drivers/scsi/sg.c

   1 /*
   2  *  History:
   3  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   4  *           to allow user process control of SCSI devices.
   5  *  Development Sponsored by Killy Corp. NY NY
   6  *
   7  * Original driver (sg.c):
   8  *        Copyright (C) 1992 Lawrence Foard
   9  * Version 2 and 3 extensions to driver:
  10  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License as published by
  14  * the Free Software Foundation; either version 2, or (at your option)
  15  * any later version.
  16  *
  17  */
  18
  19 static int sg_version_num = 30536;      /* 2 digits for each component */
  20 #define SG_VERSION_STR "3.5.36"
  21
  22 /*
  23  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  24  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  25  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  26  *        (otherwise the macros compile to empty statements).
  27  *
  28  */
  29 #include <linux/module.h>
  30
  31 #include <linux/fs.h>
  32 #include <linux/kernel.h>
  33 #include <linux/sched.h>
  34 #include <linux/string.h>
  35 #include <linux/mm.h>
  36 #include <linux/errno.h>
  37 #include <linux/mtio.h>
  38 #include <linux/ioctl.h>
  39 #include <linux/slab.h>
  40 #include <linux/fcntl.h>
  41 #include <linux/init.h>
  42 #include <linux/poll.h>
  43 #include <linux/moduleparam.h>
  44 #include <linux/cdev.h>
  45 #include <linux/idr.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/delay.h>
  49 #include <linux/blktrace_api.h>
  50 #include <linux/mutex.h>
  51 #include <linux/atomic.h>
  52 #include <linux/ratelimit.h>
  53 #include <linux/uio.h>
  54 #include <linux/cred.h> /* for sg_check_file_access() */
  55
  56 #include "scsi.h"
  57 #include <scsi/scsi_dbg.h>
  58 #include <scsi/scsi_host.h>
  59 #include <scsi/scsi_driver.h>
  60 #include <scsi/scsi_ioctl.h>
  61 #include <scsi/sg.h>
  62
  63 #include "scsi_logging.h"
  64
  65 #ifdef CONFIG_SCSI_PROC_FS
  66 #include <linux/proc_fs.h>
  67 static char *sg_version_date = "20140603";
  68
  69 static int sg_proc_init(void);
  70 static void sg_proc_cleanup(void);
  71 #endif
  72
  73 #define SG_ALLOW_DIO_DEF 0
  74
  75 #define SG_MAX_DEVS 32768
  76
  77 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  78  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  79  * than 16 bytes are "variable length" whose length is a multiple of 4
  80  */
  81 #define SG_MAX_CDB_SIZE 252
  82
  83 /*
  84  * Suppose you want to calculate the formula muldiv(x,m,d)=int(x * m / d)
  85  * Then when using 32 bit integers x * m may overflow during the calculation.
  86  * Replacing muldiv(x) by muldiv(x)=((x % d) * m) / d + int(x / d) * m
  87  * calculates the same, but prevents the overflow when both m and d
  88  * are "small" numbers (like HZ and USER_HZ).
  89  * Of course an overflow is inavoidable if the result of muldiv doesn't fit
  90  * in 32 bits.
