drivers/scsi/sg.c

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