drivers/scsi/sg.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  History:
   4  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   5  *           to allow user process control of SCSI devices.
   6  *  Development Sponsored by Killy Corp. NY NY
   7  *
   8  * Original driver (sg.c):
   9  *        Copyright (C) 1992 Lawrence Foard
  10  * Version 2 and 3 extensions to driver:
  11  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  12  */
  13
  14 static int sg_version_num = 30536;      /* 2 digits for each component */
  15 #define SG_VERSION_STR "3.5.36"
  16
  17 /*
  18  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  19  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  20  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  21  *        (otherwise the macros compile to empty statements).
  22  *
  23  */
  24 #include <linux/module.h>
  25
  26 #include <linux/fs.h>
  27 #include <linux/kernel.h>
  28 #include <linux/sched.h>
  29 #include <linux/string.h>
  30 #include <linux/mm.h>
  31 #include <linux/errno.h>
  32 #include <linux/mtio.h>
  33 #include <linux/ioctl.h>
  34 #include <linux/major.h>
  35 #include <linux/slab.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/init.h>
  38 #include <linux/poll.h>
  39 #include <linux/moduleparam.h>
  40 #include <linux/cdev.h>
  41 #include <linux/idr.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/delay.h>
  45 #include <linux/blktrace_api.h>
  46 #include <linux/mutex.h>
  47 #include <linux/atomic.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/uio.h>
  50 #include <linux/cred.h> /* for sg_check_file_access() */
  51
  52 #include "scsi.h"
  53 #include <scsi/scsi_dbg.h>
  54 #include <scsi/scsi_host.h>
  55 #include <scsi/scsi_driver.h>
  56 #include <scsi/scsi_ioctl.h>
  57 #include <scsi/sg.h>
  58
  59 #include "scsi_logging.h"
  60
  61 #ifdef CONFIG_SCSI_PROC_FS
  62 #include <linux/proc_fs.h>
  63 static char *sg_version_date = "20140603";
  64
  65 static int sg_proc_init(void);
  66 #endif
  67
  68 #define SG_ALLOW_DIO_DEF 0
  69
  70 #define SG_MAX_DEVS 32768
  71
  72 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  73  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  74  * than 16 bytes are "variable length" whose length is a multiple of 4
  75  */
  76 #define SG_MAX_CDB_SIZE 252
  77
  78 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  79
  80 int sg_big_buff = SG_DEF_RESERVED_SIZE;
  81 /* N.B. This variable is readable and writeable via
  82    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  83    of this size (or less if there is not enough memory) will be reserved
  84    for use by this file descriptor. [Deprecated usage: this variable is also
  85    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
  86    the kernel (i.e. it is not a module).] */
  87 static int def_reserved_size = -1;      /* picks up init parameter */
  88 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
  89
  90 static int scatter_elem_sz = SG_SCATTER_SZ;
  91 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
  92
  93 #define SG_SECTOR_SZ 512
  94
  95 static int sg_add_device(struct device *, struct class_interface *);
  96 static void sg_remove_device(struct device *, struct class_interface *);
  97
  98 static DEFINE_IDR(sg_index_idr);
  99 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
 100                                                            file descriptor list for device */
 101
 102 static struct class_interface sg_interface = {
 103         .add_dev        = sg_add_device,
 104         .remove_dev     = sg_remove_device,
 105 };
 106
 107 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 108         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 109         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 110         unsigned bufflen;       /* Size of (aggregate) data buffer */
 111         struct page **pages;
 112         int page_order;
 113         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 114         unsigned char cmd_opcode; /* first byte of command */
 115 } Sg_scatter_hold;
 116
 117 struct sg_device;               /* forward declarations */
 118 struct sg_fd;
 119
 120 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 121         struct list_head entry; /* list entry */
 122         struct sg_fd *parentfp; /* NULL -> not in use */
 123         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 124         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 125         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 126         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 127         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 128         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 129         /* done protected by rq_list_lock */
 130         char done;              /* 0->before bh, 1->before read, 2->read */
 131         struct request *rq;
 132         struct bio *bio;
 133         struct execute_work ew;
 134 } Sg_request;
 135
 136 typedef struct sg_fd {          /* holds the state of a file descriptor */
 137         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 138         struct sg_device *parentdp;     /* owning device */
 139         wait_queue_head_t read_wait;    /* queue read until command done */
 140         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 141         struct mutex f_mutex;   /* protect against changes in this fd */
 142         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 143         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 144         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 145         struct list_head rq_list; /* head of request list */
 146         struct fasync_struct *async_qp; /* used by asynchronous notification */
 147         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 148         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 149         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 150         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 151         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 152         char mmap_called;       /* 0 -> mmap() never called on this fd */
 153         char res_in_use;        /* 1 -> 'reserve' array in use */
 154         struct kref f_ref;
 155         struct execute_work ew;
 156 } Sg_fd;
 157
 158 typedef struct sg_device { /* holds the state of each scsi generic device */
 159         struct scsi_device *device;
 160         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 161         struct mutex open_rel_lock;     /* held when in open() or release() */
 162         int sg_tablesize;       /* adapter's max scatter-gather table size */
 163         u32 index;              /* device index number */
 164         struct list_head sfds;
 165         rwlock_t sfd_lock;      /* protect access to sfd list */
 166         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 167         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 168         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 169         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 170         char name[DISK_NAME_LEN];
 171         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 172         struct kref d_ref;
 173 } Sg_device;
 174
 175 /* tasklet or soft irq callback */
 176 static void sg_rq_end_io(struct request *rq, blk_status_t status);
 177 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 178 static int sg_finish_rem_req(Sg_request * srp);
 179 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 180 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 181                            Sg_request * srp);
 182 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 183                         const char __user *buf, size_t count, int blocking,
 184                         int read_only, int sg_io_owned, Sg_request **o_srp);
 185 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 186                            unsigned char *cmnd, int timeout, int blocking);
 187 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 188 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 189 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 190 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 191 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 192 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 193 static void sg_remove_sfp(struct kref *);
 194 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id, bool *busy);
 195 static Sg_request *sg_add_request(Sg_fd * sfp);
 196 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 197 static Sg_device *sg_get_dev(int dev);
 198 static void sg_device_destroy(struct kref *kref);
 199
 200 #define SZ_SG_HEADER sizeof(struct sg_header)
 201 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 202 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 203 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 204
 205 #define sg_printk(prefix, sdp, fmt, a...) \
 206         sdev_prefix_printk(prefix, (sdp)->device, (sdp)->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         if (!scsi_cmd_allowed(cmd, filp->f_mode))
 242                 return -EPERM;
 243         return 0;
 244 }
 245
 246 static int
 247 open_wait(Sg_device *sdp, int flags)
 248 {
 249         int retval = 0;
 250
 251         if (flags & O_EXCL) {
 252                 while (sdp->open_cnt > 0) {
 253                         mutex_unlock(&sdp->open_rel_lock);
 254                         retval = wait_event_interruptible(sdp->open_wait,
 255                                         (atomic_read(&sdp->detaching) ||
 256                                          !sdp->open_cnt));
 257                         mutex_lock(&sdp->open_rel_lock);
 258
 259                         if (retval) /* -ERESTARTSYS */
 260                                 return retval;
 261                         if (atomic_read(&sdp->detaching))
 262                                 return -ENODEV;
 263                 }
 264         } else {
 265                 while (sdp->exclude) {
 266                         mutex_unlock(&sdp->open_rel_lock);
 267                         retval = wait_event_interruptible(sdp->open_wait,
 268                                         (atomic_read(&sdp->detaching) ||
 269                                          !sdp->exclude));
 270                         mutex_lock(&sdp->open_rel_lock);
 271
 272                         if (retval) /* -ERESTARTSYS */
 273                                 return retval;
 274                         if (atomic_read(&sdp->detaching))
 275                                 return -ENODEV;
 276                 }
 277         }
 278
 279         return retval;
 280 }
 281
 282 /* Returns 0 on success, else a negated errno value */
 283 static int
 284 sg_open(struct inode *inode, struct file *filp)
 285 {
 286         int dev = iminor(inode);
 287         int flags = filp->f_flags;
 288         struct request_queue *q;
 289         Sg_device *sdp;
 290         Sg_fd *sfp;
 291         int retval;
 292
 293         nonseekable_open(inode, filp);
 294         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 295                 return -EPERM; /* Can't lock it with read only access */
 296         sdp = sg_get_dev(dev);
 297         if (IS_ERR(sdp))
 298                 return PTR_ERR(sdp);
 299
 300         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 301                                       "sg_open: flags=0x%x\n", flags));
 302
 303         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 304         /* Prevent the device driver from vanishing while we sleep */
 305         retval = scsi_device_get(sdp->device);
 306         if (retval)
 307                 goto sg_put;
 308
 309         retval = scsi_autopm_get_device(sdp->device);
 310         if (retval)
 311                 goto sdp_put;
 312
 313         /* scsi_block_when_processing_errors() may block so bypass
 314          * check if O_NONBLOCK. Permits SCSI commands to be issued
 315          * during error recovery. Tread carefully. */
 316         if (!((flags & O_NONBLOCK) ||
