drivers/scsi/sg.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  History:
   4  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   5  *           to allow user process control of SCSI devices.
   6  *  Development Sponsored by Killy Corp. NY NY
   7  *
   8  * Original driver (sg.c):
   9  *        Copyright (C) 1992 Lawrence Foard
  10  * Version 2 and 3 extensions to driver:
  11  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  12  */
  13
  14 static int sg_version_num = 30536;      /* 2 digits for each component */
  15 #define SG_VERSION_STR "3.5.36"
  16
  17 /*
  18  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  19  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  20  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  21  *        (otherwise the macros compile to empty statements).
  22  *
  23  */
  24 #include <linux/module.h>
  25
  26 #include <linux/fs.h>
  27 #include <linux/kernel.h>
  28 #include <linux/sched.h>
  29 #include <linux/string.h>
  30 #include <linux/mm.h>
  31 #include <linux/errno.h>
  32 #include <linux/mtio.h>
  33 #include <linux/ioctl.h>
  34 #include <linux/major.h>
  35 #include <linux/slab.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/init.h>
  38 #include <linux/poll.h>
  39 #include <linux/moduleparam.h>
  40 #include <linux/cdev.h>
  41 #include <linux/idr.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/delay.h>
  45 #include <linux/blktrace_api.h>
  46 #include <linux/mutex.h>
  47 #include <linux/atomic.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/uio.h>
  50 #include <linux/cred.h> /* for sg_check_file_access() */
  51
  52 #include <scsi/scsi.h>
  53 #include <scsi/scsi_cmnd.h>
  54 #include <scsi/scsi_dbg.h>
  55 #include <scsi/scsi_device.h>
  56 #include <scsi/scsi_driver.h>
  57 #include <scsi/scsi_eh.h>
  58 #include <scsi/scsi_host.h>
  59 #include <scsi/scsi_ioctl.h>
  60 #include <scsi/scsi_tcq.h>
  61 #include <scsi/sg.h>
  62
  63 #include "scsi_logging.h"
  64
  65 #ifdef CONFIG_SCSI_PROC_FS
  66 #include <linux/proc_fs.h>
  67 static char *sg_version_date = "20140603";
  68
  69 static int sg_proc_init(void);
  70 #endif
  71
  72 #define SG_ALLOW_DIO_DEF 0
  73
  74 #define SG_MAX_DEVS 32768
  75
  76 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  77  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  78  * than 16 bytes are "variable length" whose length is a multiple of 4
  79  */
  80 #define SG_MAX_CDB_SIZE 252
  81
  82 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  83
  84 static int sg_big_buff = SG_DEF_RESERVED_SIZE;
  85 /* N.B. This variable is readable and writeable via
  86    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  87    of this size (or less if there is not enough memory) will be reserved
  88    for use by this file descriptor. [Deprecated usage: this variable is also
  89    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
  90    the kernel (i.e. it is not a module).] */
  91 static int def_reserved_size = -1;      /* picks up init parameter */
  92 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
  93
  94 static int scatter_elem_sz = SG_SCATTER_SZ;
  95 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
  96
  97 #define SG_SECTOR_SZ 512
  98
  99 static int sg_add_device(struct device *, struct class_interface *);
 100 static void sg_remove_device(struct device *, struct class_interface *);
 101
 102 static DEFINE_IDR(sg_index_idr);
 103 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
 104                                                            file descriptor list for device */
 105
 106 static struct class_interface sg_interface = {
 107         .add_dev        = sg_add_device,
 108         .remove_dev     = sg_remove_device,
 109 };
 110
 111 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 112         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 113         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 114         unsigned bufflen;       /* Size of (aggregate) data buffer */
 115         struct page **pages;
 116         int page_order;
 117         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 118         unsigned char cmd_opcode; /* first byte of command */
 119 } Sg_scatter_hold;
 120
 121 struct sg_device;               /* forward declarations */
 122 struct sg_fd;
 123
 124 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 125         struct list_head entry; /* list entry */
 126         struct sg_fd *parentfp; /* NULL -> not in use */
 127         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 128         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 129         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 130         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 131         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 132         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 133         /* done protected by rq_list_lock */
 134         char done;              /* 0->before bh, 1->before read, 2->read */
 135         struct request *rq;
 136         struct bio *bio;
 137         struct execute_work ew;
 138 } Sg_request;
 139
 140 typedef struct sg_fd {          /* holds the state of a file descriptor */
 141         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 142         struct sg_device *parentdp;     /* owning device */
 143         wait_queue_head_t read_wait;    /* queue read until command done */
 144         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 145         struct mutex f_mutex;   /* protect against changes in this fd */
 146         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 147         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 148         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 149         struct list_head rq_list; /* head of request list */
 150         struct fasync_struct *async_qp; /* used by asynchronous notification */
 151         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 152         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 153         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 154         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 155         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 156         char mmap_called;       /* 0 -> mmap() never called on this fd */
 157         char res_in_use;        /* 1 -> 'reserve' array in use */
 158         struct kref f_ref;
 159         struct execute_work ew;
 160 } Sg_fd;
 161
 162 typedef struct sg_device { /* holds the state of each scsi generic device */
 163         struct scsi_device *device;
 164         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 165         struct mutex open_rel_lock;     /* held when in open() or release() */
 166         int sg_tablesize;       /* adapter's max scatter-gather table size */
 167         u32 index;              /* device index number */
 168         struct list_head sfds;
 169         rwlock_t sfd_lock;      /* protect access to sfd list */
 170         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 171         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 172         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 173         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 174         char name[DISK_NAME_LEN];
 175         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 176         struct kref d_ref;
 177 } Sg_device;
 178
 179 /* tasklet or soft irq callback */
 180 static void sg_rq_end_io(struct request *rq, blk_status_t status);
 181 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 182 static int sg_finish_rem_req(Sg_request * srp);
 183 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 184 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 185                            Sg_request * srp);
 186 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 187                         const char __user *buf, size_t count, int blocking,
 188                         int read_only, int sg_io_owned, Sg_request **o_srp);
 189 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 190                            unsigned char *cmnd, int timeout, int blocking);
 191 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 192 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 193 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 194 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 195 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 196 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 197 static void sg_remove_sfp(struct kref *);
 198 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 199 static Sg_request *sg_add_request(Sg_fd * sfp);
 200 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 201 static Sg_device *sg_get_dev(int dev);
 202 static void sg_device_destroy(struct kref *kref);
 203
 204 #define SZ_SG_HEADER sizeof(struct sg_header)
 205 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 206 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 207 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 208
 209 #define sg_printk(prefix, sdp, fmt, a...) \
 210         sdev_prefix_printk(prefix, (sdp)->device, (sdp)->name, fmt, ##a)
 211
 212 /*
 213  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 214  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 215  * to trigger read() and write() calls from various contexts with elevated
 216  * privileges. This can lead to kernel memory corruption (e.g. if these
 217  * interfaces are called through splice()) and privilege escalation inside
 218  * userspace (e.g. if a process with access to such a device passes a file
 219  * descriptor to a SUID binary as stdin/stdout/stderr).
 220  *
 221  * This function provides protection for the legacy API by restricting the
 222  * calling context.
 223  */
 224 static int sg_check_file_access(struct file *filp, const char *caller)
 225 {
 226         if (filp->f_cred != current_real_cred()) {
 227                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 228                         caller, task_tgid_vnr(current), current->comm);
 229                 return -EPERM;
 230         }
 231         return 0;
 232 }
 233
 234 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 235 {
 236         struct sg_fd *sfp = filp->private_data;
 237
 238         if (sfp->parentdp->device->type == TYPE_SCANNER)
 239                 return 0;
 240         if (!scsi_cmd_allowed(cmd, filp->f_mode))
 241                 return -EPERM;
 242         return 0;
 243 }
 244
 245 static int
 246 open_wait(Sg_device *sdp, int flags)
 247 {
 248         int retval = 0;
 249
 250         if (flags & O_EXCL) {
 251                 while (sdp->open_cnt > 0) {
 252                         mutex_unlock(&sdp->open_rel_lock);
 253                         retval = wait_event_interruptible(sdp->open_wait,
 254                                         (atomic_read(&sdp->detaching) ||
 255                                          !sdp->open_cnt));
 256                         mutex_lock(&sdp->open_rel_lock);
 257
 258                         if (retval) /* -ERESTARTSYS */
 259                                 return retval;
 260                         if (atomic_read(&sdp->detaching))
 261                                 return -ENODEV;
 262                 }
 263         } else {
 264                 while (sdp->exclude) {
 265                         mutex_unlock(&sdp->open_rel_lock);
 266                         retval = wait_event_interruptible(sdp->open_wait,
 267                                         (atomic_read(&sdp->detaching) ||
 268                                          !sdp->exclude));
 269                         mutex_lock(&sdp->open_rel_lock);
 270
 271                         if (retval) /* -ERESTARTSYS */
 272                                 return retval;
 273                         if (atomic_read(&sdp->detaching))
 274                                 return -ENODEV;
 275                 }
 276         }
 277
 278         return retval;
 279 }
 280
 281 /* Returns 0 on success, else a negated errno value */
 282 static int
 283 sg_open(struct inode *inode, struct file *filp)
 284 {
 285         int dev = iminor(inode);
 286         int flags = filp->f_flags;
 287         struct request_queue *q;
 288         Sg_device *sdp;
 289         Sg_fd *sfp;
 290         int retval;
 291
 292         nonseekable_open(inode, filp);
 293         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 294                 return -EPERM; /* Can't lock it with read only access */
 295         sdp = sg_get_dev(dev);
 296         if (IS_ERR(sdp))
 297                 return PTR_ERR(sdp);
 298
 299         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 300                                       "sg_open: flags=0x%x\n", flags));
 301
 302         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 303         /* Prevent the device driver from vanishing while we sleep */
 304         retval = scsi_device_get(sdp->device);
 305         if (retval)
 306                 goto sg_put;
 307
 308         retval = scsi_autopm_get_device(sdp->device);
 309         if (retval)
 310                 goto sdp_put;
 311
 312         /* scsi_block_when_processing_errors() may block so bypass
 313          * check if O_NONBLOCK. Permits SCSI commands to be issued
 314          * during error recovery. Tread carefully. */
 315         if (!((flags & O_NONBLOCK) ||
 316               scsi_block_when_processing_errors(sdp->device))) {
 317                 retval = -ENXIO;
 318                 /* we are in error recovery for this device */
 319                 goto error_out;
 320         }
 321
 322         mutex_lock(&sdp->open_rel_lock);
 323         if (flags & O_NONBLOCK) {
 324                 if (flags & O_EXCL) {
 325                         if (sdp->open_cnt > 0) {
 326                                 retval = -EBUSY;
 327                                 goto error_mutex_locked;
 328                         }
 329                 } else {
 330                         if (sdp->exclude) {
 331                                 retval = -EBUSY;
 332                                 goto error_mutex_locked;
 333                         }
 334                 }
 335         } else {
 336                 retval = open_wait(sdp, flags);
 337                 if (retval) /* -ERESTARTSYS or -ENODEV */
 338                         goto error_mutex_locked;
 339         }