  91  */
  92 #define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
  93
  94 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  95
  96 int sg_big_buff = SG_DEF_RESERVED_SIZE;
  97 /* N.B. This variable is readable and writeable via
  98    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  99    of this size (or less if there is not enough memory) will be reserved
 100    for use by this file descriptor. [Deprecated usage: this variable is also
 101    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
 102    the kernel (i.e. it is not a module).] */
 103 static int def_reserved_size = -1;      /* picks up init parameter */
 104 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
 105
 106 static int scatter_elem_sz = SG_SCATTER_SZ;
 107 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
 108
 109 #define SG_SECTOR_SZ 512
 110
 111 static int sg_add_device(struct device *, struct class_interface *);
 112 static void sg_remove_device(struct device *, struct class_interface *);
 113
 114 static DEFINE_IDR(sg_index_idr);
 115 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
 116                                                            file descriptor list for device */
 117
 118 static struct class_interface sg_interface = {
 119         .add_dev        = sg_add_device,
 120         .remove_dev     = sg_remove_device,
 121 };
 122
 123 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 124         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 125         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 126         unsigned bufflen;       /* Size of (aggregate) data buffer */
 127         struct page **pages;
 128         int page_order;
 129         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 130         unsigned char cmd_opcode; /* first byte of command */
 131 } Sg_scatter_hold;
 132
 133 struct sg_device;               /* forward declarations */
 134 struct sg_fd;
 135
 136 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 137         struct list_head entry; /* list entry */
 138         struct sg_fd *parentfp; /* NULL -> not in use */
 139         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 140         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 141         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 142         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 143         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 144         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 145         /* done protected by rq_list_lock */
 146         char done;              /* 0->before bh, 1->before read, 2->read */
 147         struct request *rq;
 148         struct bio *bio;
 149         struct execute_work ew;
 150 } Sg_request;
 151
 152 typedef struct sg_fd {          /* holds the state of a file descriptor */
 153         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 154         struct sg_device *parentdp;     /* owning device */
 155         wait_queue_head_t read_wait;    /* queue read until command done */
 156         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 157         struct mutex f_mutex;   /* protect against changes in this fd */
 158         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 159         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 160         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 161         struct list_head rq_list; /* head of request list */
 162         struct fasync_struct *async_qp; /* used by asynchronous notification */
 163         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 164         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 165         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 166         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 167         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 168         char mmap_called;       /* 0 -> mmap() never called on this fd */
 169         char res_in_use;        /* 1 -> 'reserve' array in use */
 170         struct kref f_ref;
 171         struct execute_work ew;
 172 } Sg_fd;
 173
 174 typedef struct sg_device { /* holds the state of each scsi generic device */
 175         struct scsi_device *device;
 176         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 177         struct mutex open_rel_lock;     /* held when in open() or release() */
 178         int sg_tablesize;       /* adapter's max scatter-gather table size */
 179         u32 index;              /* device index number */
 180         struct list_head sfds;
 181         rwlock_t sfd_lock;      /* protect access to sfd list */
 182         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 183         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 184         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 185         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 186         struct gendisk *disk;
 187         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 188         struct kref d_ref;
 189 } Sg_device;
 190
 191 /* tasklet or soft irq callback */
 192 static void sg_rq_end_io(struct request *rq, int uptodate);
 193 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 194 static int sg_finish_rem_req(Sg_request * srp);
 195 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 196 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 197                            Sg_request * srp);
 198 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 199                         const char __user *buf, size_t count, int blocking,
 200                         int read_only, int sg_io_owned, Sg_request **o_srp);
 201 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 202                            unsigned char *cmnd, int timeout, int blocking);
 203 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 204 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 205 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 206 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 207 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 208 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 209 static void sg_remove_sfp(struct kref *);
 210 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 211 static Sg_request *sg_add_request(Sg_fd * sfp);
 212 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 213 static Sg_device *sg_get_dev(int dev);
 214 static void sg_device_destroy(struct kref *kref);
 215
 216 #define SZ_SG_HEADER sizeof(struct sg_header)
 217 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 218 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 219 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 220
 221 #define sg_printk(prefix, sdp, fmt, a...) \
 222         sdev_prefix_printk(prefix, (sdp)->device,               \
 223                            (sdp)->disk->disk_name, fmt, ##a)
 224
 225 /*
 226  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 227  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 228  * to trigger read() and write() calls from various contexts with elevated
 229  * privileges. This can lead to kernel memory corruption (e.g. if these
 230  * interfaces are called through splice()) and privilege escalation inside
 231  * userspace (e.g. if a process with access to such a device passes a file
 232  * descriptor to a SUID binary as stdin/stdout/stderr).
 233  *
 234  * This function provides protection for the legacy API by restricting the
 235  * calling context.