 317               scsi_block_when_processing_errors(sdp->device))) {
 318                 retval = -ENXIO;
 319                 /* we are in error recovery for this device */
 320                 goto error_out;
 321         }
 322
 323         mutex_lock(&sdp->open_rel_lock);
 324         if (flags & O_NONBLOCK) {
 325                 if (flags & O_EXCL) {
 326                         if (sdp->open_cnt > 0) {
 327                                 retval = -EBUSY;
 328                                 goto error_mutex_locked;
 329                         }
 330                 } else {
 331                         if (sdp->exclude) {
 332                                 retval = -EBUSY;
 333                                 goto error_mutex_locked;
 334                         }
 335                 }
 336         } else {
 337                 retval = open_wait(sdp, flags);
 338                 if (retval) /* -ERESTARTSYS or -ENODEV */
 339                         goto error_mutex_locked;
 340         }
 341
 342         /* N.B. at this point we are holding the open_rel_lock */
 343         if (flags & O_EXCL)
 344                 sdp->exclude = true;
 345
 346         if (sdp->open_cnt < 1) {  /* no existing opens */
 347                 sdp->sgdebug = 0;
 348                 q = sdp->device->request_queue;
 349                 sdp->sg_tablesize = queue_max_segments(q);
 350         }
 351         sfp = sg_add_sfp(sdp);
 352         if (IS_ERR(sfp)) {
 353                 retval = PTR_ERR(sfp);
 354                 goto out_undo;
 355         }
 356
 357         filp->private_data = sfp;
 358         sdp->open_cnt++;
 359         mutex_unlock(&sdp->open_rel_lock);
 360
 361         retval = 0;
 362 sg_put:
 363         kref_put(&sdp->d_ref, sg_device_destroy);
 364         return retval;
 365
 366 out_undo:
 367         if (flags & O_EXCL) {
 368                 sdp->exclude = false;   /* undo if error */
 369                 wake_up_interruptible(&sdp->open_wait);
 370         }
 371 error_mutex_locked:
 372         mutex_unlock(&sdp->open_rel_lock);
 373 error_out:
 374         scsi_autopm_put_device(sdp->device);
 375 sdp_put:
 376         scsi_device_put(sdp->device);
 377         goto sg_put;
 378 }
 379
 380 /* Release resources associated with a successful sg_open()
 381  * Returns 0 on success, else a negated errno value */
 382 static int
 383 sg_release(struct inode *inode, struct file *filp)
 384 {
 385         Sg_device *sdp;
 386         Sg_fd *sfp;
 387
 388         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 389                 return -ENXIO;
 390         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 391
 392         mutex_lock(&sdp->open_rel_lock);
 393         scsi_autopm_put_device(sdp->device);
 394         kref_put(&sfp->f_ref, sg_remove_sfp);
 395         sdp->open_cnt--;
 396
 397         /* possibly many open()s waiting on exlude clearing, start many;
 398          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 399         if (sdp->exclude) {
 400                 sdp->exclude = false;
 401                 wake_up_interruptible_all(&sdp->open_wait);
 402         } else if (0 == sdp->open_cnt) {
 403                 wake_up_interruptible(&sdp->open_wait);
 404         }
 405         mutex_unlock(&sdp->open_rel_lock);
 406         return 0;
 407 }
 408
 409 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 410 {
 411         struct sg_header __user *old_hdr = buf;
 412         int reply_len;
 413
 414         if (count >= SZ_SG_HEADER) {
 415                 /* negative reply_len means v3 format, otherwise v1/v2 */
 416                 if (get_user(reply_len, &old_hdr->reply_len))
 417                         return -EFAULT;
 418
 419                 if (reply_len >= 0)
 420                         return get_user(*pack_id, &old_hdr->pack_id);
 421
 422                 if (in_compat_syscall() &&
 423                     count >= sizeof(struct compat_sg_io_hdr)) {
 424                         struct compat_sg_io_hdr __user *hp = buf;
 425
 426                         return get_user(*pack_id, &hp->pack_id);
 427                 }
 428
 429                 if (count >= sizeof(struct sg_io_hdr)) {
 430                         struct sg_io_hdr __user *hp = buf;
 431
 432                         return get_user(*pack_id, &hp->pack_id);
 433                 }
 434         }
 435
 436         /* no valid header was passed, so ignore the pack_id */
 437         *pack_id = -1;
 438         return 0;
 439 }
 440
 441 static ssize_t
 442 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 443 {
 444         Sg_device *sdp;
 445         Sg_fd *sfp;
 446         Sg_request *srp;
 447         int req_pack_id = -1;
 448         bool busy;
 449         sg_io_hdr_t *hp;
 450         struct sg_header *old_hdr;
 451         int retval;
 452
 453         /*
 454          * This could cause a response to be stranded. Close the associated
 455          * file descriptor to free up any resources being held.
 456          */
 457         retval = sg_check_file_access(filp, __func__);
 458         if (retval)
 459                 return retval;
 460
 461         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 462                 return -ENXIO;
 463         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 464                                       "sg_read: count=%d\n", (int) count));
 465
 466         if (sfp->force_packid)
 467                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 468         if (retval)
 469                 return retval;
 470
 471         srp = sg_get_rq_mark(sfp, req_pack_id, &busy);
 472         if (!srp) {             /* now wait on packet to arrive */
 473                 if (filp->f_flags & O_NONBLOCK)
 474                         return -EAGAIN;
 475                 retval = wait_event_interruptible(sfp->read_wait,
 476                         ((srp = sg_get_rq_mark(sfp, req_pack_id, &busy)) ||
 477                         (!busy && atomic_read(&sdp->detaching))));
 478                 if (!srp)
 479                         /* signal or detaching */
 480                         return retval ? retval : -ENODEV;
 481         }
 482         if (srp->header.interface_id != '\0')
 483                 return sg_new_read(sfp, buf, count, srp);
 484
 485         hp = &srp->header;
 486         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 487         if (!old_hdr)
 488                 return -ENOMEM;
 489
 490         old_hdr->reply_len = (int) hp->timeout;
 491         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 492         old_hdr->pack_id = hp->pack_id;
 493         old_hdr->twelve_byte =
 494             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 495         old_hdr->target_status = hp->masked_status;
 496         old_hdr->host_status = hp->host_status;
 497         old_hdr->driver_status = hp->driver_status;
 498         if ((CHECK_CONDITION & hp->masked_status) ||
 499             (srp->sense_b[0] & 0x70) == 0x70) {
 500                 old_hdr->driver_status = DRIVER_SENSE;
 501                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 502                        sizeof (old_hdr->sense_buffer));
 503         }
 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                     (srp->sense_b[0] & 0x70) == 0x70) {
 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->driver_status = DRIVER_SENSE;
 587                         hp->sb_len_wr = len;
 588                 }
 589         }
 590         if (hp->masked_status || hp->host_status || hp->driver_status)
 591                 hp->info |= SG_INFO_CHECK;
 592         err = put_sg_io_hdr(hp, buf);
 593 err_out:
 594         err2 = sg_finish_rem_req(srp);
 595         sg_remove_request(sfp, srp);
 596         return err ? : err2 ? : count;
 597 }
 598
 599 static ssize_t
 600 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 601 {
 602         int mxsize, cmd_size, k;
 603         int input_size, blocking;
 604         unsigned char opcode;
 605         Sg_device *sdp;
 606         Sg_fd *sfp;
 607         Sg_request *srp;
 608         struct sg_header old_hdr;
 609         sg_io_hdr_t *hp;
 610         unsigned char cmnd[SG_MAX_CDB_SIZE];
 611         int retval;
 612
 613         retval = sg_check_file_access(filp, __func__);
 614         if (retval)
 615                 return retval;
 616
 617         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 618                 return -ENXIO;
 619         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 620                                       "sg_write: count=%d\n", (int) count));
 621         if (atomic_read(&sdp->detaching))
 622                 return -ENODEV;
 623         if (!((filp->f_flags & O_NONBLOCK) ||
 624               scsi_block_when_processing_errors(sdp->device)))
 625                 return -ENXIO;
 626
 627         if (count < SZ_SG_HEADER)
 628                 return -EIO;
 629         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 630                 return -EFAULT;
 631         blocking = !(filp->f_flags & O_NONBLOCK);
 632         if (old_hdr.reply_len < 0)
 633                 return sg_new_write(sfp, filp, buf, count,
 634                                     blocking, 0, 0, NULL);
 635         if (count < (SZ_SG_HEADER + 6))
 636                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 637
 638         buf += SZ_SG_HEADER;
 639         if (get_user(opcode, buf))
 640                 return -EFAULT;
 641
 642         if (!(srp = sg_add_request(sfp))) {
 643                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 644                                               "sg_write: queue full\n"));
 645                 return -EDOM;
 646         }
 647         mutex_lock(&sfp->f_mutex);
 648         if (sfp->next_cmd_len > 0) {
 649                 cmd_size = sfp->next_cmd_len;
 650                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 651         } else {
 652                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 653                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 654                         cmd_size = 12;
 655         }
 656         mutex_unlock(&sfp->f_mutex);
 657         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 658                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 659 /* Determine buffer size.  */
 660         input_size = count - cmd_size;
 661         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 662         mxsize -= SZ_SG_HEADER;
 663         input_size -= SZ_SG_HEADER;
 664         if (input_size < 0) {
 665                 sg_remove_request(sfp, srp);
 666                 return -EIO;    /* User did not pass enough bytes for this command. */
 667         }
 668         hp = &srp->header;
 669         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 670         hp->cmd_len = (unsigned char) cmd_size;
 671         hp->iovec_count = 0;
 672         hp->mx_sb_len = 0;
 673         if (input_size > 0)
 674                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 675                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 676         else
 677                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 678         hp->dxfer_len = mxsize;
 679         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 680             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 681                 hp->dxferp = (char __user *)buf + cmd_size;
 682         else
 683                 hp->dxferp = NULL;
 684         hp->sbp = NULL;
 685         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 686         hp->flags = input_size; /* structure abuse ... */
 687         hp->pack_id = old_hdr.pack_id;
 688         hp->usr_ptr = NULL;
 689         if (copy_from_user(cmnd, buf, cmd_size)) {
 690                 sg_remove_request(sfp, srp);
 691                 return -EFAULT;
 692         }
 693         /*
 694          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 695          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 696          * is a non-zero input_size, so emit a warning.