 340
 341         /* N.B. at this point we are holding the open_rel_lock */
 342         if (flags & O_EXCL)
 343                 sdp->exclude = true;
 344
 345         if (sdp->open_cnt < 1) {  /* no existing opens */
 346                 sdp->sgdebug = 0;
 347                 q = sdp->device->request_queue;
 348                 sdp->sg_tablesize = queue_max_segments(q);
 349         }
 350         sfp = sg_add_sfp(sdp);
 351         if (IS_ERR(sfp)) {
 352                 retval = PTR_ERR(sfp);
 353                 goto out_undo;
 354         }
 355
 356         filp->private_data = sfp;
 357         sdp->open_cnt++;
 358         mutex_unlock(&sdp->open_rel_lock);
 359
 360         retval = 0;
 361 sg_put:
 362         kref_put(&sdp->d_ref, sg_device_destroy);
 363         return retval;
 364
 365 out_undo:
 366         if (flags & O_EXCL) {
 367                 sdp->exclude = false;   /* undo if error */
 368                 wake_up_interruptible(&sdp->open_wait);
 369         }
 370 error_mutex_locked:
 371         mutex_unlock(&sdp->open_rel_lock);
 372 error_out:
 373         scsi_autopm_put_device(sdp->device);
 374 sdp_put:
 375         scsi_device_put(sdp->device);
 376         goto sg_put;
 377 }
 378
 379 /* Release resources associated with a successful sg_open()
 380  * Returns 0 on success, else a negated errno value */
 381 static int
 382 sg_release(struct inode *inode, struct file *filp)
 383 {
 384         Sg_device *sdp;
 385         Sg_fd *sfp;
 386
 387         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 388                 return -ENXIO;
 389         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 390
 391         mutex_lock(&sdp->open_rel_lock);
 392         scsi_autopm_put_device(sdp->device);
 393         kref_put(&sfp->f_ref, sg_remove_sfp);
 394         sdp->open_cnt--;
 395
 396         /* possibly many open()s waiting on exlude clearing, start many;
 397          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 398         if (sdp->exclude) {
 399                 sdp->exclude = false;
 400                 wake_up_interruptible_all(&sdp->open_wait);
 401         } else if (0 == sdp->open_cnt) {
 402                 wake_up_interruptible(&sdp->open_wait);
 403         }
 404         mutex_unlock(&sdp->open_rel_lock);
 405         return 0;
 406 }
 407
 408 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 409 {
 410         struct sg_header __user *old_hdr = buf;
 411         int reply_len;
 412
 413         if (count >= SZ_SG_HEADER) {
 414                 /* negative reply_len means v3 format, otherwise v1/v2 */
 415                 if (get_user(reply_len, &old_hdr->reply_len))
 416                         return -EFAULT;
 417
 418                 if (reply_len >= 0)
 419                         return get_user(*pack_id, &old_hdr->pack_id);
 420
 421                 if (in_compat_syscall() &&
 422                     count >= sizeof(struct compat_sg_io_hdr)) {
 423                         struct compat_sg_io_hdr __user *hp = buf;
 424
 425                         return get_user(*pack_id, &hp->pack_id);
 426                 }
 427
 428                 if (count >= sizeof(struct sg_io_hdr)) {
 429                         struct sg_io_hdr __user *hp = buf;
 430
 431                         return get_user(*pack_id, &hp->pack_id);
 432                 }
 433         }
 434
 435         /* no valid header was passed, so ignore the pack_id */
 436         *pack_id = -1;
 437         return 0;
 438 }
 439
 440 static ssize_t
 441 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 442 {
 443         Sg_device *sdp;
 444         Sg_fd *sfp;
 445         Sg_request *srp;
 446         int req_pack_id = -1;
 447         sg_io_hdr_t *hp;
 448         struct sg_header *old_hdr;
 449         int retval;
 450
 451         /*
 452          * This could cause a response to be stranded. Close the associated
 453          * file descriptor to free up any resources being held.
 454          */
 455         retval = sg_check_file_access(filp, __func__);
 456         if (retval)
 457                 return retval;
 458
 459         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 460                 return -ENXIO;
 461         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 462                                       "sg_read: count=%d\n", (int) count));
 463
 464         if (sfp->force_packid)
 465                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 466         if (retval)
 467                 return retval;
 468
 469         srp = sg_get_rq_mark(sfp, req_pack_id);
 470         if (!srp) {             /* now wait on packet to arrive */
 471                 if (atomic_read(&sdp->detaching))
 472                         return -ENODEV;
 473                 if (filp->f_flags & O_NONBLOCK)
 474                         return -EAGAIN;
 475                 retval = wait_event_interruptible(sfp->read_wait,
 476                         (atomic_read(&sdp->detaching) ||
 477                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 478                 if (atomic_read(&sdp->detaching))
 479                         return -ENODEV;
 480                 if (retval)
 481                         /* -ERESTARTSYS as signal hit process */
 482                         return retval;
 483         }
 484         if (srp->header.interface_id != '\0')
 485                 return sg_new_read(sfp, buf, count, srp);
 486
 487         hp = &srp->header;
 488         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 489         if (!old_hdr)
 490                 return -ENOMEM;
 491
 492         old_hdr->reply_len = (int) hp->timeout;
 493         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 494         old_hdr->pack_id = hp->pack_id;
 495         old_hdr->twelve_byte =
 496             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 497         old_hdr->target_status = hp->masked_status;
 498         old_hdr->host_status = hp->host_status;
 499         old_hdr->driver_status = hp->driver_status;
 500         if ((CHECK_CONDITION & hp->masked_status) ||
 501             (srp->sense_b[0] & 0x70) == 0x70) {
 502                 old_hdr->driver_status = DRIVER_SENSE;
 503                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 504                        sizeof (old_hdr->sense_buffer));
 505         }
 506         switch (hp->host_status) {
 507         /* This setup of 'result' is for backward compatibility and is best
 508            ignored by the user who should use target, host + driver status */
 509         case DID_OK:
 510         case DID_PASSTHROUGH:
 511         case DID_SOFT_ERROR:
 512                 old_hdr->result = 0;
 513                 break;
 514         case DID_NO_CONNECT:
 515         case DID_BUS_BUSY:
 516         case DID_TIME_OUT:
 517                 old_hdr->result = EBUSY;
 518                 break;
 519         case DID_BAD_TARGET:
 520         case DID_ABORT:
 521         case DID_PARITY:
 522         case DID_RESET:
 523         case DID_BAD_INTR:
 524                 old_hdr->result = EIO;
 525                 break;
 526         case DID_ERROR:
 527                 old_hdr->result = (srp->sense_b[0] == 0 &&
 528                                   hp->masked_status == GOOD) ? 0 : EIO;
 529                 break;
 530         default:
 531                 old_hdr->result = EIO;
 532                 break;
 533         }
 534
 535         /* Now copy the result back to the user buffer.  */
 536         if (count >= SZ_SG_HEADER) {
 537                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 538                         retval = -EFAULT;
 539                         goto free_old_hdr;
 540                 }
 541                 buf += SZ_SG_HEADER;
 542                 if (count > old_hdr->reply_len)
 543                         count = old_hdr->reply_len;
 544                 if (count > SZ_SG_HEADER) {
 545                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 546                                 retval = -EFAULT;
 547                                 goto free_old_hdr;
 548                         }
 549                 }
 550         } else
 551                 count = (old_hdr->result == 0) ? 0 : -EIO;
 552         sg_finish_rem_req(srp);
 553         sg_remove_request(sfp, srp);
 554         retval = count;
 555 free_old_hdr:
 556         kfree(old_hdr);
 557         return retval;
 558 }
 559
 560 static ssize_t
 561 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 562 {
 563         sg_io_hdr_t *hp = &srp->header;
 564         int err = 0, err2;
 565         int len;
 566
 567         if (in_compat_syscall()) {
 568                 if (count < sizeof(struct compat_sg_io_hdr)) {
 569                         err = -EINVAL;
 570                         goto err_out;
 571                 }
 572         } else if (count < SZ_SG_IO_HDR) {
 573                 err = -EINVAL;
 574                 goto err_out;
 575         }
 576         hp->sb_len_wr = 0;
 577         if ((hp->mx_sb_len > 0) && hp->sbp) {
 578                 if ((CHECK_CONDITION & hp->masked_status) ||
 579                     (srp->sense_b[0] & 0x70) == 0x70) {
 580                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 581                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 582                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 583                         len = (len > sb_len) ? sb_len : len;
 584                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 585                                 err = -EFAULT;
 586                                 goto err_out;
 587                         }
 588                         hp->driver_status = DRIVER_SENSE;
 589                         hp->sb_len_wr = len;
 590                 }
 591         }
 592         if (hp->masked_status || hp->host_status || hp->driver_status)
 593                 hp->info |= SG_INFO_CHECK;
 594         err = put_sg_io_hdr(hp, buf);
 595 err_out:
 596         err2 = sg_finish_rem_req(srp);
 597         sg_remove_request(sfp, srp);
 598         return err ? : err2 ? : count;
 599 }
 600
 601 static ssize_t
 602 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 603 {
 604         int mxsize, cmd_size, k;
 605         int input_size, blocking;
 606         unsigned char opcode;
 607         Sg_device *sdp;
 608         Sg_fd *sfp;
 609         Sg_request *srp;
 610         struct sg_header old_hdr;
 611         sg_io_hdr_t *hp;
 612         unsigned char cmnd[SG_MAX_CDB_SIZE];
 613         int retval;
 614
 615         retval = sg_check_file_access(filp, __func__);
 616         if (retval)
 617                 return retval;
 618
 619         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 620                 return -ENXIO;
 621         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 622                                       "sg_write: count=%d\n", (int) count));
 623         if (atomic_read(&sdp->detaching))
 624                 return -ENODEV;
 625         if (!((filp->f_flags & O_NONBLOCK) ||
 626               scsi_block_when_processing_errors(sdp->device)))
 627                 return -ENXIO;
 628
 629         if (count < SZ_SG_HEADER)
 630                 return -EIO;
 631         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 632                 return -EFAULT;
 633         blocking = !(filp->f_flags & O_NONBLOCK);
 634         if (old_hdr.reply_len < 0)
 635                 return sg_new_write(sfp, filp, buf, count,
 636                                     blocking, 0, 0, NULL);
 637         if (count < (SZ_SG_HEADER + 6))
 638                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 639
 640         buf += SZ_SG_HEADER;
 641         if (get_user(opcode, buf))
 642                 return -EFAULT;
 643
 644         if (!(srp = sg_add_request(sfp))) {
 645                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 646                                               "sg_write: queue full\n"));
 647                 return -EDOM;
 648         }
 649         mutex_lock(&sfp->f_mutex);
 650         if (sfp->next_cmd_len > 0) {
 651                 cmd_size = sfp->next_cmd_len;
 652                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 653         } else {
 654                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 655                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 656                         cmd_size = 12;
 657         }
 658         mutex_unlock(&sfp->f_mutex);
 659         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 660                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 661 /* Determine buffer size.  */
 662         input_size = count - cmd_size;
 663         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 664         mxsize -= SZ_SG_HEADER;
 665         input_size -= SZ_SG_HEADER;
 666         if (input_size < 0) {
 667                 sg_remove_request(sfp, srp);
 668                 return -EIO;    /* User did not pass enough bytes for this command. */
 669         }
 670         hp = &srp->header;
 671         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 672         hp->cmd_len = (unsigned char) cmd_size;
 673         hp->iovec_count = 0;
 674         hp->mx_sb_len = 0;
 675         if (input_size > 0)
 676                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 677                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 678         else
 679                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 680         hp->dxfer_len = mxsize;
 681         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 682             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 683                 hp->dxferp = (char __user *)buf + cmd_size;
 684         else
 685                 hp->dxferp = NULL;
 686         hp->sbp = NULL;
 687         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 688         hp->flags = input_size; /* structure abuse ... */
 689         hp->pack_id = old_hdr.pack_id;
 690         hp->usr_ptr = NULL;
 691         if (copy_from_user(cmnd, buf, cmd_size)) {
 692                 sg_remove_request(sfp, srp);
 693                 return -EFAULT;
 694         }
 695         /*
 696          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 697          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 698          * is a non-zero input_size, so emit a warning.
