sgat=%d low_dma=%d\n", k,
2648                            jiffies_to_msecs(fp->timeout),
2649                            fp->reserve.bufflen,
2650                            (int) fp->reserve.k_use_sg,
2651                            (int) sdp->device->host->unchecked_isa_dma);
2652                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2653                            (int) fp->cmd_q, (int) fp->force_packid,
2654                            (int) fp->keep_orphan);
2655                 list_for_each_entry(srp, &fp->rq_list, entry) {
2656                         hp = &srp->header;
2657                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2658                         if (srp->res_used) {
2659                                 if (new_interface &&
2660                                     (SG_FLAG_MMAP_IO & hp->flags))
2661                                         cp = "     mmap>> ";
2662                                 else
2663                                         cp = "     rb>> ";
2664                         } else {
2665                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2666                                         cp = "     dio>> ";
2667                                 else
2668                                         cp = "     ";
2669                         }
2670                         seq_puts(s, cp);
2671                         blen = srp->data.bufflen;
2672                         usg = srp->data.k_use_sg;
2673                         seq_puts(s, srp->done ?
2674                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2675                                   : "act:");
2676                         seq_printf(s, " id=%d blen=%d",
2677                                    srp->header.pack_id, blen);
2678                         if (srp->done)
2679                                 seq_printf(s, " dur=%d", hp->duration);
2680                         else {
2681                                 ms = jiffies_to_msecs(jiffies);
2682                                 seq_printf(s, " t_o/elap=%d/%d",
2683                                         (new_interface ? hp->timeout :
2684                                                   jiffies_to_msecs(fp->timeout)),
2685                                         (ms > hp->duration ? ms - hp->duration : 0));
2686                         }
2687                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2688                                    (int) srp->data.cmd_opcode);
2689                 }
2690                 if (list_empty(&fp->rq_list))
2691                         seq_puts(s, "     No requests active\n");
2692                 read_unlock(&fp->rq_list_lock);
2693         }
2694 }
2695
2696 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2697 {
2698         return seq_open(file, &debug_seq_ops);
2699 }
2700
2701 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2702 {
2703         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2704         Sg_device *sdp;
2705         unsigned long iflags;
2706
2707         if (it && (0 == it->index))
2708                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2709                            (int)it->max, sg_big_buff);
2710
2711         read_lock_irqsave(&sg_index_lock, iflags);
2712         sdp = it ? sg_lookup_dev(it->index) : NULL;
2713         if (NULL == sdp)
2714                 goto skip;
2715         read_lock(&sdp->sfd_lock);
2716         if (!list_empty(&sdp->sfds)) {
2717                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2718                 if (atomic_read(&sdp->detaching))
2719                         seq_puts(s, "detaching pending close ");
2720                 else if (sdp->device) {
2721                         struct scsi_device *scsidp = sdp->device;
2722
2723                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2724                                    scsidp->host->host_no,
2725                                    scsidp->channel, scsidp->id,
2726                                    scsidp->lun,
2727                                    scsidp->host->hostt->emulated);
2728                 }
2729                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2730                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2731                 sg_proc_debug_helper(s, sdp);
2732         }
2733         read_unlock(&sdp->sfd_lock);
2734 skip:
2735         read_unlock_irqrestore(&sg_index_lock, iflags);
2736         return 0;
2737 }
2738
2739 #endif                          /* CONFIG_SCSI_PROC_FS */
2740
2741 module_init(init_sg);
2742 module_exit(exit_sg);