 697          */
 698         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 699                 printk_ratelimited(KERN_WARNING
 700                                    "sg_write: data in/out %d/%d bytes "
 701                                    "for SCSI command 0x%x-- guessing "
 702                                    "data in;\n   program %s not setting "
 703                                    "count and/or reply_len properly\n",
 704                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 705                                    input_size, (unsigned int) cmnd[0],
 706                                    current->comm);
 707         }
 708         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 709         return (k < 0) ? k : count;
 710 }
 711
 712 static ssize_t
 713 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 714                  size_t count, int blocking, int read_only, int sg_io_owned,
 715                  Sg_request **o_srp)
 716 {
 717         int k;
 718         Sg_request *srp;
 719         sg_io_hdr_t *hp;
 720         unsigned char cmnd[SG_MAX_CDB_SIZE];
 721         int timeout;
 722         unsigned long ul_timeout;
 723
 724         if (count < SZ_SG_IO_HDR)
 725                 return -EINVAL;
 726
 727         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 728         if (!(srp = sg_add_request(sfp))) {
 729                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 730                                               "sg_new_write: queue full\n"));
 731                 return -EDOM;
 732         }
 733         srp->sg_io_owned = sg_io_owned;
 734         hp = &srp->header;
 735         if (get_sg_io_hdr(hp, buf)) {
 736                 sg_remove_request(sfp, srp);
 737                 return -EFAULT;
 738         }
 739         if (hp->interface_id != 'S') {
 740                 sg_remove_request(sfp, srp);
 741                 return -ENOSYS;
 742         }
 743         if (hp->flags & SG_FLAG_MMAP_IO) {
 744                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 745                         sg_remove_request(sfp, srp);
 746                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 747                 }
 748                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 749                         sg_remove_request(sfp, srp);
 750                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 751                 }
 752                 if (sfp->res_in_use) {
 753                         sg_remove_request(sfp, srp);
 754                         return -EBUSY;  /* reserve buffer already being used */
 755                 }
 756         }
 757         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 758         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 759         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 760                 sg_remove_request(sfp, srp);
 761                 return -EMSGSIZE;
 762         }
 763         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 764                 sg_remove_request(sfp, srp);
 765                 return -EFAULT;
 766         }
 767         if (read_only && sg_allow_access(file, cmnd)) {
 768                 sg_remove_request(sfp, srp);
 769                 return -EPERM;
 770         }
 771         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 772         if (k < 0)
 773                 return k;
 774         if (o_srp)
 775                 *o_srp = srp;
 776         return count;
 777 }
 778
 779 static int
 780 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 781                 unsigned char *cmnd, int timeout, int blocking)
 782 {
 783         int k, at_head;
 784         Sg_device *sdp = sfp->parentdp;
 785         sg_io_hdr_t *hp = &srp->header;
 786
 787         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 788         hp->status = 0;
 789         hp->masked_status = 0;
 790         hp->msg_status = 0;
 791         hp->info = 0;
 792         hp->host_status = 0;
 793         hp->driver_status = 0;
 794         hp->resid = 0;
 795         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 796                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 797                         (int) cmnd[0], (int) hp->cmd_len));
 798
 799         if (hp->dxfer_len >= SZ_256M) {
 800                 sg_remove_request(sfp, srp);
 801                 return -EINVAL;
 802         }
 803
 804         k = sg_start_req(srp, cmnd);
 805         if (k) {
 806                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 807                         "sg_common_write: start_req err=%d\n", k));
 808                 sg_finish_rem_req(srp);
 809                 sg_remove_request(sfp, srp);
 810                 return k;       /* probably out of space --> ENOMEM */
 811         }
 812         if (atomic_read(&sdp->detaching)) {
 813                 if (srp->bio) {
 814                         scsi_req_free_cmd(scsi_req(srp->rq));
 815                         blk_put_request(srp->rq);
 816                         srp->rq = NULL;
 817                 }
 818
 819                 sg_finish_rem_req(srp);
 820                 sg_remove_request(sfp, srp);
 821                 return -ENODEV;
 822         }
 823
 824         hp->duration = jiffies_to_msecs(jiffies);
 825         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 826             (SG_FLAG_Q_AT_TAIL & hp->flags))
 827                 at_head = 0;
 828         else
 829                 at_head = 1;
 830
 831         srp->rq->timeout = timeout;
 832         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 833         blk_execute_rq_nowait(NULL, 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));
 942                 write_lock_irq(&sfp->rq_list_lock);
 943                 if (srp->done) {
 944                         srp->done = 2;
 945                         write_unlock_irq(&sfp->rq_list_lock);
 946                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 947                         return (result < 0) ? result : 0;
 948                 }
 949                 srp->orphan = 1;
 950                 write_unlock_irq(&sfp->rq_list_lock);
 951                 return result;  /* -ERESTARTSYS because signal hit process */
 952         case SG_SET_TIMEOUT:
 953                 result = get_user(val, ip);
 954                 if (result)
 955                         return result;
 956                 if (val < 0)
 957                         return -EIO;
 958                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 959                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 960                                     INT_MAX);
 961                 sfp->timeout_user = val;
 962                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 963
 964                 return 0;
 965         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 966                                 /* strange ..., for backward compatibility */
 967                 return sfp->timeout_user;
 968         case SG_SET_FORCE_LOW_DMA:
 969                 /*
 970                  * N.B. This ioctl never worked properly, but failed to
 971                  * return an error value. So returning '0' to keep compability
 972                  * with legacy applications.
 973                  */
 974                 return 0;
 975         case SG_GET_LOW_DMA:
 976                 return put_user(0, ip);
 977         case SG_GET_SCSI_ID:
 978                 {
 979                         sg_scsi_id_t v;
 980
 981                         if (atomic_read(&sdp->detaching))
 982                                 return -ENODEV;
 983                         memset(&v, 0, sizeof(v));
 984                         v.host_no = sdp->device->host->host_no;
 985                         v.channel = sdp->device->channel;
 986                         v.scsi_id = sdp->device->id;
 987                         v.lun = sdp->device->lun;
 988                         v.scsi_type = sdp->device->type;
 989                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 990                         v.d_queue_depth = sdp->device->queue_depth;
 991                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 992                                 return -EFAULT;
 993                         return 0;
 994                 }
 995         case SG_SET_FORCE_PACK_ID:
 996                 result = get_user(val, ip);
 997                 if (result)
 998                         return result;
 999                 sfp->force_packid = val ? 1 : 0;
1000                 return 0;
1001         case SG_GET_PACK_ID:
1002                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1003                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1004                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1005                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1006                                                        iflags);
1007                                 return put_user(srp->header.pack_id, ip);
1008                         }
1009                 }
1010                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1011                 return put_user(-1, ip);
1012         case SG_GET_NUM_WAITING:
1013                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1014                 val = 0;
1015                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1016                         if ((1 == srp->done) && (!srp->sg_io_owned))
1017                                 ++val;
1018                 }
1019                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1020                 return put_user(val, ip);
1021         case SG_GET_SG_TABLESIZE:
1022                 return put_user(sdp->sg_tablesize, ip);
1023         case SG_SET_RESERVED_SIZE:
1024                 result = get_user(val, ip);
1025                 if (result)
1026                         return result;
1027                 if (val < 0)
1028                         return -EINVAL;
1029                 val = min_t(int, val,
1030                             max_sectors_bytes(sdp->device->request_queue));
1031                 mutex_lock(&sfp->f_mutex);
1032                 if (val != sfp->reserve.bufflen) {
1033                         if (sfp->mmap_called ||
1034                             sfp->res_in_use) {
1035                                 mutex_unlock(&sfp->f_mutex);
1036                                 return -EBUSY;
1037                         }
1038
1039                         sg_remove_scat(sfp, &sfp->reserve);
1040                         sg_build_reserve(sfp, val);
1041                 }
1042                 mutex_unlock(&sfp->f_mutex);
1043                 return 0;
1044         case SG_GET_RESERVED_SIZE:
1045                 val = min_t(int, sfp->reserve.bufflen,
1046                             max_sectors_bytes(sdp->device->request_queue));
1047                 return put_user(val, ip);
1048         case SG_SET_COMMAND_Q:
1049                 result = get_user(val, ip);
1050                 if (result)
1051                         return result;
1052                 sfp->cmd_q = val ? 1 : 0;
1053                 return 0;
1054         case SG_GET_COMMAND_Q:
1055                 return put_user((int) sfp->cmd_q, ip);
1056         case SG_SET_KEEP_ORPHAN:
1057                 result = get_user(val, ip);
1058                 if (result)