 699          */
 700         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 701                 printk_ratelimited(KERN_WARNING
 702                                    "sg_write: data in/out %d/%d bytes "
 703                                    "for SCSI command 0x%x-- guessing "
 704                                    "data in;\n   program %s not setting "
 705                                    "count and/or reply_len properly\n",
 706                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 707                                    input_size, (unsigned int) cmnd[0],
 708                                    current->comm);
 709         }
 710         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 711         return (k < 0) ? k : count;
 712 }
 713
 714 static ssize_t
 715 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 716                  size_t count, int blocking, int read_only, int sg_io_owned,
 717                  Sg_request **o_srp)
 718 {
 719         int k;
 720         Sg_request *srp;
 721         sg_io_hdr_t *hp;
 722         unsigned char cmnd[SG_MAX_CDB_SIZE];
 723         int timeout;
 724         unsigned long ul_timeout;
 725
 726         if (count < SZ_SG_IO_HDR)
 727                 return -EINVAL;
 728
 729         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 730         if (!(srp = sg_add_request(sfp))) {
 731                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 732                                               "sg_new_write: queue full\n"));
 733                 return -EDOM;
 734         }
 735         srp->sg_io_owned = sg_io_owned;
 736         hp = &srp->header;
 737         if (get_sg_io_hdr(hp, buf)) {
 738                 sg_remove_request(sfp, srp);
 739                 return -EFAULT;
 740         }
 741         if (hp->interface_id != 'S') {
 742                 sg_remove_request(sfp, srp);
 743                 return -ENOSYS;
 744         }
 745         if (hp->flags & SG_FLAG_MMAP_IO) {
 746                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 747                         sg_remove_request(sfp, srp);
 748                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 749                 }
 750                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 751                         sg_remove_request(sfp, srp);
 752                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 753                 }
 754                 if (sfp->res_in_use) {
 755                         sg_remove_request(sfp, srp);
 756                         return -EBUSY;  /* reserve buffer already being used */
 757                 }
 758         }
 759         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 760         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 761         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 762                 sg_remove_request(sfp, srp);
 763                 return -EMSGSIZE;
 764         }
 765         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 766                 sg_remove_request(sfp, srp);
 767                 return -EFAULT;
 768         }
 769         if (read_only && sg_allow_access(file, cmnd)) {
 770                 sg_remove_request(sfp, srp);
 771                 return -EPERM;
 772         }
 773         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 774         if (k < 0)
 775                 return k;
 776         if (o_srp)
 777                 *o_srp = srp;
 778         return count;
 779 }
 780
 781 static int
 782 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 783                 unsigned char *cmnd, int timeout, int blocking)
 784 {
 785         int k, at_head;
 786         Sg_device *sdp = sfp->parentdp;
 787         sg_io_hdr_t *hp = &srp->header;
 788
 789         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 790         hp->status = 0;
 791         hp->masked_status = 0;
 792         hp->msg_status = 0;
 793         hp->info = 0;
 794         hp->host_status = 0;
 795         hp->driver_status = 0;
 796         hp->resid = 0;
 797         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 798                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 799                         (int) cmnd[0], (int) hp->cmd_len));
 800
 801         if (hp->dxfer_len >= SZ_256M) {
 802                 sg_remove_request(sfp, srp);
 803                 return -EINVAL;
 804         }
 805
 806         k = sg_start_req(srp, cmnd);
 807         if (k) {
 808                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 809                         "sg_common_write: start_req err=%d\n", k));
 810                 sg_finish_rem_req(srp);
 811                 sg_remove_request(sfp, srp);
 812                 return k;       /* probably out of space --> ENOMEM */
 813         }
 814         if (atomic_read(&sdp->detaching)) {
 815                 if (srp->bio) {
 816                         blk_mq_free_request(srp->rq);
 817                         srp->rq = NULL;
 818                 }
 819
 820                 sg_finish_rem_req(srp);
 821                 sg_remove_request(sfp, srp);
 822                 return -ENODEV;
 823         }
 824
 825         hp->duration = jiffies_to_msecs(jiffies);
 826         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 827             (SG_FLAG_Q_AT_TAIL & hp->flags))
 828                 at_head = 0;
 829         else
 830                 at_head = 1;
 831
 832         srp->rq->timeout = timeout;
 833         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 834         blk_execute_rq_nowait(srp->rq, at_head, sg_rq_end_io);
 835         return 0;
 836 }
 837
 838 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 839 {
 840         unsigned long flags;
 841         int ret;
 842
 843         read_lock_irqsave(&sfp->rq_list_lock, flags);
 844         ret = srp->done;
 845         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 846         return ret;
 847 }
 848
 849 static int max_sectors_bytes(struct request_queue *q)
 850 {
 851         unsigned int max_sectors = queue_max_sectors(q);
 852
 853         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 854
 855         return max_sectors << 9;
 856 }
 857
 858 static void
 859 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 860 {
 861         Sg_request *srp;
 862         int val;
 863         unsigned int ms;
 864
 865         val = 0;
 866         list_for_each_entry(srp, &sfp->rq_list, entry) {
 867                 if (val >= SG_MAX_QUEUE)
 868                         break;
 869                 rinfo[val].req_state = srp->done + 1;
 870                 rinfo[val].problem =
 871                         srp->header.masked_status &
 872                         srp->header.host_status &
 873                         srp->header.driver_status;
 874                 if (srp->done)
 875                         rinfo[val].duration =
 876                                 srp->header.duration;
 877                 else {
 878                         ms = jiffies_to_msecs(jiffies);
 879                         rinfo[val].duration =
 880                                 (ms > srp->header.duration) ?
 881                                 (ms - srp->header.duration) : 0;
 882                 }
 883                 rinfo[val].orphan = srp->orphan;
 884                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 885                 rinfo[val].pack_id = srp->header.pack_id;
 886                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 887                 val++;
 888         }
 889 }
 890
 891 #ifdef CONFIG_COMPAT
 892 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 893         char req_state;
 894         char orphan;
 895         char sg_io_owned;
 896         char problem;
 897         int pack_id;
 898         compat_uptr_t usr_ptr;
 899         unsigned int duration;
 900         int unused;
 901 };
 902
 903 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 904                                     struct sg_req_info *rinfo)
 905 {
 906         int i;
 907         for (i = 0; i < SG_MAX_QUEUE; i++) {
 908                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 909                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 910                     put_user(rinfo[i].duration, &o[i].duration) ||
 911                     put_user(rinfo[i].unused, &o[i].unused))
 912                         return -EFAULT;
 913         }
 914         return 0;
 915 }
 916 #endif
 917
 918 static long
 919 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 920                 unsigned int cmd_in, void __user *p)
 921 {
 922         int __user *ip = p;
 923         int result, val, read_only;
 924         Sg_request *srp;
 925         unsigned long iflags;
 926
 927         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 928                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 929         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 930
 931         switch (cmd_in) {
 932         case SG_IO:
 933                 if (atomic_read(&sdp->detaching))
 934                         return -ENODEV;
 935                 if (!scsi_block_when_processing_errors(sdp->device))
 936                         return -ENXIO;
 937                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 938                                  1, read_only, 1, &srp);
 939                 if (result < 0)
 940                         return result;
 941                 result = wait_event_interruptible(sfp->read_wait,
 942                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 943                 if (atomic_read(&sdp->detaching))
 944                         return -ENODEV;
 945                 write_lock_irq(&sfp->rq_list_lock);
 946                 if (srp->done) {
 947                         srp->done = 2;
 948                         write_unlock_irq(&sfp->rq_list_lock);
 949                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 950                         return (result < 0) ? result : 0;
 951                 }
 952                 srp->orphan = 1;
 953                 write_unlock_irq(&sfp->rq_list_lock);
 954                 return result;  /* -ERESTARTSYS because signal hit process */
 955         case SG_SET_TIMEOUT:
 956                 result = get_user(val, ip);
 957                 if (result)
 958                         return result;
 959                 if (val < 0)
 960                         return -EIO;
 961                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 962                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 963                                     INT_MAX);
 964                 sfp->timeout_user = val;
 965                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 966
 967                 return 0;
 968         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 969                                 /* strange ..., for backward compatibility */
 970                 return sfp->timeout_user;
 971         case SG_SET_FORCE_LOW_DMA:
 972                 /*
 973                  * N.B. This ioctl never worked properly, but failed to
 974                  * return an error value. So returning '0' to keep compability
 975                  * with legacy applications.