 236  */
 237 static int sg_check_file_access(struct file *filp, const char *caller)
 238 {
 239         if (filp->f_cred != current_real_cred()) {
 240                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 241                         caller, task_tgid_vnr(current), current->comm);
 242                 return -EPERM;
 243         }
 244         if (unlikely(segment_eq(get_fs(), KERNEL_DS))) {
 245                 pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
 246                         caller, task_tgid_vnr(current), current->comm);
 247                 return -EACCES;
 248         }
 249         return 0;
 250 }
 251
 252 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 253 {
 254         struct sg_fd *sfp = filp->private_data;
 255
 256         if (sfp->parentdp->device->type == TYPE_SCANNER)
 257                 return 0;
 258
 259         return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
 260 }
 261
 262 static int
 263 open_wait(Sg_device *sdp, int flags)
 264 {
 265         int retval = 0;
 266
 267         if (flags & O_EXCL) {
 268                 while (sdp->open_cnt > 0) {
 269                         mutex_unlock(&sdp->open_rel_lock);
 270                         retval = wait_event_interruptible(sdp->open_wait,
 271                                         (atomic_read(&sdp->detaching) ||
 272                                          !sdp->open_cnt));
 273                         mutex_lock(&sdp->open_rel_lock);
 274
 275                         if (retval) /* -ERESTARTSYS */
 276                                 return retval;
 277                         if (atomic_read(&sdp->detaching))
 278                                 return -ENODEV;
 279                 }
 280         } else {
 281                 while (sdp->exclude) {
 282                         mutex_unlock(&sdp->open_rel_lock);
 283                         retval = wait_event_interruptible(sdp->open_wait,
 284                                         (atomic_read(&sdp->detaching) ||
 285                                          !sdp->exclude));
 286                         mutex_lock(&sdp->open_rel_lock);
 287
 288                         if (retval) /* -ERESTARTSYS */
 289                                 return retval;
 290                         if (atomic_read(&sdp->detaching))
 291                                 return -ENODEV;
 292                 }
 293         }
 294
 295         return retval;
 296 }
 297
 298 /* Returns 0 on success, else a negated errno value */
 299 static int
 300 sg_open(struct inode *inode, struct file *filp)
 301 {
 302         int dev = iminor(inode);
 303         int flags = filp->f_flags;
 304         struct request_queue *q;
 305         Sg_device *sdp;
 306         Sg_fd *sfp;
 307         int retval;
 308
 309         nonseekable_open(inode, filp);
 310         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 311                 return -EPERM; /* Can't lock it with read only access */
 312         sdp = sg_get_dev(dev);
 313         if (IS_ERR(sdp))
 314                 return PTR_ERR(sdp);
 315
 316         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 317                                       "sg_open: flags=0x%x\n", flags));
 318
 319         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 320         /* Prevent the device driver from vanishing while we sleep */
 321         retval = scsi_device_get(sdp->device);
 322         if (retval)
 323                 goto sg_put;
 324
 325         retval = scsi_autopm_get_device(sdp->device);
 326         if (retval)
 327                 goto sdp_put;
 328
 329         /* scsi_block_when_processing_errors() may block so bypass
 330          * check if O_NONBLOCK. Permits SCSI commands to be issued
 331          * during error recovery. Tread carefully. */
 332         if (!((flags & O_NONBLOCK) ||
 333               scsi_block_when_processing_errors(sdp->device))) {
 334                 retval = -ENXIO;
 335                 /* we are in error recovery for this device */
 336                 goto error_out;
 337         }
 338
 339         mutex_lock(&sdp->open_rel_lock);
 340         if (flags & O_NONBLOCK) {
 341                 if (flags & O_EXCL) {
 342                         if (sdp->open_cnt > 0) {
 343                                 retval = -EBUSY;
 344                                 goto error_mutex_locked;
 345                         }
 346                 } else {
 347                         if (sdp->exclude) {
 348                                 retval = -EBUSY;
 349                                 goto error_mutex_locked;
 350                         }
 351                 }
 352         } else {
 353                 retval = open_wait(sdp, flags);
 354                 if (retval) /* -ERESTARTSYS or -ENODEV */
 355                         goto error_mutex_locked;
 356         }
 357
 358         /* N.B. at this point we are holding the open_rel_lock */
 359         if (flags & O_EXCL)