1059                         return result;
1060                 sfp->keep_orphan = val;
1061                 return 0;
1062         case SG_GET_KEEP_ORPHAN:
1063                 return put_user((int) sfp->keep_orphan, ip);
1064         case SG_NEXT_CMD_LEN:
1065                 result = get_user(val, ip);
1066                 if (result)
1067                         return result;
1068                 if (val > SG_MAX_CDB_SIZE)
1069                         return -ENOMEM;
1070                 sfp->next_cmd_len = (val > 0) ? val : 0;
1071                 return 0;
1072         case SG_GET_VERSION_NUM:
1073                 return put_user(sg_version_num, ip);
1074         case SG_GET_ACCESS_COUNT:
1075                 /* faked - we don't have a real access count anymore */
1076                 val = (sdp->device ? 1 : 0);
1077                 return put_user(val, ip);
1078         case SG_GET_REQUEST_TABLE:
1079                 {
1080                         sg_req_info_t *rinfo;
1081
1082                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1083                                         GFP_KERNEL);
1084                         if (!rinfo)
1085                                 return -ENOMEM;
1086                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1087                         sg_fill_request_table(sfp, rinfo);
1088                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1089         #ifdef CONFIG_COMPAT
1090                         if (in_compat_syscall())
1091                                 result = put_compat_request_table(p, rinfo);
1092                         else
1093         #endif
1094                                 result = copy_to_user(p, rinfo,
1095                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1096                         result = result ? -EFAULT : 0;
1097                         kfree(rinfo);
1098                         return result;
1099                 }
1100         case SG_EMULATED_HOST:
1101                 if (atomic_read(&sdp->detaching))
1102                         return -ENODEV;
1103                 return put_user(sdp->device->host->hostt->emulated, ip);
1104         case SCSI_IOCTL_SEND_COMMAND:
1105                 if (atomic_read(&sdp->detaching))
1106                         return -ENODEV;
1107                 return scsi_ioctl(sdp->device, NULL, filp->f_mode, cmd_in, p);
1108         case SG_SET_DEBUG:
1109                 result = get_user(val, ip);
1110                 if (result)
1111                         return result;
1112                 sdp->sgdebug = (char) val;
1113                 return 0;
1114         case BLKSECTGET:
1115                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1116                                 ip);
1117         case BLKTRACESETUP:
1118                 return blk_trace_setup(sdp->device->request_queue, sdp->name,
1119                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1120                                        NULL, p);
1121         case BLKTRACESTART:
1122                 return blk_trace_startstop(sdp->device->request_queue, 1);
1123         case BLKTRACESTOP:
1124                 return blk_trace_startstop(sdp->device->request_queue, 0);
1125         case BLKTRACETEARDOWN:
1126                 return blk_trace_remove(sdp->device->request_queue);
1127         case SCSI_IOCTL_GET_IDLUN:
1128         case SCSI_IOCTL_GET_BUS_NUMBER:
1129         case SCSI_IOCTL_PROBE_HOST:
1130         case SG_GET_TRANSFORM:
1131         case SG_SCSI_RESET:
1132                 if (atomic_read(&sdp->detaching))
1133                         return -ENODEV;
1134                 break;
1135         default:
1136                 if (read_only)
1137                         return -EPERM;  /* don't know so take safe approach */
1138                 break;
1139         }
1140
1141         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1142                         cmd_in, filp->f_flags & O_NDELAY);
1143         if (result)
1144                 return result;
1145
1146         return -ENOIOCTLCMD;
1147 }
1148
1149 static long
1150 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1151 {
1152         void __user *p = (void __user *)arg;
1153         Sg_device *sdp;
1154         Sg_fd *sfp;
1155         int ret;
1156
1157         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1158                 return -ENXIO;
1159
1160         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1161         if (ret != -ENOIOCTLCMD)
1162                 return ret;
1163         return scsi_ioctl(sdp->device, NULL, filp->f_mode, cmd_in, p);
1164 }
1165
1166 static __poll_t
1167 sg_poll(struct file *filp, poll_table * wait)
1168 {
1169         __poll_t res = 0;
1170         Sg_device *sdp;
1171         Sg_fd *sfp;
1172         Sg_request *srp;
1173         int count = 0;
1174         unsigned long iflags;
1175
1176         sfp = filp->private_data;
1177         if (!sfp)
1178                 return EPOLLERR;
1179         sdp = sfp->parentdp;
1180         if (!sdp)
1181                 return EPOLLERR;
1182         poll_wait(filp, &sfp->read_wait, wait);
1183         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1184         list_for_each_entry(srp, &sfp->rq_list, entry) {
1185                 /* if any read waiting, flag it */
1186                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1187                         res = EPOLLIN | EPOLLRDNORM;
1188                 ++count;
1189         }
1190         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1191
1192         if (atomic_read(&sdp->detaching))
1193                 res |= EPOLLHUP;
1194         else if (!sfp->cmd_q) {
1195                 if (0 == count)
1196                         res |= EPOLLOUT | EPOLLWRNORM;
1197         } else if (count < SG_MAX_QUEUE)
1198                 res |= EPOLLOUT | EPOLLWRNORM;
1199         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1200                                       "sg_poll: res=0x%x\n", (__force u32) res));
1201         return res;
1202 }
1203
1204 static int
1205 sg_fasync(int fd, struct file *filp, int mode)
1206 {
1207         Sg_device *sdp;
1208         Sg_fd *sfp;
1209
1210         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1211                 return -ENXIO;
1212         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1213                                       "sg_fasync: mode=%d\n", mode));
1214
1215         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1216 }
1217
1218 static vm_fault_t
1219 sg_vma_fault(struct vm_fault *vmf)
1220 {
1221         struct vm_area_struct *vma = vmf->vma;
1222         Sg_fd *sfp;
1223         unsigned long offset, len, sa;
1224         Sg_scatter_hold *rsv_schp;
1225         int k, length;
1226
1227         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1228                 return VM_FAULT_SIGBUS;
1229         rsv_schp = &sfp->reserve;
1230         offset = vmf->pgoff << PAGE_SHIFT;
1231         if (offset >= rsv_schp->bufflen)
1232                 return VM_FAULT_SIGBUS;
1233         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1234                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1235                                       offset, rsv_schp->k_use_sg));
1236         sa = vma->vm_start;
1237         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1238         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1239                 len = vma->vm_end - sa;
1240                 len = (len < length) ? len : length;
1241                 if (offset < len) {
1242                         struct page *page = nth_page(rsv_schp->pages[k],
1243                                                      offset >> PAGE_SHIFT);
1244                         get_page(page); /* increment page count */
1245                         vmf->page = page;
1246                         return 0; /* success */
1247                 }
1248                 sa += len;
1249                 offset -= len;
1250         }
1251
1252         return VM_FAULT_SIGBUS;
1253 }
1254
1255 static const struct vm_operations_struct sg_mmap_vm_ops = {
1256         .fault = sg_vma_fault,
1257 };
1258
1259 static int
1260 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1261 {
1262         Sg_fd *sfp;
1263         unsigned long req_sz, len, sa;
1264         Sg_scatter_hold *rsv_schp;
1265         int k, length;
1266         int ret = 0;
1267
1268         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1269                 return -ENXIO;
1270         req_sz = vma->vm_end - vma->vm_start;
1271         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1272                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1273                                       (void *) vma->vm_start, (int) req_sz));
1274         if (vma->vm_pgoff)
1275                 return -EINVAL; /* want no offset */
1276         rsv_schp = &sfp->reserve;
1277         mutex_lock(&sfp->f_mutex);
1278         if (req_sz > rsv_schp->bufflen) {
1279                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1280                 goto out;
1281         }
1282
1283         sa = vma->vm_start;
1284         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1285         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1286                 len = vma->vm_end - sa;
1287                 len = (len < length) ? len : length;
1288                 sa += len;
1289         }
1290
1291         sfp->mmap_called = 1;
1292         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1293         vma->vm_private_data = sfp;
1294         vma->vm_ops = &sg_mmap_vm_ops;
1295 out:
1296         mutex_unlock(&sfp->f_mutex);
1297         return ret;
1298 }
1299
1300 static void
1301 sg_rq_end_io_usercontext(struct work_struct *work)
1302 {
1303         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1304         struct sg_fd *sfp = srp->parentfp;
1305
1306         sg_finish_rem_req(srp);
1307         sg_remove_request(sfp, srp);
1308         kref_put(&sfp->f_ref, sg_remove_sfp);
1309 }
1310
1311 /*
1312  * This function is a "bottom half" handler that is called by the mid
1313  * level when a command is completed (or has failed).
1314  */
1315 static void
1316 sg_rq_end_io(struct request *rq, blk_status_t status)
1317 {
1318         struct sg_request *srp = rq->end_io_data;
1319         struct scsi_request *req = scsi_req(rq);
1320         Sg_device *sdp;
1321         Sg_fd *sfp;
1322         unsigned long iflags;
1323         unsigned int ms;
1324         char *sense;
1325         int result, resid, done = 1;
1326
1327         if (WARN_ON(srp->done != 0))
1328                 return;
1329
1330         sfp = srp->parentfp;
1331         if (WARN_ON(sfp == NULL))
1332                 return;
1333
1334         sdp = sfp->parentdp;
1335         if (unlikely(atomic_read(&sdp->detaching)))
1336                 pr_info("%s: device detaching\n", __func__);
1337
1338         sense = req->sense;
1339         result = req->result;
1340         resid = req->resid_len;
1341
1342         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1343                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1344                                       srp->header.pack_id, result));
1345         srp->header.resid = resid;
1346         ms = jiffies_to_msecs(jiffies);
1347         srp->header.duration = (ms > srp->header.duration) ?