 976                  */
 977                 return 0;
 978         case SG_GET_LOW_DMA:
 979                 return put_user(0, ip);
 980         case SG_GET_SCSI_ID:
 981                 {
 982                         sg_scsi_id_t v;
 983
 984                         if (atomic_read(&sdp->detaching))
 985                                 return -ENODEV;
 986                         memset(&v, 0, sizeof(v));
 987                         v.host_no = sdp->device->host->host_no;
 988                         v.channel = sdp->device->channel;
 989                         v.scsi_id = sdp->device->id;
 990                         v.lun = sdp->device->lun;
 991                         v.scsi_type = sdp->device->type;
 992                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 993                         v.d_queue_depth = sdp->device->queue_depth;
 994                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 995                                 return -EFAULT;
 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                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1006                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1007                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1008                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1009                                                        iflags);
1010                                 return put_user(srp->header.pack_id, ip);
1011                         }
1012                 }
1013                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1014                 return put_user(-1, ip);
1015         case SG_GET_NUM_WAITING:
1016                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1017                 val = 0;
1018                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1019                         if ((1 == srp->done) && (!srp->sg_io_owned))
1020                                 ++val;
1021                 }
1022                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1023                 return put_user(val, ip);
1024         case SG_GET_SG_TABLESIZE:
1025                 return put_user(sdp->sg_tablesize, ip);
1026         case SG_SET_RESERVED_SIZE:
1027                 result = get_user(val, ip);
1028                 if (result)
1029                         return result;
1030                 if (val < 0)
1031                         return -EINVAL;
1032                 val = min_t(int, val,
1033                             max_sectors_bytes(sdp->device->request_queue));
1034                 mutex_lock(&sfp->f_mutex);
1035                 if (val != sfp->reserve.bufflen) {
1036                         if (sfp->mmap_called ||
1037                             sfp->res_in_use) {
1038                                 mutex_unlock(&sfp->f_mutex);
1039                                 return -EBUSY;
1040                         }
1041
1042                         sg_remove_scat(sfp, &sfp->reserve);
1043                         sg_build_reserve(sfp, val);
1044                 }
1045                 mutex_unlock(&sfp->f_mutex);
1046                 return 0;
1047         case SG_GET_RESERVED_SIZE:
1048                 val = min_t(int, sfp->reserve.bufflen,
1049                             max_sectors_bytes(sdp->device->request_queue));
1050                 return put_user(val, ip);
1051         case SG_SET_COMMAND_Q:
1052                 result = get_user(val, ip);
1053                 if (result)
1054                         return result;
1055                 sfp->cmd_q = val ? 1 : 0;
1056                 return 0;
1057         case SG_GET_COMMAND_Q:
1058                 return put_user((int) sfp->cmd_q, ip);
1059         case SG_SET_KEEP_ORPHAN:
1060                 result = get_user(val, ip);
1061                 if (result)
1062                         return result;
1063                 sfp->keep_orphan = val;
1064                 return 0;
1065         case SG_GET_KEEP_ORPHAN:
1066                 return put_user((int) sfp->keep_orphan, ip);
1067         case SG_NEXT_CMD_LEN:
1068                 result = get_user(val, ip);
1069                 if (result)
1070                         return result;
1071                 if (val > SG_MAX_CDB_SIZE)
1072                         return -ENOMEM;
1073                 sfp->next_cmd_len = (val > 0) ? val : 0;
1074                 return 0;
1075         case SG_GET_VERSION_NUM:
1076                 return put_user(sg_version_num, ip);
1077         case SG_GET_ACCESS_COUNT:
1078                 /* faked - we don't have a real access count anymore */
1079                 val = (sdp->device ? 1 : 0);
1080                 return put_user(val, ip);
1081         case SG_GET_REQUEST_TABLE:
1082                 {
1083                         sg_req_info_t *rinfo;
1084
1085                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1086                                         GFP_KERNEL);
1087                         if (!rinfo)
1088                                 return -ENOMEM;
1089                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1090                         sg_fill_request_table(sfp, rinfo);
1091                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1092         #ifdef CONFIG_COMPAT
1093                         if (in_compat_syscall())
1094                                 result = put_compat_request_table(p, rinfo);
1095                         else
1096         #endif
1097                                 result = copy_to_user(p, rinfo,
1098                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1099                         result = result ? -EFAULT : 0;
1100                         kfree(rinfo);
1101                         return result;
1102                 }
1103         case SG_EMULATED_HOST:
1104                 if (atomic_read(&sdp->detaching))
1105                         return -ENODEV;
1106                 return put_user(sdp->device->host->hostt->emulated, ip);
1107         case SCSI_IOCTL_SEND_COMMAND:
1108                 if (atomic_read(&sdp->detaching))
1109                         return -ENODEV;
1110                 return scsi_ioctl(sdp->device, filp->f_mode, cmd_in, p);
1111         case SG_SET_DEBUG:
1112                 result = get_user(val, ip);
1113                 if (result)
1114                         return result;
1115                 sdp->sgdebug = (char) val;
1116                 return 0;
1117         case BLKSECTGET:
1118                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1119                                 ip);
1120         case BLKTRACESETUP:
1121                 return blk_trace_setup(sdp->device->request_queue, sdp->name,
1122                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1123                                        NULL, p);
1124         case BLKTRACESTART:
1125                 return blk_trace_startstop(sdp->device->request_queue, 1);
1126         case BLKTRACESTOP:
1127                 return blk_trace_startstop(sdp->device->request_queue, 0);
1128         case BLKTRACETEARDOWN:
1129                 return blk_trace_remove(sdp->device->request_queue);
1130         case SCSI_IOCTL_GET_IDLUN:
1131         case SCSI_IOCTL_GET_BUS_NUMBER:
1132         case SCSI_IOCTL_PROBE_HOST:
1133         case SG_GET_TRANSFORM:
1134         case SG_SCSI_RESET:
1135                 if (atomic_read(&sdp->detaching))
1136                         return -ENODEV;
1137                 break;
1138         default:
1139                 if (read_only)
1140                         return -EPERM;  /* don't know so take safe approach */
1141                 break;
1142         }
1143
1144         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1145                         cmd_in, filp->f_flags & O_NDELAY);
1146         if (result)
1147                 return result;
1148
1149         return -ENOIOCTLCMD;
1150 }
1151
1152 static long
1153 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1154 {
1155         void __user *p = (void __user *)arg;
1156         Sg_device *sdp;
1157         Sg_fd *sfp;
1158         int ret;
1159
1160         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1161                 return -ENXIO;
1162
1163         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1164         if (ret != -ENOIOCTLCMD)
1165                 return ret;
1166         return scsi_ioctl(sdp->device, filp->f_mode, cmd_in, p);
1167 }
1168
1169 static __poll_t
1170 sg_poll(struct file *filp, poll_table * wait)
1171 {
1172         __poll_t res = 0;
1173         Sg_device *sdp;
1174         Sg_fd *sfp;
1175         Sg_request *srp;
1176         int count = 0;
1177         unsigned long iflags;
1178
1179         sfp = filp->private_data;
1180         if (!sfp)
1181                 return EPOLLERR;
1182         sdp = sfp->parentdp;
1183         if (!sdp)
1184                 return EPOLLERR;
1185         poll_wait(filp, &sfp->read_wait, wait);
1186         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1187         list_for_each_entry(srp, &sfp->rq_list, entry) {
1188                 /* if any read waiting, flag it */
1189                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1190                         res = EPOLLIN | EPOLLRDNORM;
1191                 ++count;
1192         }
1193         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1194
1195         if (atomic_read(&sdp->detaching))
1196                 res |= EPOLLHUP;
1197         else if (!sfp->cmd_q) {
1198                 if (0 == count)
1199                         res |= EPOLLOUT | EPOLLWRNORM;
1200         } else if (count < SG_MAX_QUEUE)
1201                 res |= EPOLLOUT | EPOLLWRNORM;
1202         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1203                                       "sg_poll: res=0x%x\n", (__force u32) res));
1204         return res;
1205 }
1206
1207 static int
1208 sg_fasync(int fd, struct file *filp, int mode)
1209 {
1210         Sg_device *sdp;
1211         Sg_fd *sfp;
1212
1213         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1214                 return -ENXIO;
1215         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1216                                       "sg_fasync: mode=%d\n", mode));
1217
1218         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1219 }
1220
1221 static vm_fault_t
1222 sg_vma_fault(struct vm_fault *vmf)
1223 {
1224         struct vm_area_struct *vma = vmf->vma;
1225         Sg_fd *sfp;
1226         unsigned long offset, len, sa;
1227         Sg_scatter_hold *rsv_schp;
1228         int k, length;
1229
1230         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1231                 return VM_FAULT_SIGBUS;
1232         rsv_schp = &sfp->reserve;
1233         offset = vmf->pgoff << PAGE_SHIFT;
1234         if (offset >= rsv_schp->bufflen)
1235                 return VM_FAULT_SIGBUS;
1236         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1237                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1238                                       offset, rsv_schp->k_use_sg));
1239         sa = vma->vm_start;
1240         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1241         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1242                 len = vma->vm_end - sa;
1243                 len = (len < length) ? len : length;
1244                 if (offset < len) {
1245                         struct page *page = nth_page(rsv_schp->pages[k],
1246                                                      offset >> PAGE_SHIFT);
1247                         get_page(page); /* increment page count */
1248                         vmf->page = page;
1249                         return 0; /* success */
1250                 }
1251                 sa += len;
1252                 offset -= len;
1253         }
1254
1255         return VM_FAULT_SIGBUS;
1256 }
1257
1258 static const struct vm_operations_struct sg_mmap_vm_ops = {
1259         .fault = sg_vma_fault,
1260 };
1261
1262 static int
1263 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1264 {
1265         Sg_fd *sfp;
1266         unsigned long req_sz, len, sa;
1267         Sg_scatter_hold *rsv_schp;
1268         int k, length;
1269         int ret = 0;
1270
1271         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1272                 return -ENXIO;
1273         req_sz = vma->vm_end - vma->vm_start;
1274         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1275                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1276                                       (void *) vma->vm_start, (int) req_sz));
1277         if (vma->vm_pgoff)
1278                 return -EINVAL; /* want no offset */
1279         rsv_schp = &sfp->reserve;
1280         mutex_lock(&sfp->f_mutex);
1281         if (req_sz > rsv_schp->bufflen) {
1282                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1283                 goto out;
1284         }
1285
1286         sa = vma->vm_start;
1287         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1288         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1289                 len = vma->vm_end - sa;
1290                 len = (len < length) ? len : length;
1291                 sa += len;
1292         }
1293
1294         sfp->mmap_called = 1;
1295         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1296         vma->vm_private_data = sfp;
1297         vma->vm_ops = &sg_mmap_vm_ops;
1298 out:
1299         mutex_unlock(&sfp->f_mutex);
1300         return ret;
1301 }
1302
1303 static void
1304 sg_rq_end_io_usercontext(struct work_struct *work)
1305 {
1306         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1307         struct sg_fd *sfp = srp->parentfp;
1308
1309         sg_finish_rem_req(srp);
1310         sg_remove_request(sfp, srp);
1311         kref_put(&sfp->f_ref, sg_remove_sfp);
1312 }
1313
1314 /*
1315  * This function is a "bottom half" handler that is called by the mid
1316  * level when a command is completed (or has failed).