 360                 sdp->exclude = true;
 361
 362         if (sdp->open_cnt < 1) {  /* no existing opens */
 363                 sdp->sgdebug = 0;
 364                 q = sdp->device->request_queue;
 365                 sdp->sg_tablesize = queue_max_segments(q);
 366         }
 367         sfp = sg_add_sfp(sdp);
 368         if (IS_ERR(sfp)) {
 369                 retval = PTR_ERR(sfp);
 370                 goto out_undo;
 371         }
 372
 373         filp->private_data = sfp;
 374         sdp->open_cnt++;
 375         mutex_unlock(&sdp->open_rel_lock);
 376
 377         retval = 0;
 378 sg_put:
 379         kref_put(&sdp->d_ref, sg_device_destroy);
 380         return retval;
 381
 382 out_undo:
 383         if (flags & O_EXCL) {
 384                 sdp->exclude = false;   /* undo if error */
 385                 wake_up_interruptible(&sdp->open_wait);
 386         }
 387 error_mutex_locked:
 388         mutex_unlock(&sdp->open_rel_lock);
 389 error_out:
 390         scsi_autopm_put_device(sdp->device);
 391 sdp_put:
 392         scsi_device_put(sdp->device);
 393         goto sg_put;
 394 }
 395
 396 /* Release resources associated with a successful sg_open()
 397  * Returns 0 on success, else a negated errno value */
 398 static int
 399 sg_release(struct inode *inode, struct file *filp)
 400 {
 401         Sg_device *sdp;
 402         Sg_fd *sfp;
 403
 404         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 405                 return -ENXIO;
 406         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 407
 408         mutex_lock(&sdp->open_rel_lock);
 409         scsi_autopm_put_device(sdp->device);
 410         kref_put(&sfp->f_ref, sg_remove_sfp);
 411         sdp->open_cnt--;
 412
 413         /* possibly many open()s waiting on exlude clearing, start many;
 414          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 415         if (sdp->exclude) {
 416                 sdp->exclude = false;
 417                 wake_up_interruptible_all(&sdp->open_wait);
 418         } else if (0 == sdp->open_cnt) {
 419                 wake_up_interruptible(&sdp->open_wait);
 420         }
 421         mutex_unlock(&sdp->open_rel_lock);
 422         return 0;
 423 }
 424
 425 static ssize_t
 426 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 427 {
 428         Sg_device *sdp;
 429         Sg_fd *sfp;
 430         Sg_request *srp;
 431         int req_pack_id = -1;
 432         sg_io_hdr_t *hp;
 433         struct sg_header *old_hdr = NULL;
 434         int retval = 0;
 435
 436         /*
 437          * This could cause a response to be stranded. Close the associated
 438          * file descriptor to free up any resources being held.
 439          */
 440         retval = sg_check_file_access(filp, __func__);
 441         if (retval)
 442                 return retval;
 443
 444         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 445                 return -ENXIO;
 446         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 447                                       "sg_read: count=%d\n", (int) count));
 448
 449         if (!access_ok(VERIFY_WRITE, buf, count))
 450                 return -EFAULT;
 451         if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
 452                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 453                 if (!old_hdr)
 454                         return -ENOMEM;
 455                 if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
 456                         retval = -EFAULT;
 457                         goto free_old_hdr;
 458                 }
 459                 if (old_hdr->reply_len < 0) {
 460                         if (count >= SZ_SG_IO_HDR) {
 461                                 sg_io_hdr_t *new_hdr;
 462                                 new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
 463                                 if (!new_hdr) {
 464                                         retval = -ENOMEM;
 465                                         goto free_old_hdr;
 466                                 }
 467                                 retval =__copy_from_user
 468                                     (new_hdr, buf, SZ_SG_IO_HDR);
 469                                 req_pack_id = new_hdr->pack_id;
 470                                 kfree(new_hdr);
 471                                 if (retval) {
 472                                         retval = -EFAULT;
 473                                         goto free_old_hdr;
 474                                 }
 475                         }
 476                 } else
 477                         req_pack_id = old_hdr->pack_id;
 478         }
 479         srp = sg_get_rq_mark(sfp, req_pack_id);
 480         if (!srp) {             /* now wait on packet to arrive */
 481                 if (atomic_read(&sdp->detaching)) {
 482                         retval = -ENODEV;
 483                         goto free_old_hdr;
 484                 }