1348                                 (ms - srp->header.duration) : 0;
1349         if (0 != result) {
1350                 struct scsi_sense_hdr sshdr;
1351
1352                 srp->header.status = 0xff & result;
1353                 srp->header.masked_status = status_byte(result);
1354                 srp->header.msg_status = COMMAND_COMPLETE;
1355                 srp->header.host_status = host_byte(result);
1356                 srp->header.driver_status = driver_byte(result);
1357                 if ((sdp->sgdebug > 0) &&
1358                     ((CHECK_CONDITION == srp->header.masked_status) ||
1359                      (COMMAND_TERMINATED == srp->header.masked_status)))
1360                         __scsi_print_sense(sdp->device, __func__, sense,
1361                                            SCSI_SENSE_BUFFERSIZE);
1362
1363                 /* Following if statement is a patch supplied by Eric Youngdale */
1364                 if (driver_byte(result) != 0
1365                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1366                     && !scsi_sense_is_deferred(&sshdr)
1367                     && sshdr.sense_key == UNIT_ATTENTION
1368                     && sdp->device->removable) {
1369                         /* Detected possible disc change. Set the bit - this */
1370                         /* may be used if there are filesystems using this device */
1371                         sdp->device->changed = 1;
1372                 }
1373         }
1374
1375         if (req->sense_len)
1376                 memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1377
1378         /* Rely on write phase to clean out srp status values, so no "else" */
1379
1380         /*
1381          * Free the request as soon as it is complete so that its resources
1382          * can be reused without waiting for userspace to read() the
1383          * result.  But keep the associated bio (if any) around until
1384          * blk_rq_unmap_user() can be called from user context.
1385          */
1386         srp->rq = NULL;
1387         scsi_req_free_cmd(scsi_req(rq));
1388         blk_put_request(rq);
1389
1390         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1391         if (unlikely(srp->orphan)) {
1392                 if (sfp->keep_orphan)
1393                         srp->sg_io_owned = 0;
1394                 else
1395                         done = 0;
1396         }
1397         srp->done = done;
1398         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1399
1400         if (likely(done)) {
1401                 /* Now wake up any sg_read() that is waiting for this
1402                  * packet.
1403                  */
1404                 wake_up_interruptible(&sfp->read_wait);
1405                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1406                 kref_put(&sfp->f_ref, sg_remove_sfp);
1407         } else {
1408                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1409                 schedule_work(&srp->ew.work);
1410         }
1411 }
1412
1413 static const struct file_operations sg_fops = {
1414         .owner = THIS_MODULE,
1415         .read = sg_read,
1416         .write = sg_write,
1417         .poll = sg_poll,
1418         .unlocked_ioctl = sg_ioctl,
1419         .compat_ioctl = compat_ptr_ioctl,
1420         .open = sg_open,
1421         .mmap = sg_mmap,
1422         .release = sg_release,
1423         .fasync = sg_fasync,
1424         .llseek = no_llseek,
1425 };
1426
1427 static struct class *sg_sysfs_class;
1428
1429 static int sg_sysfs_valid = 0;
1430
1431 static Sg_device *
1432 sg_alloc(struct scsi_device *scsidp)
1433 {
1434         struct request_queue *q = scsidp->request_queue;
1435         Sg_device *sdp;
1436         unsigned long iflags;
1437         int error;
1438         u32 k;
1439
1440         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1441         if (!sdp) {
1442                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1443                             "failure\n", __func__);
1444                 return ERR_PTR(-ENOMEM);
1445         }
1446
1447         idr_preload(GFP_KERNEL);
1448         write_lock_irqsave(&sg_index_lock, iflags);
1449
1450         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1451         if (error < 0) {
1452                 if (error == -ENOSPC) {
1453                         sdev_printk(KERN_WARNING, scsidp,
1454                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1455                                     scsidp->type, SG_MAX_DEVS - 1);
1456                         error = -ENODEV;
1457                 } else {
1458                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1459                                     "allocation Sg_device failure: %d\n",
1460                                     __func__, error);
1461                 }
1462                 goto out_unlock;
1463         }
1464         k = error;
1465
1466         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1467                                         "sg_alloc: dev=%d \n", k));
1468         sprintf(sdp->name, "sg%d", k);
1469         sdp->device = scsidp;
1470         mutex_init(&sdp->open_rel_lock);
1471         INIT_LIST_HEAD(&sdp->sfds);
1472         init_waitqueue_head(&sdp->open_wait);
1473         atomic_set(&sdp->detaching, 0);
1474         rwlock_init(&sdp->sfd_lock);
1475         sdp->sg_tablesize = queue_max_segments(q);
1476         sdp->index = k;
1477         kref_init(&sdp->d_ref);
1478         error = 0;
1479
1480 out_unlock:
1481         write_unlock_irqrestore(&sg_index_lock, iflags);
1482         idr_preload_end();
1483
1484         if (error) {
1485                 kfree(sdp);
1486                 return ERR_PTR(error);
1487         }
1488         return sdp;
1489 }
1490
1491 static int
1492 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1493 {
1494         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1495         Sg_device *sdp = NULL;
1496         struct cdev * cdev = NULL;
1497         int error;
1498         unsigned long iflags;
1499
1500         error = -ENOMEM;
1501         cdev = cdev_alloc();
1502         if (!cdev) {
1503                 pr_warn("%s: cdev_alloc failed\n", __func__);
1504                 goto out;
1505         }
1506         cdev->owner = THIS_MODULE;
1507         cdev->ops = &sg_fops;
1508
1509         sdp = sg_alloc(scsidp);
1510         if (IS_ERR(sdp)) {
1511                 pr_warn("%s: sg_alloc failed\n", __func__);
1512                 error = PTR_ERR(sdp);
1513                 goto out;
1514         }
1515
1516         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1517         if (error)
1518                 goto cdev_add_err;
1519
1520         sdp->cdev = cdev;
1521         if (sg_sysfs_valid) {
1522                 struct device *sg_class_member;
1523
1524                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1525                                                 MKDEV(SCSI_GENERIC_MAJOR,
1526                                                       sdp->index),
1527                                                 sdp, "%s", sdp->name);
1528                 if (IS_ERR(sg_class_member)) {
1529                         pr_err("%s: device_create failed\n", __func__);
1530                         error = PTR_ERR(sg_class_member);
1531                         goto cdev_add_err;
1532                 }
1533                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1534                                           &sg_class_member->kobj, "generic");
1535                 if (error)
1536                         pr_err("%s: unable to make symlink 'generic' back "
1537                                "to sg%d\n", __func__, sdp->index);
1538         } else
1539                 pr_warn("%s: sg_sys Invalid\n", __func__);
1540
1541         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1542                     "type %d\n", sdp->index, scsidp->type);
1543
1544         dev_set_drvdata(cl_dev, sdp);
1545
1546         return 0;
1547
1548 cdev_add_err:
1549         write_lock_irqsave(&sg_index_lock, iflags);
1550         idr_remove(&sg_index_idr, sdp->index);
1551         write_unlock_irqrestore(&sg_index_lock, iflags);
1552         kfree(sdp);
1553
1554 out:
1555         if (cdev)
1556                 cdev_del(cdev);
1557         return error;
1558 }
1559
1560 static void
1561 sg_device_destroy(struct kref *kref)
1562 {
1563         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1564         unsigned long flags;
1565
1566         /* CAUTION!  Note that the device can still be found via idr_find()
1567          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1568          * any other cleanup.
1569          */
1570
1571         write_lock_irqsave(&sg_index_lock, flags);
1572         idr_remove(&sg_index_idr, sdp->index);
1573         write_unlock_irqrestore(&sg_index_lock, flags);
1574
1575         SCSI_LOG_TIMEOUT(3,
1576                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1577
1578         kfree(sdp);
1579 }
1580
1581 static void
1582 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1583 {
1584         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1585         Sg_device *sdp = dev_get_drvdata(cl_dev);
1586         unsigned long iflags;
1587         Sg_fd *sfp;
1588         int val;
1589
1590         if (!sdp)
1591                 return;
1592         /* want sdp->detaching non-zero as soon as possible */
1593         val = atomic_inc_return(&sdp->detaching);
1594         if (val > 1)
1595                 return; /* only want to do following once per device */
1596
1597         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1598                                       "%s\n", __func__));
1599
1600         read_lock_irqsave(&sdp->sfd_lock, iflags);
1601         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1602                 wake_up_interruptible_all(&sfp->read_wait);
1603                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1604         }
1605         wake_up_interruptible_all(&sdp->open_wait);
1606         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1607
1608         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1609         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1610         cdev_del(sdp->cdev);
1611         sdp->cdev = NULL;
1612
1613         kref_put(&sdp->d_ref, sg_device_destroy);
1614 }
1615
1616 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1617 module_param_named(def_reserved_size, def_reserved_size, int,
1618                    S_IRUGO | S_IWUSR);
1619 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1620
1621 MODULE_AUTHOR("Douglas Gilbert");
1622 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1623 MODULE_LICENSE("GPL");
1624 MODULE_VERSION(SG_VERSION_STR);
1625 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1626
1627 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1628                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1629 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1630 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1631
1632 static int __init
1633 init_sg(void)
1634 {
1635         int rc;
1636
1637         if (scatter_elem_sz < PAGE_SIZE) {
1638                 scatter_elem_sz = PAGE_SIZE;
1639                 scatter_elem_sz_prev = scatter_elem_sz;
1640         }
1641         if (def_reserved_size >= 0)
1642                 sg_big_buff = def_reserved_size;
1643         else
1644                 def_reserved_size = sg_big_buff;
1645
1646         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1647                                     SG_MAX_DEVS, "sg");
1648         if (rc)
1649                 return rc;
1650         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1651         if ( IS_ERR(sg_sysfs_class) ) {
1652                 rc = PTR_ERR(sg_sysfs_class);
1653                 goto err_out;
1654         }
1655         sg_sysfs_valid = 1;
1656         rc = scsi_register_interface(&sg_interface);
1657         if (0 == rc) {
1658 #ifdef CONFIG_SCSI_PROC_FS
1659                 sg_proc_init();
1660 #endif                          /* CONFIG_SCSI_PROC_FS */
1661                 return 0;
1662         }
1663         class_destroy(sg_sysfs_class);
1664 err_out:
1665         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1666         return rc;
1667 }
1668
1669 static void __exit
1670 exit_sg(void)
1671 {
1672 #ifdef CONFIG_SCSI_PROC_FS
1673         remove_proc_subtree("scsi/sg", NULL);
1674 #endif                          /* CONFIG_SCSI_PROC_FS */
1675         scsi_unregister_interface(&sg_interface);
1676         class_destroy(sg_sysfs_class);
1677         sg_sysfs_valid = 0;
1678         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1679                                  SG_MAX_DEVS);
1680         idr_destroy(&sg_index_idr);
1681 }
1682
1683 static int
1684 sg_start_req(Sg_request *srp, unsigned char *cmd)
1685 {
1686         int res;
1687         struct request *rq;
1688         struct scsi_request *req;
1689         Sg_fd *sfp = srp->parentfp;
1690         sg_io_hdr_t *hp = &srp->header;
1691         int dxfer_len = (int) hp->dxfer_len;
1692         int dxfer_dir = hp->dxfer_direction;
1693         unsigned int iov_count = hp->iovec_count;
1694         Sg_scatter_hold *req_schp = &srp->data;
1695         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1696         struct request_queue *q = sfp->parentdp->device->request_queue;
1697         struct rq_map_data *md, map_data;
1698         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1699         unsigned char *long_cmdp = NULL;
1700
1701         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1702                                       "sg_start_req: dxfer_len=%d\n",
1703                                       dxfer_len));
1704
1705         if (hp->cmd_len > BLK_MAX_CDB) {
1706                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1707                 if (!long_cmdp)
1708                         return -ENOMEM;
1709         }
1710
1711         /*
1712          * NOTE
1713          *
1714          * With scsi-mq enabled, there are a fixed number of preallocated
1715          * requests equal in number to shost->can_queue.  If all of the
1716          * preallocated requests are already in use, then blk_get_request()
1717          * will sleep until an active command completes, freeing up a request.