1317  */
1318 static void
1319 sg_rq_end_io(struct request *rq, blk_status_t status)
1320 {
1321         struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(rq);
1322         struct sg_request *srp = rq->end_io_data;
1323         Sg_device *sdp;
1324         Sg_fd *sfp;
1325         unsigned long iflags;
1326         unsigned int ms;
1327         char *sense;
1328         int result, resid, done = 1;
1329
1330         if (WARN_ON(srp->done != 0))
1331                 return;
1332
1333         sfp = srp->parentfp;
1334         if (WARN_ON(sfp == NULL))
1335                 return;
1336
1337         sdp = sfp->parentdp;
1338         if (unlikely(atomic_read(&sdp->detaching)))
1339                 pr_info("%s: device detaching\n", __func__);
1340
1341         sense = scmd->sense_buffer;
1342         result = scmd->result;
1343         resid = scmd->resid_len;
1344
1345         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1346                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1347                                       srp->header.pack_id, result));
1348         srp->header.resid = resid;
1349         ms = jiffies_to_msecs(jiffies);
1350         srp->header.duration = (ms > srp->header.duration) ?
1351                                 (ms - srp->header.duration) : 0;
1352         if (0 != result) {
1353                 struct scsi_sense_hdr sshdr;
1354
1355                 srp->header.status = 0xff & result;
1356                 srp->header.masked_status = status_byte(result);
1357                 srp->header.msg_status = COMMAND_COMPLETE;
1358                 srp->header.host_status = host_byte(result);
1359                 srp->header.driver_status = driver_byte(result);
1360                 if ((sdp->sgdebug > 0) &&
1361                     ((CHECK_CONDITION == srp->header.masked_status) ||
1362                      (COMMAND_TERMINATED == srp->header.masked_status)))
1363                         __scsi_print_sense(sdp->device, __func__, sense,
1364                                            SCSI_SENSE_BUFFERSIZE);
1365
1366                 /* Following if statement is a patch supplied by Eric Youngdale */
1367                 if (driver_byte(result) != 0
1368                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1369                     && !scsi_sense_is_deferred(&sshdr)
1370                     && sshdr.sense_key == UNIT_ATTENTION
1371                     && sdp->device->removable) {
1372                         /* Detected possible disc change. Set the bit - this */
1373                         /* may be used if there are filesystems using this device */
1374                         sdp->device->changed = 1;
1375                 }
1376         }
1377
1378         if (scmd->sense_len)
1379                 memcpy(srp->sense_b, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
1380
1381         /* Rely on write phase to clean out srp status values, so no "else" */
1382
1383         /*
1384          * Free the request as soon as it is complete so that its resources
1385          * can be reused without waiting for userspace to read() the
1386          * result.  But keep the associated bio (if any) around until
1387          * blk_rq_unmap_user() can be called from user context.
1388          */
1389         srp->rq = NULL;
1390         blk_mq_free_request(rq);
1391
1392         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1393         if (unlikely(srp->orphan)) {
1394                 if (sfp->keep_orphan)
1395                         srp->sg_io_owned = 0;
1396                 else
1397                         done = 0;
1398         }
1399         srp->done = done;
1400         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1401
1402         if (likely(done)) {
1403                 /* Now wake up any sg_read() that is waiting for this
1404                  * packet.
1405                  */
1406                 wake_up_interruptible(&sfp->read_wait);
1407                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1408                 kref_put(&sfp->f_ref, sg_remove_sfp);
1409         } else {
1410                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1411                 schedule_work(&srp->ew.work);
1412         }
1413 }
1414
1415 static const struct file_operations sg_fops = {
1416         .owner = THIS_MODULE,
1417         .read = sg_read,
1418         .write = sg_write,
1419         .poll = sg_poll,
1420         .unlocked_ioctl = sg_ioctl,
1421         .compat_ioctl = compat_ptr_ioctl,
1422         .open = sg_open,
1423         .mmap = sg_mmap,
1424         .release = sg_release,
1425         .fasync = sg_fasync,
1426         .llseek = no_llseek,
1427 };
1428
1429 static struct class *sg_sysfs_class;
1430
1431 static int sg_sysfs_valid = 0;
1432
1433 static Sg_device *
1434 sg_alloc(struct scsi_device *scsidp)
1435 {
1436         struct request_queue *q = scsidp->request_queue;
1437         Sg_device *sdp;
1438         unsigned long iflags;
1439         int error;
1440         u32 k;
1441
1442         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1443         if (!sdp) {
1444                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1445                             "failure\n", __func__);
1446                 return ERR_PTR(-ENOMEM);
1447         }
1448
1449         idr_preload(GFP_KERNEL);
1450         write_lock_irqsave(&sg_index_lock, iflags);
1451
1452         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1453         if (error < 0) {
1454                 if (error == -ENOSPC) {
1455                         sdev_printk(KERN_WARNING, scsidp,
1456                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1457                                     scsidp->type, SG_MAX_DEVS - 1);
1458                         error = -ENODEV;
1459                 } else {
1460                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1461                                     "allocation Sg_device failure: %d\n",
1462                                     __func__, error);
1463                 }
1464                 goto out_unlock;
1465         }
1466         k = error;
1467
1468         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1469                                         "sg_alloc: dev=%d \n", k));
1470         sprintf(sdp->name, "sg%d", k);
1471         sdp->device = scsidp;
1472         mutex_init(&sdp->open_rel_lock);
1473         INIT_LIST_HEAD(&sdp->sfds);
1474         init_waitqueue_head(&sdp->open_wait);
1475         atomic_set(&sdp->detaching, 0);
1476         rwlock_init(&sdp->sfd_lock);
1477         sdp->sg_tablesize = queue_max_segments(q);
1478         sdp->index = k;
1479         kref_init(&sdp->d_ref);
1480         error = 0;
1481
1482 out_unlock:
1483         write_unlock_irqrestore(&sg_index_lock, iflags);
1484         idr_preload_end();
1485
1486         if (error) {
1487                 kfree(sdp);
1488                 return ERR_PTR(error);
1489         }
1490         return sdp;
1491 }
1492
1493 static int
1494 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1495 {
1496         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1497         Sg_device *sdp = NULL;
1498         struct cdev * cdev = NULL;
1499         int error;
1500         unsigned long iflags;
1501
1502         error = -ENOMEM;
1503         cdev = cdev_alloc();
1504         if (!cdev) {
1505                 pr_warn("%s: cdev_alloc failed\n", __func__);
1506                 goto out;
1507         }
1508         cdev->owner = THIS_MODULE;
1509         cdev->ops = &sg_fops;
1510
1511         sdp = sg_alloc(scsidp);
1512         if (IS_ERR(sdp)) {
1513                 pr_warn("%s: sg_alloc failed\n", __func__);
1514                 error = PTR_ERR(sdp);
1515                 goto out;
1516         }
1517
1518         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1519         if (error)
1520                 goto cdev_add_err;
1521
1522         sdp->cdev = cdev;
1523         if (sg_sysfs_valid) {
1524                 struct device *sg_class_member;
1525
1526                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1527                                                 MKDEV(SCSI_GENERIC_MAJOR,
1528                                                       sdp->index),
1529                                                 sdp, "%s", sdp->name);
1530                 if (IS_ERR(sg_class_member)) {
1531                         pr_err("%s: device_create failed\n", __func__);
1532                         error = PTR_ERR(sg_class_member);
1533                         goto cdev_add_err;
1534                 }
1535                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1536                                           &sg_class_member->kobj, "generic");
1537                 if (error)
1538                         pr_err("%s: unable to make symlink 'generic' back "
1539                                "to sg%d\n", __func__, sdp->index);
1540         } else
1541                 pr_warn("%s: sg_sys Invalid\n", __func__);
1542
1543         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1544                     "type %d\n", sdp->index, scsidp->type);
1545
1546         dev_set_drvdata(cl_dev, sdp);
1547
1548         return 0;
1549
1550 cdev_add_err:
1551         write_lock_irqsave(&sg_index_lock, iflags);
1552         idr_remove(&sg_index_idr, sdp->index);
1553         write_unlock_irqrestore(&sg_index_lock, iflags);
1554         kfree(sdp);
1555
1556 out:
1557         if (cdev)
1558                 cdev_del(cdev);
1559         return error;
1560 }
1561
1562 static void
1563 sg_device_destroy(struct kref *kref)
1564 {
1565         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1566         unsigned long flags;
1567
1568         /* CAUTION!  Note that the device can still be found via idr_find()
1569          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1570          * any other cleanup.