 485                 if (filp->f_flags & O_NONBLOCK) {
 486                         retval = -EAGAIN;
 487                         goto free_old_hdr;
 488                 }
 489                 retval = wait_event_interruptible(sfp->read_wait,
 490                         (atomic_read(&sdp->detaching) ||
 491                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 492                 if (atomic_read(&sdp->detaching)) {
 493                         retval = -ENODEV;
 494                         goto free_old_hdr;
 495                 }
 496                 if (retval) {
 497                         /* -ERESTARTSYS as signal hit process */
 498                         goto free_old_hdr;
 499                 }
 500         }
 501         if (srp->header.interface_id != '\0') {
 502                 retval = sg_new_read(sfp, buf, count, srp);
 503                 goto free_old_hdr;
 504         }
 505
 506         hp = &srp->header;
 507         if (old_hdr == NULL) {
 508                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 509                 if (! old_hdr) {
 510                         retval = -ENOMEM;
 511                         goto free_old_hdr;
 512                 }
 513         }
 514         memset(old_hdr, 0, SZ_SG_HEADER);
 515         old_hdr->reply_len = (int) hp->timeout;
 516         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 517         old_hdr->pack_id = hp->pack_id;
 518         old_hdr->twelve_byte =
 519             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 520         old_hdr->target_status = hp->masked_status;
 521         old_hdr->host_status = hp->host_status;
 522         old_hdr->driver_status = hp->driver_status;
 523         if ((CHECK_CONDITION & hp->masked_status) ||
 524             (DRIVER_SENSE & hp->driver_status))
 525                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 526                        sizeof (old_hdr->sense_buffer));
 527         switch (hp->host_status) {
 528         /* This setup of 'result' is for backward compatibility and is best
 529            ignored by the user who should use target, host + driver status */
 530         case DID_OK:
 531         case DID_PASSTHROUGH:
 532         case DID_SOFT_ERROR:
 533                 old_hdr->result = 0;
 534                 break;
 535         case DID_NO_CONNECT:
 536         case DID_BUS_BUSY:
 537         case DID_TIME_OUT:
 538                 old_hdr->result = EBUSY;
 539                 break;
 540         case DID_BAD_TARGET:
 541         case DID_ABORT:
 542         case DID_PARITY:
 543         case DID_RESET:
 544         case DID_BAD_INTR:
 545                 old_hdr->result = EIO;
 546                 break;
 547         case DID_ERROR:
 548                 old_hdr->result = (srp->sense_b[0] == 0 &&
 549                                   hp->masked_status == GOOD) ? 0 : EIO;
 550                 break;
 551         default:
 552                 old_hdr->result = EIO;
 553                 break;
 554         }
 555
 556         /* Now copy the result back to the user buffer.  */
 557         if (count >= SZ_SG_HEADER) {
 558                 if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 559                         retval = -EFAULT;
 560                         goto free_old_hdr;
 561                 }
 562                 buf += SZ_SG_HEADER;
 563                 if (count > old_hdr->reply_len)
 564                         count = old_hdr->reply_len;
 565                 if (count > SZ_SG_HEADER) {
 566                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 567                                 retval = -EFAULT;
 568                                 goto free_old_hdr;
 569                         }
 570                 }
 571         } else
 572                 count = (old_hdr->result == 0) ? 0 : -EIO;
 573         sg_finish_rem_req(srp);
 574         sg_remove_request(sfp, srp);
 575         retval = count;
 576 free_old_hdr:
 577         kfree(old_hdr);
 578         return retval;
 579 }
 580
 581 static ssize_t
 582 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 583 {
 584         sg_io_hdr_t *hp = &srp->header;
 585         int err = 0, err2;
 586         int len;
 587
 588         if (count < SZ_SG_IO_HDR) {
 589                 err = -EINVAL;
 590                 goto err_out;
 591         }
 592         hp->sb_len_wr = 0;
 593         if ((hp->mx_sb_len > 0) && hp->sbp) {
 594                 if ((CHECK_CONDITION & hp->masked_status) ||
 595                     (DRIVER_SENSE & hp->driver_status)) {
 596                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 597                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 598                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 599                         len = (len > sb_len) ? sb_len : len;
 600                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 601                                 err = -EFAULT;
 602                                 goto err_out;