1718          * Although waiting in an asynchronous interface is less than ideal, we
1719          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1720          * not expect an EWOULDBLOCK from this condition.
1721          */
1722         rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1723                         REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1724         if (IS_ERR(rq)) {
1725                 kfree(long_cmdp);
1726                 return PTR_ERR(rq);
1727         }
1728         req = scsi_req(rq);
1729
1730         if (hp->cmd_len > BLK_MAX_CDB)
1731                 req->cmd = long_cmdp;
1732         memcpy(req->cmd, cmd, hp->cmd_len);
1733         req->cmd_len = hp->cmd_len;
1734
1735         srp->rq = rq;
1736         rq->end_io_data = srp;
1737         req->retries = SG_DEFAULT_RETRIES;
1738
1739         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1740                 return 0;
1741
1742         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1743             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1744             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1745                 md = NULL;
1746         else
1747                 md = &map_data;
1748
1749         if (md) {
1750                 mutex_lock(&sfp->f_mutex);
1751                 if (dxfer_len <= rsv_schp->bufflen &&
1752                     !sfp->res_in_use) {
1753                         sfp->res_in_use = 1;
1754                         sg_link_reserve(sfp, srp, dxfer_len);
1755                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1756                         res = -EBUSY; /* sfp->res_in_use == 1 */
1757                         if (dxfer_len > rsv_schp->bufflen)
1758                                 res = -ENOMEM;
1759                         mutex_unlock(&sfp->f_mutex);
1760                         return res;
1761                 } else {
1762                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1763                         if (res) {
1764                                 mutex_unlock(&sfp->f_mutex);
1765                                 return res;
1766                         }
1767                 }
1768                 mutex_unlock(&sfp->f_mutex);
1769
1770                 md->pages = req_schp->pages;
1771                 md->page_order = req_schp->page_order;
1772                 md->nr_entries = req_schp->k_use_sg;
1773                 md->offset = 0;
1774                 md->null_mapped = hp->dxferp ? 0 : 1;
1775                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1776                         md->from_user = 1;
1777                 else
1778                         md->from_user = 0;
1779         }
1780
1781         if (iov_count) {
1782                 struct iovec *iov = NULL;
1783                 struct iov_iter i;
1784
1785                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1786                 if (res < 0)
1787                         return res;
1788
1789                 iov_iter_truncate(&i, hp->dxfer_len);
1790                 if (!iov_iter_count(&i)) {
1791                         kfree(iov);
1792                         return -EINVAL;
1793                 }
1794
1795                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1796                 kfree(iov);
1797         } else
1798                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1799                                       hp->dxfer_len, GFP_ATOMIC);
1800
1801         if (!res) {
1802                 srp->bio = rq->bio;
1803
1804                 if (!md) {
1805                         req_schp->dio_in_use = 1;
1806                         hp->info |= SG_INFO_DIRECT_IO;
1807                 }
1808         }
1809         return res;
1810 }
1811
1812 static int
1813 sg_finish_rem_req(Sg_request *srp)
1814 {
1815         int ret = 0;
1816
1817         Sg_fd *sfp = srp->parentfp;
1818         Sg_scatter_hold *req_schp = &srp->data;
1819
1820         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1821                                       "sg_finish_rem_req: res_used=%d\n",
1822                                       (int) srp->res_used));
1823         if (srp->bio)
1824                 ret = blk_rq_unmap_user(srp->bio);
1825
1826         if (srp->rq) {
1827                 scsi_req_free_cmd(scsi_req(srp->rq));
1828                 blk_put_request(srp->rq);
1829         }
1830
1831         if (srp->res_used)
1832                 sg_unlink_reserve(sfp, srp);
1833         else
1834                 sg_remove_scat(sfp, req_schp);
1835
1836         return ret;
1837 }
1838
1839 static int
1840 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1841 {
1842         int sg_bufflen = tablesize * sizeof(struct page *);
1843         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1844
1845         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1846         if (!schp->pages)
1847                 return -ENOMEM;
1848         schp->sglist_len = sg_bufflen;
1849         return tablesize;       /* number of scat_gath elements allocated */
1850 }
1851
1852 static int
1853 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1854 {
1855         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1856         int sg_tablesize = sfp->parentdp->sg_tablesize;
1857         int blk_size = buff_size, order;
1858         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1859
1860         if (blk_size < 0)
1861                 return -EFAULT;
1862         if (0 == blk_size)
1863                 ++blk_size;     /* don't know why */
1864         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1865         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1866         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1867                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1868                 buff_size, blk_size));
1869
1870         /* N.B. ret_sz carried into this block ... */
1871         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1872         if (mx_sc_elems < 0)
1873                 return mx_sc_elems;     /* most likely -ENOMEM */
1874
1875         num = scatter_elem_sz;
1876         if (unlikely(num != scatter_elem_sz_prev)) {
1877                 if (num < PAGE_SIZE) {
1878                         scatter_elem_sz = PAGE_SIZE;
1879                         scatter_elem_sz_prev = PAGE_SIZE;
1880                 } else
1881                         scatter_elem_sz_prev = num;
1882         }
1883
1884         order = get_order(num);
1885 retry:
1886         ret_sz = 1 << (PAGE_SHIFT + order);
1887
1888         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1889              k++, rem_sz -= ret_sz) {
1890
1891                 num = (rem_sz > scatter_elem_sz_prev) ?