1571          */
1572
1573         write_lock_irqsave(&sg_index_lock, flags);
1574         idr_remove(&sg_index_idr, sdp->index);
1575         write_unlock_irqrestore(&sg_index_lock, flags);
1576
1577         SCSI_LOG_TIMEOUT(3,
1578                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1579
1580         kfree(sdp);
1581 }
1582
1583 static void
1584 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1585 {
1586         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1587         Sg_device *sdp = dev_get_drvdata(cl_dev);
1588         unsigned long iflags;
1589         Sg_fd *sfp;
1590         int val;
1591
1592         if (!sdp)
1593                 return;
1594         /* want sdp->detaching non-zero as soon as possible */
1595         val = atomic_inc_return(&sdp->detaching);
1596         if (val > 1)
1597                 return; /* only want to do following once per device */
1598
1599         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1600                                       "%s\n", __func__));
1601
1602         read_lock_irqsave(&sdp->sfd_lock, iflags);
1603         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1604                 wake_up_interruptible_all(&sfp->read_wait);
1605                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1606         }
1607         wake_up_interruptible_all(&sdp->open_wait);
1608         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1609
1610         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1611         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1612         cdev_del(sdp->cdev);
1613         sdp->cdev = NULL;
1614
1615         kref_put(&sdp->d_ref, sg_device_destroy);
1616 }
1617
1618 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1619 module_param_named(def_reserved_size, def_reserved_size, int,
1620                    S_IRUGO | S_IWUSR);
1621 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1622
1623 MODULE_AUTHOR("Douglas Gilbert");
1624 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1625 MODULE_LICENSE("GPL");
1626 MODULE_VERSION(SG_VERSION_STR);
1627 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1628
1629 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1630                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1631 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1632 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1633
1634 #ifdef CONFIG_SYSCTL
1635 #include <linux/sysctl.h>
1636
1637 static struct ctl_table sg_sysctls[] = {
1638         {
1639                 .procname       = "sg-big-buff",
1640                 .data           = &sg_big_buff,
1641                 .maxlen         = sizeof(int),
1642                 .mode           = 0444,
1643                 .proc_handler   = proc_dointvec,
1644         },
1645         {}
1646 };
1647
1648 static struct ctl_table_header *hdr;
1649 static void register_sg_sysctls(void)
1650 {
1651         if (!hdr)
1652                 hdr = register_sysctl("kernel", sg_sysctls);
1653 }
1654
1655 static void unregister_sg_sysctls(void)
1656 {
1657         if (hdr)
1658                 unregister_sysctl_table(hdr);
1659 }
1660 #else
1661 #define register_sg_sysctls() do { } while (0)
1662 #define unregister_sg_sysctls() do { } while (0)
1663 #endif /* CONFIG_SYSCTL */
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         register_sg_sysctls();
1698 err_out:
1699         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1700         return rc;
1701 }
1702
1703 static void __exit
1704 exit_sg(void)
1705 {
1706         unregister_sg_sysctls();
1707 #ifdef CONFIG_SCSI_PROC_FS
1708         remove_proc_subtree("scsi/sg", NULL);
1709 #endif                          /* CONFIG_SCSI_PROC_FS */
1710         scsi_unregister_interface(&sg_interface);
1711         class_destroy(sg_sysfs_class);
1712         sg_sysfs_valid = 0;
1713         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1714                                  SG_MAX_DEVS);
1715         idr_destroy(&sg_index_idr);
1716 }
1717
1718 static int
1719 sg_start_req(Sg_request *srp, unsigned char *cmd)
1720 {
1721         int res;
1722         struct request *rq;
1723         Sg_fd *sfp = srp->parentfp;
1724         sg_io_hdr_t *hp = &srp->header;
1725         int dxfer_len = (int) hp->dxfer_len;
1726         int dxfer_dir = hp->dxfer_direction;
1727         unsigned int iov_count = hp->iovec_count;
1728         Sg_scatter_hold *req_schp = &srp->data;
1729         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1730         struct request_queue *q = sfp->parentdp->device->request_queue;
1731         struct rq_map_data *md, map_data;
1732         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1733         struct scsi_cmnd *scmd;
1734
1735         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1736                                       "sg_start_req: dxfer_len=%d\n",
1737                                       dxfer_len));
1738
1739         /*
1740          * NOTE
1741          *
1742          * With scsi-mq enabled, there are a fixed number of preallocated
1743          * requests equal in number to shost->can_queue.  If all of the
1744          * preallocated requests are already in use, then scsi_alloc_request()
1745          * will sleep until an active command completes, freeing up a request.
1746          * Although waiting in an asynchronous interface is less than ideal, we
1747          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1748          * not expect an EWOULDBLOCK from this condition.
1749          */
1750         rq = scsi_alloc_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1751                         REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1752         if (IS_ERR(rq))
1753                 return PTR_ERR(rq);
1754         scmd = blk_mq_rq_to_pdu(rq);
1755
1756         if (hp->cmd_len > sizeof(scmd->cmnd)) {
1757                 blk_mq_free_request(rq);
1758                 return -EINVAL;
1759         }
1760
1761         memcpy(scmd->cmnd, cmd, hp->cmd_len);
1762         scmd->cmd_len = hp->cmd_len;
1763
1764         srp->rq = rq;
1765         rq->end_io_data = srp;
1766         scmd->allowed = SG_DEFAULT_RETRIES;
1767
1768         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1769                 return 0;
1770
1771         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1772             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1773             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1774                 md = NULL;
1775         else
1776                 md = &map_data;
1777
1778         if (md) {
1779                 mutex_lock(&sfp->f_mutex);
1780                 if (dxfer_len <= rsv_schp->bufflen &&
1781                     !sfp->res_in_use) {
1782                         sfp->res_in_use = 1;
1783                         sg_link_reserve(sfp, srp, dxfer_len);
1784                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1785                         res = -EBUSY; /* sfp->res_in_use == 1 */
1786                         if (dxfer_len > rsv_schp->bufflen)
1787                                 res = -ENOMEM;
1788                         mutex_unlock(&sfp->f_mutex);
1789                         return res;
1790                 } else {
1791                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1792                         if (res) {
1793                                 mutex_unlock(&sfp->f_mutex);
1794                                 return res;
1795                         }
1796                 }
1797                 mutex_unlock(&sfp->f_mutex);
1798
1799                 md->pages = req_schp->pages;
1800                 md->page_order = req_schp->page_order;
1801                 md->nr_entries = req_schp->k_use_sg;
1802                 md->offset = 0;
1803                 md->null_mapped = hp->dxferp ? 0 : 1;
1804                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1805                         md->from_user = 1;
1806                 else
1807                         md->from_user = 0;
1808         }
1809
1810         if (iov_count) {
1811                 struct iovec *iov = NULL;
1812                 struct iov_iter i;
1813
1814                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1815                 if (res < 0)
1816                         return res;
1817
1818                 iov_iter_truncate(&i, hp->dxfer_len);
1819                 if (!iov_iter_count(&i)) {
1820                         kfree(iov);
1821                         return -EINVAL;
1822                 }
1823
1824                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1825                 kfree(iov);
1826         } else
1827                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1828                                       hp->dxfer_len, GFP_ATOMIC);
1829
1830         if (!res) {
1831                 srp->bio = rq->bio;
1832
1833                 if (!md) {
1834                         req_schp->dio_in_use = 1;
1835                         hp->info |= SG_INFO_DIRECT_IO;
1836                 }
1837         }
1838         return res;
1839 }
1840
1841 static int
1842 sg_finish_rem_req(Sg_request *srp)
1843 {
1844         int ret = 0;
1845
1846         Sg_fd *sfp = srp->parentfp;
1847         Sg_scatter_hold *req_schp = &srp->data;
1848
1849         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1850                                       "sg_finish_rem_req: res_used=%d\n",
1851                                       (int) srp->res_used));
1852         if (srp->bio)
1853                 ret = blk_rq_unmap_user(srp->bio);
1854
1855         if (srp->rq)
1856                 blk_mq_free_request(srp->rq);
1857
1858         if (srp->res_used)
1859                 sg_unlink_reserve(sfp, srp);
1860         else
1861                 sg_remove_scat(sfp, req_schp);
1862
1863         return ret;
1864 }
1865
1866 static int
1867 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1868 {
1869         int sg_bufflen = tablesize * sizeof(struct page *);
1870         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1871
1872         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1873         if (!schp->pages)
1874                 return -ENOMEM;
1875         schp->sglist_len = sg_bufflen;
1876         return tablesize;       /* number of scat_gath elements allocated */
1877 }
1878
1879 static int
1880 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1881 {
1882         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1883         int sg_tablesize = sfp->parentdp->sg_tablesize;
1884         int blk_size = buff_size, order;
1885         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1886
1887         if (blk_size < 0)
1888                 return -EFAULT;
1889         if (0 == blk_size)
1890                 ++blk_size;     /* don't know why */
1891         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1892         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1893         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1894                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1895                 buff_size, blk_size));
1896
1897         /* N.B. ret_sz carried into this block ... */
1898         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1899         if (mx_sc_elems < 0)
1900                 return mx_sc_elems;     /* most likely -ENOMEM */
1901
1902         num = scatter_elem_sz;
1903         if (unlikely(num != scatter_elem_sz_prev)) {
1904                 if (num < PAGE_SIZE) {
1905                         scatter_elem_sz = PAGE_SIZE;
1906                         scatter_elem_sz_prev = PAGE_SIZE;
1907                 } else
1908                         scatter_elem_sz_prev = num;
1909         }
1910
1911         order = get_order(num);
1912 retry:
1913         ret_sz = 1 << (PAGE_SHIFT + order);
1914
1915         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1916              k++, rem_sz -= ret_sz) {
1917
1918                 num = (rem_sz > scatter_elem_sz_prev) ?