 603                         }
 604                         hp->sb_len_wr = len;
 605                 }
 606         }
 607         if (hp->masked_status || hp->host_status || hp->driver_status)
 608                 hp->info |= SG_INFO_CHECK;
 609         if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
 610                 err = -EFAULT;
 611                 goto err_out;
 612         }
 613 err_out:
 614         err2 = sg_finish_rem_req(srp);
 615         sg_remove_request(sfp, srp);
 616         return err ? : err2 ? : count;
 617 }
 618
 619 static ssize_t
 620 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 621 {
 622         int mxsize, cmd_size, k;
 623         int input_size, blocking;
 624         unsigned char opcode;
 625         Sg_device *sdp;
 626         Sg_fd *sfp;
 627         Sg_request *srp;
 628         struct sg_header old_hdr;
 629         sg_io_hdr_t *hp;
 630         unsigned char cmnd[SG_MAX_CDB_SIZE];
 631         int retval;
 632
 633         retval = sg_check_file_access(filp, __func__);
 634         if (retval)
 635                 return retval;
 636
 637         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 638                 return -ENXIO;
 639         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 640                                       "sg_write: count=%d\n", (int) count));
 641         if (atomic_read(&sdp->detaching))
 642                 return -ENODEV;
 643         if (!((filp->f_flags & O_NONBLOCK) ||
 644               scsi_block_when_processing_errors(sdp->device)))
 645                 return -ENXIO;
 646
 647         if (!access_ok(VERIFY_READ, buf, count))
 648                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 649         if (count < SZ_SG_HEADER)
 650                 return -EIO;
 651         if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 652                 return -EFAULT;
 653         blocking = !(filp->f_flags & O_NONBLOCK);
 654         if (old_hdr.reply_len < 0)
 655                 return sg_new_write(sfp, filp, buf, count,
 656                                     blocking, 0, 0, NULL);
 657         if (count < (SZ_SG_HEADER + 6))
 658                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 659
 660         if (!(srp = sg_add_request(sfp))) {
 661                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 662                                               "sg_write: queue full\n"));
 663                 return -EDOM;
 664         }
 665         buf += SZ_SG_HEADER;
 666         __get_user(opcode, buf);
 667         mutex_lock(&sfp->f_mutex);
 668         if (sfp->next_cmd_len > 0) {
 669                 cmd_size = sfp->next_cmd_len;
 670                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 671         } else {
 672                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 673                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 674                         cmd_size = 12;
 675         }
 676         mutex_unlock(&sfp->f_mutex);
 677         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 678                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 679 /* Determine buffer size.  */
 680         input_size = count - cmd_size;
 681         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 682         mxsize -= SZ_SG_HEADER;
 683         input_size -= SZ_SG_HEADER;
 684         if (input_size < 0) {
 685                 sg_remove_request(sfp, srp);
 686                 return -EIO;    /* User did not pass enough bytes for this command. */
 687         }
 688         hp = &srp->header;
 689         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 690         hp->cmd_len = (unsigned char) cmd_size;
 691         hp->iovec_count = 0;
 692         hp->mx_sb_len = 0;
 693         if (input_size > 0)
 694                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 695                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 696         else
 697                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 698         hp->dxfer_len = mxsize;
 699         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 700             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 701                 hp->dxferp = (char __user *)buf + cmd_size;
 702         else
 703                 hp->dxferp = NULL;
 704         hp->sbp = NULL;
 705         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 706         hp->flags = input_size; /* structure abuse ... */
 707         hp->pack_id = old_hdr.pack_id;
 708         hp->usr_ptr = NULL;
 709         if (__copy_from_user(cmnd, buf, cmd_size)) {
 710                 sg_remove_request(sfp, srp);
 711                 return -EFAULT;
 712         }
 713         /*
 714          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 715          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 716          * is a non-zero input_size, so emit a warning.