1892                         scatter_elem_sz_prev : rem_sz;
1893
1894                 schp->pages[k] = alloc_pages(gfp_mask, order);
1895                 if (!schp->pages[k])
1896                         goto out;
1897
1898                 if (num == scatter_elem_sz_prev) {
1899                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1900                                 scatter_elem_sz = ret_sz;
1901                                 scatter_elem_sz_prev = ret_sz;
1902                         }
1903                 }
1904
1905                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1906                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1907                                  k, num, ret_sz));
1908         }               /* end of for loop */
1909
1910         schp->page_order = order;
1911         schp->k_use_sg = k;
1912         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1913                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1914                          k, rem_sz));
1915
1916         schp->bufflen = blk_size;
1917         if (rem_sz > 0) /* must have failed */
1918                 return -ENOMEM;
1919         return 0;
1920 out:
1921         for (i = 0; i < k; i++)
1922                 __free_pages(schp->pages[i], order);
1923
1924         if (--order >= 0)
1925                 goto retry;
1926
1927         return -ENOMEM;
1928 }
1929
1930 static void
1931 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1932 {
1933         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1934                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1935         if (schp->pages && schp->sglist_len > 0) {
1936                 if (!schp->dio_in_use) {
1937                         int k;
1938
1939                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1940                                 SCSI_LOG_TIMEOUT(5,
1941                                         sg_printk(KERN_INFO, sfp->parentdp,
1942                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1943                                         k, schp->pages[k]));
1944                                 __free_pages(schp->pages[k], schp->page_order);
1945                         }
1946
1947                         kfree(schp->pages);
1948                 }
1949         }
1950         memset(schp, 0, sizeof (*schp));
1951 }
1952
1953 static int
1954 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1955 {
1956         Sg_scatter_hold *schp = &srp->data;
1957         int k, num;
1958
1959         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1960                          "sg_read_oxfer: num_read_xfer=%d\n",
1961                          num_read_xfer));
1962         if ((!outp) || (num_read_xfer <= 0))
1963                 return 0;
1964
1965         num = 1 << (PAGE_SHIFT + schp->page_order);
1966         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1967                 if (num > num_read_xfer) {
1968                         if (copy_to_user(outp, page_address(schp->pages[k]),
1969                                            num_read_xfer))
1970                                 return -EFAULT;
1971                         break;
1972                 } else {
1973                         if (copy_to_user(outp, page_address(schp->pages[k]),
1974                                            num))
1975                                 return -EFAULT;
1976                         num_read_xfer -= num;
1977                         if (num_read_xfer <= 0)
1978                                 break;
1979                         outp += num;
1980                 }
1981         }
1982
1983         return 0;
1984 }
1985
1986 static void
1987 sg_build_reserve(Sg_fd * sfp, int req_size)
1988 {
1989         Sg_scatter_hold *schp = &sfp->reserve;
1990
1991         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1992                          "sg_build_reserve: req_size=%d\n", req_size));
1993         do {
1994                 if (req_size < PAGE_SIZE)
1995                         req_size = PAGE_SIZE;
1996                 if (0 == sg_build_indirect(schp, sfp, req_size))
1997                         return;
1998                 else
1999                         sg_remove_scat(sfp, schp);
2000                 req_size >>= 1; /* divide by 2 */
2001         } while (req_size > (PAGE_SIZE / 2));
2002 }
2003
2004 static void
2005 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2006 {
2007         Sg_scatter_hold *req_schp = &srp->data;
2008         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2009         int k, num, rem;
2010
2011         srp->res_used = 1;
2012         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2013                          "sg_link_reserve: size=%d\n", size));
2014         rem = size;
2015
2016         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2017         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2018                 if (rem <= num) {
2019                         req_schp->k_use_sg = k + 1;
2020                         req_schp->sglist_len = rsv_schp->sglist_len;
2021                         req_schp->pages = rsv_schp->pages;
2022
2023                         req_schp->bufflen = size;
2024                         req_schp->page_order = rsv_schp->page_order;
2025                         break;
2026                 } else
2027                         rem -= num;
2028         }
2029
2030         if (k >= rsv_schp->k_use_sg)
2031                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2032                                  "sg_link_reserve: BAD size\n"));
2033 }
2034
2035 static void
2036 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2037 {
2038         Sg_scatter_hold *req_schp = &srp->data;
2039
2040         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2041                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2042                                       (int) req_schp->k_use_sg));
2043         req_schp->k_use_sg = 0;
2044         req_schp->bufflen = 0;
2045         req_schp->pages = NULL;
2046         req_schp->page_order = 0;
2047         req_schp->sglist_len = 0;
2048         srp->res_used = 0;
2049         /* Called without mutex lock to avoid deadlock */
2050         sfp->res_in_use = 0;
2051 }
2052
2053 static Sg_request *
2054 sg_get_rq_mark(Sg_fd * sfp, int pack_id, bool *busy)
2055 {
2056         Sg_request *resp;
2057         unsigned long iflags;
2058
2059         *busy = false;
2060         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2061         list_for_each_entry(resp, &sfp->rq_list, entry) {
2062                 /* look for requests that are not SG_IO owned */
2063                 if ((!resp->sg_io_owned) &&
2064                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2065                         switch (resp->done) {
2066                         case 0: /* request active */
2067                                 *busy = true;
2068                                 break;
2069                         case 1: /* request done; response ready to return */
2070                                 resp->done = 2; /* guard against other readers */
2071                                 write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2072                                 return resp;
2073                         case 2: /* response already being returned */
2074                                 break;
2075                         }
2076                 }
2077         }
2078         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2079         return NULL;
2080 }
2081
2082 /* always adds to end of list */
2083 static Sg_request *
2084 sg_add_request(Sg_fd * sfp)
2085 {
2086         int k;
2087         unsigned long iflags;
2088         Sg_request *rp = sfp->req_arr;
2089
2090         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2091         if (!list_empty(&sfp->rq_list)) {
2092                 if (!sfp->cmd_q)
2093                         goto out_unlock;
2094
2095                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2096                         if (!rp->parentfp)
2097                                 break;
2098                 }
2099                 if (k >= SG_MAX_QUEUE)
2100                         goto out_unlock;
2101         }
2102         memset(rp, 0, sizeof (Sg_request));
2103         rp->parentfp = sfp;
2104         rp->header.duration = jiffies_to_msecs(jiffies);
2105         list_add_tail(&rp->entry, &sfp->rq_list);
2106         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2107         return rp;
2108 out_unlock:
2109         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2110         return NULL;
2111 }
2112
2113 /* Return of 1 for found; 0 for not found */
2114 static int
2115 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2116 {
2117         unsigned long iflags;
2118         int res = 0;
2119
2120         if (!sfp || !srp || list_empty(&sfp->rq_list))
2121                 return res;
2122         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2123         if (!list_empty(&srp->entry)) {
2124                 list_del(&srp->entry);
2125                 srp->parentfp = NULL;
2126                 res = 1;
2127         }
2128         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2129
2130         /*
2131          * If the device is detaching, wakeup any readers in case we just
2132          * removed the last response, which would leave nothing for them to
2133          * return other than -ENODEV.
2134          */
2135         if (unlikely(atomic_read(&sfp->parentdp->detaching)))
2136                 wake_up_interruptible_all(&sfp->read_wait);
2137
2138         return res;
2139 }
2140
2141 static Sg_fd *
2142 sg_add_sfp(Sg_device * sdp)
2143 {
2144         Sg_fd *sfp;
2145         unsigned long iflags;
2146         int bufflen;
2147
2148         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2149         if (!sfp)
2150                 return ERR_PTR(-ENOMEM);
2151
2152         init_waitqueue_head(&sfp->read_wait);
2153         rwlock_init(&sfp->rq_list_lock);
2154         INIT_LIST_HEAD(&sfp->rq_list);
2155         kref_init(&sfp->f_ref);
2156         mutex_init(&sfp->f_mutex);
2157         sfp->timeout = SG_DEFAULT_TIMEOUT;
2158         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2159         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2160         sfp->cmd_q = SG_DEF_COMMAND_Q;
2161         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2162         sfp->parentdp = sdp;
2163         write_lock_irqsave(&sdp->sfd_lock, iflags);
2164         if (atomic_read(&sdp->detaching)) {
2165                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2166                 kfree(sfp);
2167                 return ERR_PTR(-ENODEV);
2168         }
2169         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2170         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2171         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2172                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2173         if (unlikely(sg_big_buff != def_reserved_size))
2174                 sg_big_buff = def_reserved_size;
2175
2176         bufflen = min_t(int, sg_big_buff,
2177                         max_sectors_bytes(sdp->device->request_queue));
2178         sg_build_reserve(sfp, bufflen);
2179         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2180                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2181                                       sfp->reserve.bufflen,
2182                                       sfp->reserve.k_use_sg));
2183
2184         kref_get(&sdp->d_ref);
2185         __module_get(THIS_MODULE);
2186         return sfp;
2187 }
2188
2189 static void
2190 sg_remove_sfp_usercontext(struct work_struct *work)
2191 {
2192         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2193         struct sg_device *sdp = sfp->parentdp;
2194         Sg_request *srp;
2195         unsigned long iflags;
2196
2197         /* Cleanup any responses which were never read(). */
2198         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2199         while (!list_empty(&sfp->rq_list)) {
2200                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2201                 sg_finish_rem_req(srp);
2202                 list_del(&srp->entry);
2203                 srp->parentfp = NULL;
2204         }
2205         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2206
2207         if (sfp->reserve.bufflen > 0) {
2208                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2209                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2210                                 (int) sfp->reserve.bufflen,
2211                                 (int) sfp->reserve.k_use_sg));
2212                 sg_remove_scat(sfp, &sfp->reserve);
2213         }
2214
2215         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2216                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2217         kfree(sfp);
2218
2219         scsi_device_put(sdp->device);
2220         kref_put(&sdp->d_ref, sg_device_destroy);
2221         module_put(THIS_MODULE);
2222 }
2223
2224 static void
2225 sg_remove_sfp(struct kref *kref)
2226 {
2227         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2228         struct sg_device *sdp = sfp->parentdp;
2229         unsigned long iflags;
2230
2231         write_lock_irqsave(&sdp->sfd_lock, iflags);
2232         list_del(&sfp->sfd_siblings);
2233         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2234
2235         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2236         schedule_work(&sfp->ew.work);
2237 }
2238
2239 #ifdef CONFIG_SCSI_PROC_FS
2240 static int
2241 sg_idr_max_id(int id, void *p, void *data)
2242 {
2243         int *k = data;
2244
2245         if (*k < id)
2246                 *k = id;
2247
2248         return 0;
2249 }
2250
2251 static int
2252 sg_last_dev(void)
2253 {
2254         int k = -1;
2255         unsigned long iflags;
2256
2257         read_lock_irqsave(&sg_index_lock, iflags);
2258         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2259         read_unlock_irqrestore(&sg_index_lock, iflags);
2260         return k + 1;           /* origin 1 */
2261 }
2262 #endif
2263
2264 /* must be called with sg_index_lock held */
2265 static Sg_device *sg_lookup_dev(int dev)
2266 {
2267         return idr_find(&sg_index_idr, dev);
2268 }
2269
2270 static Sg_device *
2271 sg_get_dev(int dev)
2272 {
2273         struct sg_device *sdp;
2274         unsigned long flags;
2275
2276         read_lock_irqsave(&sg_index_lock, flags);
2277         sdp = sg_lookup_dev(dev);
2278         if (!sdp)
2279                 sdp = ERR_PTR(-ENXIO);
2280         else if (atomic_read(&sdp->detaching)) {
2281                 /* If sdp->detaching, then the refcount may already be 0, in
2282                  * which case it would be a bug to do kref_get().