1919                         scatter_elem_sz_prev : rem_sz;
1920
1921                 schp->pages[k] = alloc_pages(gfp_mask, order);
1922                 if (!schp->pages[k])
1923                         goto out;
1924
1925                 if (num == scatter_elem_sz_prev) {
1926                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1927                                 scatter_elem_sz = ret_sz;
1928                                 scatter_elem_sz_prev = ret_sz;
1929                         }
1930                 }
1931
1932                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1933                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1934                                  k, num, ret_sz));
1935         }               /* end of for loop */
1936
1937         schp->page_order = order;
1938         schp->k_use_sg = k;
1939         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1940                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1941                          k, rem_sz));
1942
1943         schp->bufflen = blk_size;
1944         if (rem_sz > 0) /* must have failed */
1945                 return -ENOMEM;
1946         return 0;
1947 out:
1948         for (i = 0; i < k; i++)
1949                 __free_pages(schp->pages[i], order);
1950
1951         if (--order >= 0)
1952                 goto retry;
1953
1954         return -ENOMEM;
1955 }
1956
1957 static void
1958 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1959 {
1960         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1961                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1962         if (schp->pages && schp->sglist_len > 0) {
1963                 if (!schp->dio_in_use) {
1964                         int k;
1965
1966                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1967                                 SCSI_LOG_TIMEOUT(5,
1968                                         sg_printk(KERN_INFO, sfp->parentdp,
1969                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1970                                         k, schp->pages[k]));
1971                                 __free_pages(schp->pages[k], schp->page_order);
1972                         }
1973
1974                         kfree(schp->pages);
1975                 }
1976         }
1977         memset(schp, 0, sizeof (*schp));
1978 }
1979
1980 static int
1981 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1982 {
1983         Sg_scatter_hold *schp = &srp->data;
1984         int k, num;
1985
1986         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1987                          "sg_read_oxfer: num_read_xfer=%d\n",
1988                          num_read_xfer));
1989         if ((!outp) || (num_read_xfer <= 0))
1990                 return 0;
1991
1992         num = 1 << (PAGE_SHIFT + schp->page_order);
1993         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1994                 if (num > num_read_xfer) {
1995                         if (copy_to_user(outp, page_address(schp->pages[k]),
1996                                            num_read_xfer))
1997                                 return -EFAULT;
1998                         break;
1999                 } else {
2000                         if (copy_to_user(outp, page_address(schp->pages[k]),
2001                                            num))
2002                                 return -EFAULT;
2003                         num_read_xfer -= num;
2004                         if (num_read_xfer <= 0)
2005                                 break;
2006                         outp += num;
2007                 }
2008         }
2009
2010         return 0;
2011 }
2012
2013 static void
2014 sg_build_reserve(Sg_fd * sfp, int req_size)
2015 {
2016         Sg_scatter_hold *schp = &sfp->reserve;
2017
2018         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2019                          "sg_build_reserve: req_size=%d\n", req_size));
2020         do {
2021                 if (req_size < PAGE_SIZE)
2022                         req_size = PAGE_SIZE;
2023                 if (0 == sg_build_indirect(schp, sfp, req_size))
2024                         return;
2025                 else
2026                         sg_remove_scat(sfp, schp);
2027                 req_size >>= 1; /* divide by 2 */
2028         } while (req_size > (PAGE_SIZE / 2));
2029 }
2030
2031 static void
2032 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2033 {
2034         Sg_scatter_hold *req_schp = &srp->data;
2035         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2036         int k, num, rem;
2037
2038         srp->res_used = 1;
2039         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2040                          "sg_link_reserve: size=%d\n", size));
2041         rem = size;
2042
2043         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2044         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2045                 if (rem <= num) {
2046                         req_schp->k_use_sg = k + 1;
2047                         req_schp->sglist_len = rsv_schp->sglist_len;
2048                         req_schp->pages = rsv_schp->pages;
2049
2050                         req_schp->bufflen = size;
2051                         req_schp->page_order = rsv_schp->page_order;
2052                         break;
2053                 } else
2054                         rem -= num;
2055         }
2056
2057         if (k >= rsv_schp->k_use_sg)
2058                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2059                                  "sg_link_reserve: BAD size\n"));
2060 }
2061
2062 static void
2063 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2064 {
2065         Sg_scatter_hold *req_schp = &srp->data;
2066
2067         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2068                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2069                                       (int) req_schp->k_use_sg));
2070         req_schp->k_use_sg = 0;
2071         req_schp->bufflen = 0;
2072         req_schp->pages = NULL;
2073         req_schp->page_order = 0;
2074         req_schp->sglist_len = 0;
2075         srp->res_used = 0;
2076         /* Called without mutex lock to avoid deadlock */
2077         sfp->res_in_use = 0;
2078 }
2079
2080 static Sg_request *
2081 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2082 {
2083         Sg_request *resp;
2084         unsigned long iflags;
2085
2086         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2087         list_for_each_entry(resp, &sfp->rq_list, entry) {
2088                 /* look for requests that are ready + not SG_IO owned */
2089                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2090                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2091                         resp->done = 2; /* guard against other readers */
2092                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2093                         return resp;
2094                 }
2095         }
2096         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2097         return NULL;
2098 }
2099
2100 /* always adds to end of list */
2101 static Sg_request *
2102 sg_add_request(Sg_fd * sfp)
2103 {
2104         int k;
2105         unsigned long iflags;
2106         Sg_request *rp = sfp->req_arr;
2107
2108         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2109         if (!list_empty(&sfp->rq_list)) {
2110                 if (!sfp->cmd_q)
2111                         goto out_unlock;
2112
2113                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2114                         if (!rp->parentfp)
2115                                 break;
2116                 }
2117                 if (k >= SG_MAX_QUEUE)
2118                         goto out_unlock;
2119         }
2120         memset(rp, 0, sizeof (Sg_request));
2121         rp->parentfp = sfp;
2122         rp->header.duration = jiffies_to_msecs(jiffies);
2123         list_add_tail(&rp->entry, &sfp->rq_list);
2124         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2125         return rp;
2126 out_unlock:
2127         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2128         return NULL;
2129 }
2130
2131 /* Return of 1 for found; 0 for not found */
2132 static int
2133 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2134 {
2135         unsigned long iflags;
2136         int res = 0;
2137
2138         if (!sfp || !srp || list_empty(&sfp->rq_list))
2139                 return res;
2140         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2141         if (!list_empty(&srp->entry)) {
2142                 list_del(&srp->entry);
2143                 srp->parentfp = NULL;
2144                 res = 1;
2145         }
2146         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2147         return res;
2148 }
2149
2150 static Sg_fd *
2151 sg_add_sfp(Sg_device * sdp)
2152 {
2153         Sg_fd *sfp;
2154         unsigned long iflags;
2155         int bufflen;
2156
2157         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2158         if (!sfp)
2159                 return ERR_PTR(-ENOMEM);
2160
2161         init_waitqueue_head(&sfp->read_wait);
2162         rwlock_init(&sfp->rq_list_lock);
2163         INIT_LIST_HEAD(&sfp->rq_list);
2164         kref_init(&sfp->f_ref);
2165         mutex_init(&sfp->f_mutex);
2166         sfp->timeout = SG_DEFAULT_TIMEOUT;
2167         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2168         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2169         sfp->cmd_q = SG_DEF_COMMAND_Q;
2170         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2171         sfp->parentdp = sdp;
2172         write_lock_irqsave(&sdp->sfd_lock, iflags);
2173         if (atomic_read(&sdp->detaching)) {
2174                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2175                 kfree(sfp);
2176                 return ERR_PTR(-ENODEV);
2177         }
2178         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2179         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2180         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2181                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2182         if (unlikely(sg_big_buff != def_reserved_size))
2183                 sg_big_buff = def_reserved_size;
2184
2185         bufflen = min_t(int, sg_big_buff,
2186                         max_sectors_bytes(sdp->device->request_queue));
2187         sg_build_reserve(sfp, bufflen);
2188         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2189                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2190                                       sfp->reserve.bufflen,
2191                                       sfp->reserve.k_use_sg));
2192
2193         kref_get(&sdp->d_ref);
2194         __module_get(THIS_MODULE);
2195         return sfp;
2196 }
2197
2198 static void
2199 sg_remove_sfp_usercontext(struct work_struct *work)
2200 {
2201         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2202         struct sg_device *sdp = sfp->parentdp;
2203         Sg_request *srp;
2204         unsigned long iflags;
2205
2206         /* Cleanup any responses which were never read(). */
2207         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2208         while (!list_empty(&sfp->rq_list)) {
2209                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2210                 sg_finish_rem_req(srp);
2211                 list_del(&srp->entry);
2212                 srp->parentfp = NULL;
2213         }
2214         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2215
2216         if (sfp->reserve.bufflen > 0) {
2217                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2218                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2219                                 (int) sfp->reserve.bufflen,
2220                                 (int) sfp->reserve.k_use_sg));
2221                 sg_remove_scat(sfp, &sfp->reserve);
2222         }
2223
2224         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2225                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2226         kfree(sfp);
2227
2228         scsi_device_put(sdp->device);
2229         kref_put(&sdp->d_ref, sg_device_destroy);
2230         module_put(THIS_MODULE);
2231 }
2232
2233 static void
2234 sg_remove_sfp(struct kref *kref)
2235 {
2236         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2237         struct sg_device *sdp = sfp->parentdp;
2238         unsigned long iflags;
2239
2240         write_lock_irqsave(&sdp->sfd_lock, iflags);
2241         list_del(&sfp->sfd_siblings);
2242         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2243
2244         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2245         schedule_work(&sfp->ew.work);
2246 }
2247
2248 #ifdef CONFIG_SCSI_PROC_FS
2249 static int
2250 sg_idr_max_id(int id, void *p, void *data)
2251 {
2252         int *k = data;
2253
2254         if (*k < id)
2255                 *k = id;
2256
2257         return 0;
2258 }
2259
2260 static int
2261 sg_last_dev(void)
2262 {
2263         int k = -1;
2264         unsigned long iflags;
2265
2266         read_lock_irqsave(&sg_index_lock, iflags);
2267         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2268         read_unlock_irqrestore(&sg_index_lock, iflags);
2269         return k + 1;           /* origin 1 */
2270 }
2271 #endif
2272
2273 /* must be called with sg_index_lock held */
2274 static Sg_device *sg_lookup_dev(int dev)
2275 {
2276         return idr_find(&sg_index_idr, dev);
2277 }
2278
2279 static Sg_device *
2280 sg_get_dev(int dev)
2281 {
2282         struct sg_device *sdp;
2283         unsigned long flags;
2284
2285         read_lock_irqsave(&sg_index_lock, flags);
2286         sdp = sg_lookup_dev(dev);
2287         if (!sdp)
2288                 sdp = ERR_PTR(-ENXIO);
2289         else if (atomic_read(&sdp->detaching)) {
2290                 /* If sdp->detaching, then the refcount may already be 0, in
2291                  * which case it would be a bug to do kref_get().