2283                  */
2284                 sdp = ERR_PTR(-ENODEV);
2285         } else
2286                 kref_get(&sdp->d_ref);
2287         read_unlock_irqrestore(&sg_index_lock, flags);
2288
2289         return sdp;
2290 }
2291
2292 #ifdef CONFIG_SCSI_PROC_FS
2293 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2294
2295 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2296 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2297                                   size_t count, loff_t *off);
2298 static const struct proc_ops adio_proc_ops = {
2299         .proc_open      = sg_proc_single_open_adio,
2300         .proc_read      = seq_read,
2301         .proc_lseek     = seq_lseek,
2302         .proc_write     = sg_proc_write_adio,
2303         .proc_release   = single_release,
2304 };
2305
2306 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2307 static ssize_t sg_proc_write_dressz(struct file *filp,
2308                 const char __user *buffer, size_t count, loff_t *off);
2309 static const struct proc_ops dressz_proc_ops = {
2310         .proc_open      = sg_proc_single_open_dressz,
2311         .proc_read      = seq_read,
2312         .proc_lseek     = seq_lseek,
2313         .proc_write     = sg_proc_write_dressz,
2314         .proc_release   = single_release,
2315 };
2316
2317 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2318 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2319 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2320 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2321 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2322 static void dev_seq_stop(struct seq_file *s, void *v);
2323 static const struct seq_operations dev_seq_ops = {
2324         .start = dev_seq_start,
2325         .next  = dev_seq_next,
2326         .stop  = dev_seq_stop,
2327         .show  = sg_proc_seq_show_dev,
2328 };
2329
2330 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2331 static const struct seq_operations devstrs_seq_ops = {
2332         .start = dev_seq_start,
2333         .next  = dev_seq_next,
2334         .stop  = dev_seq_stop,
2335         .show  = sg_proc_seq_show_devstrs,
2336 };
2337
2338 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2339 static const struct seq_operations debug_seq_ops = {
2340         .start = dev_seq_start,
2341         .next  = dev_seq_next,
2342         .stop  = dev_seq_stop,
2343         .show  = sg_proc_seq_show_debug,
2344 };
2345
2346 static int
2347 sg_proc_init(void)
2348 {
2349         struct proc_dir_entry *p;
2350
2351         p = proc_mkdir("scsi/sg", NULL);
2352         if (!p)
2353                 return 1;
2354
2355         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2356         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2357         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2358         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2359         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2360         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2361         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2362         return 0;
2363 }
2364
2365
2366 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2367 {
2368         seq_printf(s, "%d\n", *((int *)s->private));
2369         return 0;
2370 }
2371
2372 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2373 {
2374         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2375 }
2376
2377 static ssize_t
2378 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2379                    size_t count, loff_t *off)
2380 {
2381         int err;
2382         unsigned long num;
2383
2384         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2385                 return -EACCES;
2386         err = kstrtoul_from_user(buffer, count, 0, &num);
2387         if (err)
2388                 return err;
2389         sg_allow_dio = num ? 1 : 0;
2390         return count;
2391 }
2392
2393 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2394 {
2395         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2396 }
2397
2398 static ssize_t
2399 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2400                      size_t count, loff_t *off)
2401 {
2402         int err;
2403         unsigned long k = ULONG_MAX;
2404
2405         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2406                 return -EACCES;
2407
2408         err = kstrtoul_from_user(buffer, count, 0, &k);
2409         if (err)
2410                 return err;
2411         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2412                 sg_big_buff = k;
2413                 return count;
2414         }
2415         return -ERANGE;
2416 }
2417
2418 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2419 {
2420         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2421                    sg_version_date);
2422         return 0;
2423 }
2424
2425 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2426 {
2427         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2428         return 0;
2429 }
2430
2431 struct sg_proc_deviter {
2432         loff_t  index;
2433         size_t  max;
2434 };
2435
2436 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2437 {
2438         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2439
2440         s->private = it;
2441         if (! it)
2442                 return NULL;
2443
2444         it->index = *pos;
2445         it->max = sg_last_dev();
2446         if (it->index >= it->max)
2447                 return NULL;
2448         return it;
2449 }
2450
2451 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2452 {
2453         struct sg_proc_deviter * it = s->private;
2454
2455         *pos = ++it->index;
2456         return (it->index < it->max) ? it : NULL;
2457 }
2458
2459 static void dev_seq_stop(struct seq_file *s, void *v)
2460 {
2461         kfree(s->private);
2462 }
2463
2464 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2465 {
2466         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2467         Sg_device *sdp;
2468         struct scsi_device *scsidp;
2469         unsigned long iflags;
2470
2471         read_lock_irqsave(&sg_index_lock, iflags);
2472         sdp = it ? sg_lookup_dev(it->index) : NULL;
2473         if ((NULL == sdp) || (NULL == sdp->device) ||
2474             (atomic_read(&sdp->detaching)))
2475                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2476         else {
2477                 scsidp = sdp->device;
2478                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2479                               scsidp->host->host_no, scsidp->channel,
2480                               scsidp->id, scsidp->lun, (int) scsidp->type,
2481                               1,
2482                               (int) scsidp->queue_depth,
2483                               (int) scsi_device_busy(scsidp),
2484                               (int) scsi_device_online(scsidp));
2485         }
2486         read_unlock_irqrestore(&sg_index_lock, iflags);
2487         return 0;
2488 }
2489
2490 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2491 {
2492         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2493         Sg_device *sdp;
2494         struct scsi_device *scsidp;
2495         unsigned long iflags;
2496
2497         read_lock_irqsave(&sg_index_lock, iflags);
2498         sdp = it ? sg_lookup_dev(it->index) : NULL;
2499         scsidp = sdp ? sdp->device : NULL;
2500         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2501                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2502                            scsidp->vendor, scsidp->model, scsidp->rev);
2503         else
2504                 seq_puts(s, "<no active device>\n");
2505         read_unlock_irqrestore(&sg_index_lock, iflags);
2506         return 0;
2507 }
2508
2509 /* must be called while holding sg_index_lock */
2510 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2511 {
2512         int k, new_interface, blen, usg;
2513         Sg_request *srp;
2514         Sg_fd *fp;
2515         const sg_io_hdr_t *hp;
2516         const char * cp;
2517         unsigned int ms;
2518
2519         k = 0;
2520         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2521                 k++;
2522                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2523                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2524                            "(res)sgat=%d low_dma=%d\n", k,
2525                            jiffies_to_msecs(fp->timeout),
2526                            fp->reserve.bufflen,
2527                            (int) fp->reserve.k_use_sg, 0);
2528                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2529                            (int) fp->cmd_q, (int) fp->force_packid,
2530                            (int) fp->keep_orphan);
2531                 list_for_each_entry(srp, &fp->rq_list, entry) {
2532                         hp = &srp->header;
2533                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2534                         if (srp->res_used) {
2535                                 if (new_interface &&
2536                                     (SG_FLAG_MMAP_IO & hp->flags))
2537                                         cp = "     mmap>> ";
2538                                 else
2539                                         cp = "     rb>> ";
2540                         } else {
2541                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2542                                         cp = "     dio>> ";
2543                                 else
2544                                         cp = "     ";
2545                         }
2546                         seq_puts(s, cp);
2547                         blen = srp->data.bufflen;
2548                         usg = srp->data.k_use_sg;
2549                         seq_puts(s, srp->done ?
2550                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2551                                   : "act:");
2552                         seq_printf(s, " id=%d blen=%d",
2553                                    srp->header.pack_id, blen);
2554                         if (srp->done)
2555                                 seq_printf(s, " dur=%d", hp->duration);
2556                         else {
2557                                 ms = jiffies_to_msecs(jiffies);
2558                                 seq_printf(s, " t_o/elap=%d/%d",
2559                                         (new_interface ? hp->timeout :
2560                                                   jiffies_to_msecs(fp->timeout)),
2561                                         (ms > hp->duration ? ms - hp->duration : 0));
2562                         }
2563                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2564                                    (int) srp->data.cmd_opcode);
2565                 }
2566                 if (list_empty(&fp->rq_list))
2567                         seq_puts(s, "     No requests active\n");
2568                 read_unlock(&fp->rq_list_lock);
2569         }
2570 }
2571
2572 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2573 {
2574         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2575         Sg_device *sdp;
2576         unsigned long iflags;
2577
2578         if (it && (0 == it->index))
2579                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2580                            (int)it->max, sg_big_buff);
2581
2582         read_lock_irqsave(&sg_index_lock, iflags);
2583         sdp = it ? sg_lookup_dev(it->index) : NULL;
2584         if (NULL == sdp)
2585                 goto skip;
2586         read_lock(&sdp->sfd_lock);
2587         if (!list_empty(&sdp->sfds)) {
2588                 seq_printf(s, " >>> device=%s ", sdp->name);
2589                 if (atomic_read(&sdp->detaching))
2590                         seq_puts(s, "detaching pending close ");
2591                 else if (sdp->device) {
2592                         struct scsi_device *scsidp = sdp->device;
2593
2594                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2595                                    scsidp->host->host_no,
2596                                    scsidp->channel, scsidp->id,
2597                                    scsidp->lun,
2598                                    scsidp->host->hostt->emulated);
2599                 }
2600                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2601                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2602                 sg_proc_debug_helper(s, sdp);
2603         }
2604         read_unlock(&sdp->sfd_lock);
2605 skip:
2606         read_unlock_irqrestore(&sg_index_lock, iflags);
2607         return 0;
2608 }
2609
2610 #endif                          /* CONFIG_SCSI_PROC_FS */
2611
2612 module_init(init_sg);
2613 module_exit(exit_sg);