2292                  */
2293                 sdp = ERR_PTR(-ENODEV);
2294         } else
2295                 kref_get(&sdp->d_ref);
2296         read_unlock_irqrestore(&sg_index_lock, flags);
2297
2298         return sdp;
2299 }
2300
2301 #ifdef CONFIG_SCSI_PROC_FS
2302 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2303
2304 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2305 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2306                                   size_t count, loff_t *off);
2307 static const struct proc_ops adio_proc_ops = {
2308         .proc_open      = sg_proc_single_open_adio,
2309         .proc_read      = seq_read,
2310         .proc_lseek     = seq_lseek,
2311         .proc_write     = sg_proc_write_adio,
2312         .proc_release   = single_release,
2313 };
2314
2315 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2316 static ssize_t sg_proc_write_dressz(struct file *filp,
2317                 const char __user *buffer, size_t count, loff_t *off);
2318 static const struct proc_ops dressz_proc_ops = {
2319         .proc_open      = sg_proc_single_open_dressz,
2320         .proc_read      = seq_read,
2321         .proc_lseek     = seq_lseek,
2322         .proc_write     = sg_proc_write_dressz,
2323         .proc_release   = single_release,
2324 };
2325
2326 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2327 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2328 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2329 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2330 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2331 static void dev_seq_stop(struct seq_file *s, void *v);
2332 static const struct seq_operations dev_seq_ops = {
2333         .start = dev_seq_start,
2334         .next  = dev_seq_next,
2335         .stop  = dev_seq_stop,
2336         .show  = sg_proc_seq_show_dev,
2337 };
2338
2339 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2340 static const struct seq_operations devstrs_seq_ops = {
2341         .start = dev_seq_start,
2342         .next  = dev_seq_next,
2343         .stop  = dev_seq_stop,
2344         .show  = sg_proc_seq_show_devstrs,
2345 };
2346
2347 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2348 static const struct seq_operations debug_seq_ops = {
2349         .start = dev_seq_start,
2350         .next  = dev_seq_next,
2351         .stop  = dev_seq_stop,
2352         .show  = sg_proc_seq_show_debug,
2353 };
2354
2355 static int
2356 sg_proc_init(void)
2357 {
2358         struct proc_dir_entry *p;
2359
2360         p = proc_mkdir("scsi/sg", NULL);
2361         if (!p)
2362                 return 1;
2363
2364         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2365         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2366         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2367         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2368         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2369         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2370         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2371         return 0;
2372 }
2373
2374
2375 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2376 {
2377         seq_printf(s, "%d\n", *((int *)s->private));
2378         return 0;
2379 }
2380
2381 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2382 {
2383         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2384 }
2385
2386 static ssize_t
2387 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2388                    size_t count, loff_t *off)
2389 {
2390         int err;
2391         unsigned long num;
2392
2393         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2394                 return -EACCES;
2395         err = kstrtoul_from_user(buffer, count, 0, &num);
2396         if (err)
2397                 return err;
2398         sg_allow_dio = num ? 1 : 0;
2399         return count;
2400 }
2401
2402 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2403 {
2404         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2405 }
2406
2407 static ssize_t
2408 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2409                      size_t count, loff_t *off)
2410 {
2411         int err;
2412         unsigned long k = ULONG_MAX;
2413
2414         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2415                 return -EACCES;
2416
2417         err = kstrtoul_from_user(buffer, count, 0, &k);
2418         if (err)
2419                 return err;
2420         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2421                 sg_big_buff = k;
2422                 return count;
2423         }
2424         return -ERANGE;
2425 }
2426
2427 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2428 {
2429         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2430                    sg_version_date);
2431         return 0;
2432 }
2433
2434 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2435 {
2436         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2437         return 0;
2438 }
2439
2440 struct sg_proc_deviter {
2441         loff_t  index;
2442         size_t  max;
2443 };
2444
2445 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2446 {
2447         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2448
2449         s->private = it;
2450         if (! it)
2451                 return NULL;
2452
2453         it->index = *pos;
2454         it->max = sg_last_dev();
2455         if (it->index >= it->max)
2456                 return NULL;
2457         return it;
2458 }
2459
2460 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2461 {
2462         struct sg_proc_deviter * it = s->private;
2463
2464         *pos = ++it->index;
2465         return (it->index < it->max) ? it : NULL;
2466 }
2467
2468 static void dev_seq_stop(struct seq_file *s, void *v)
2469 {
2470         kfree(s->private);
2471 }
2472
2473 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2474 {
2475         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2476         Sg_device *sdp;
2477         struct scsi_device *scsidp;
2478         unsigned long iflags;
2479
2480         read_lock_irqsave(&sg_index_lock, iflags);
2481         sdp = it ? sg_lookup_dev(it->index) : NULL;
2482         if ((NULL == sdp) || (NULL == sdp->device) ||
2483             (atomic_read(&sdp->detaching)))
2484                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2485         else {
2486                 scsidp = sdp->device;
2487                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2488                               scsidp->host->host_no, scsidp->channel,
2489                               scsidp->id, scsidp->lun, (int) scsidp->type,
2490                               1,
2491                               (int) scsidp->queue_depth,
2492                               (int) scsi_device_busy(scsidp),
2493                               (int) scsi_device_online(scsidp));
2494         }
2495         read_unlock_irqrestore(&sg_index_lock, iflags);
2496         return 0;
2497 }
2498
2499 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2500 {
2501         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2502         Sg_device *sdp;
2503         struct scsi_device *scsidp;
2504         unsigned long iflags;
2505
2506         read_lock_irqsave(&sg_index_lock, iflags);
2507         sdp = it ? sg_lookup_dev(it->index) : NULL;
2508         scsidp = sdp ? sdp->device : NULL;
2509         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2510                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2511                            scsidp->vendor, scsidp->model, scsidp->rev);
2512         else
2513                 seq_puts(s, "<no active device>\n");
2514         read_unlock_irqrestore(&sg_index_lock, iflags);
2515         return 0;
2516 }
2517
2518 /* must be called while holding sg_index_lock */
2519 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2520 {
2521         int k, new_interface, blen, usg;
2522         Sg_request *srp;
2523         Sg_fd *fp;
2524         const sg_io_hdr_t *hp;
2525         const char * cp;
2526         unsigned int ms;
2527
2528         k = 0;
2529         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2530                 k++;
2531                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2532                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2533                            "(res)sgat=%d low_dma=%d\n", k,
2534                            jiffies_to_msecs(fp->timeout),
2535                            fp->reserve.bufflen,
2536                            (int) fp->reserve.k_use_sg, 0);
2537                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2538                            (int) fp->cmd_q, (int) fp->force_packid,
2539                            (int) fp->keep_orphan);
2540                 list_for_each_entry(srp, &fp->rq_list, entry) {
2541                         hp = &srp->header;
2542                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2543                         if (srp->res_used) {
2544                                 if (new_interface &&
2545                                     (SG_FLAG_MMAP_IO & hp->flags))
2546                                         cp = "     mmap>> ";
2547                                 else
2548                                         cp = "     rb>> ";
2549                         } else {
2550                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2551                                         cp = "     dio>> ";
2552                                 else
2553                                         cp = "     ";
2554                         }
2555                         seq_puts(s, cp);
2556                         blen = srp->data.bufflen;
2557                         usg = srp->data.k_use_sg;
2558                         seq_puts(s, srp->done ?
2559                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2560                                   : "act:");
2561                         seq_printf(s, " id=%d blen=%d",
2562                                    srp->header.pack_id, blen);
2563                         if (srp->done)
2564                                 seq_printf(s, " dur=%d", hp->duration);
2565                         else {
2566                                 ms = jiffies_to_msecs(jiffies);
2567                                 seq_printf(s, " t_o/elap=%d/%d",
2568                                         (new_interface ? hp->timeout :
2569                                                   jiffies_to_msecs(fp->timeout)),
2570                                         (ms > hp->duration ? ms - hp->duration : 0));
2571                         }
2572                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2573                                    (int) srp->data.cmd_opcode);
2574                 }
2575                 if (list_empty(&fp->rq_list))
2576                         seq_puts(s, "     No requests active\n");
2577                 read_unlock(&fp->rq_list_lock);
2578         }
2579 }
2580
2581 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2582 {
2583         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2584         Sg_device *sdp;
2585         unsigned long iflags;
2586
2587         if (it && (0 == it->index))
2588                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2589                            (int)it->max, sg_big_buff);
2590
2591         read_lock_irqsave(&sg_index_lock, iflags);
2592         sdp = it ? sg_lookup_dev(it->index) : NULL;
2593         if (NULL == sdp)
2594                 goto skip;
2595         read_lock(&sdp->sfd_lock);
2596         if (!list_empty(&sdp->sfds)) {
2597                 seq_printf(s, " >>> device=%s ", sdp->name);
2598                 if (atomic_read(&sdp->detaching))
2599                         seq_puts(s, "detaching pending close ");
2600                 else if (sdp->device) {
2601                         struct scsi_device *scsidp = sdp->device;
2602
2603                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2604                                    scsidp->host->host_no,
2605                                    scsidp->channel, scsidp->id,
2606                                    scsidp->lun,
2607                                    scsidp->host->hostt->emulated);
2608                 }
2609                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2610                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2611                 sg_proc_debug_helper(s, sdp);
2612         }
2613         read_unlock(&sdp->sfd_lock);
2614 skip:
2615         read_unlock_irqrestore(&sg_index_lock, iflags);
2616         return 0;
2617 }
2618
2619 #endif                          /* CONFIG_SCSI_PROC_FS */
2620
2621 module_init(init_sg);
2622 module_exit(exit_sg);