 717          */
 718         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 719                 printk_ratelimited(KERN_WARNING
 720                                    "sg_write: data in/out %d/%d bytes "
 721                                    "for SCSI command 0x%x-- guessing "
 722                                    "data in;\n   program %s not setting "
 723                                    "count and/or reply_len properly\n",
 724                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 725                                    input_size, (unsigned int) cmnd[0],
 726                                    current->comm);
 727         }
 728         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 729         return (k < 0) ? k : count;
 730 }
 731
 732 static ssize_t
 733 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 734                  size_t count, int blocking, int read_only, int sg_io_owned,
 735                  Sg_request **o_srp)
 736 {
 737         int k;
 738         Sg_request *srp;
 739         sg_io_hdr_t *hp;
 740         unsigned char cmnd[SG_MAX_CDB_SIZE];
 741         int timeout;
 742         unsigned long ul_timeout;
 743
 744         if (count < SZ_SG_IO_HDR)
 745                 return -EINVAL;
 746         if (!access_ok(VERIFY_READ, buf, count))
 747                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 748
 749         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 750         if (!(srp = sg_add_request(sfp))) {
 751                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 752                                               "sg_new_write: queue full\n"));
 753                 return -EDOM;
 754         }
 755         srp->sg_io_owned = sg_io_owned;
 756         hp = &srp->header;
 757         if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
 758                 sg_remove_request(sfp, srp);
 759                 return -EFAULT;
 760         }
 761         if (hp->interface_id != 'S') {
 762                 sg_remove_request(sfp, srp);
 763                 return -ENOSYS;
 764         }
 765         if (hp->flags & SG_FLAG_MMAP_IO) {
 766                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 767                         sg_remove_request(sfp, srp);
 768                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 769                 }
 770                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 771                         sg_remove_request(sfp, srp);
 772                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 773                 }
 774                 if (sfp->res_in_use) {
 775                         sg_remove_request(sfp, srp);
 776                         return -EBUSY;  /* reserve buffer already being used */
 777                 }
 778         }
 779         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 780         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 781         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 782                 sg_remove_request(sfp, srp);
 783                 return -EMSGSIZE;
 784         }
 785         if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
 786                 sg_remove_request(sfp, srp);
 787                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 788         }
 789         if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 790                 sg_remove_request(sfp, srp);
 791                 return -EFAULT;
 792         }
 793         if (read_only && sg_allow_access(file, cmnd)) {
 794                 sg_remove_request(sfp, srp);
 795                 return -EPERM;
 796         }
 797         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 798         if (k < 0)
 799                 return k;
 800         if (o_srp)
 801                 *o_srp = srp;
 802         return count;
 803 }
 804
 805 static int
 806 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 807                 unsigned char *cmnd, int timeout, int blocking)
 808 {
 809         int k, at_head;
 810         Sg_device *sdp = sfp->parentdp;
 811         sg_io_hdr_t *hp = &srp->header;
 812
 813         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 814         hp->status = 0;
 815         hp->masked_status = 0;
 816         hp->msg_status = 0;
 817         hp->info = 0;
 818         hp->host_status = 0;
 819         hp->driver_status = 0;
 820         hp->resid = 0;
 821         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 822                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 823                         (int) cmnd[0], (int) hp->cmd_len));
 824
 825         if (hp->dxfer_len >= SZ_256M) {
 826                 sg_remove_request(sfp, srp);
 827                 return -EINVAL;
 828         }
 829
 830         k = sg_start_req(srp, cmnd);
 831         if (k) {
 832                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 833                         "sg_common_write: start_req err=%d\n", k));
 834                 sg_finish_rem_req(srp);
 835                 sg_remove_request(sfp, srp);
 836                 return k;       /* probably out of space --> ENOMEM */
 837         }
 838         if (atomic_read(&sdp->detaching)) {
 839                 if (srp->bio) {
 840                         if (srp->rq->cmd != srp->rq->__cmd)
 841                                 kfree(srp->rq->cmd);
 842
 843                         blk_end_request_all(srp->rq, -EIO);
 844                         srp->rq = NULL;
 845                 }
 846
 847                 sg_finish_rem_req(srp);
 848                 sg_remove_request(sfp, srp);
 849                 return -ENODEV;
 850         }
 851
 852         hp->duration = jiffies_to_msecs(jiffies);
 853         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 854             (SG_FLAG_Q_AT_TAIL & hp->flags))
 855                 at_head = 0;
 856         else
 857                 at_head = 1;
 858
 859         srp->rq->timeout = timeout;
 860         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 861         blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
 862                               srp->rq, at_head, sg_rq_end_io);
 863         return 0;
 864 }
 865
 866 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 867 {
 868         unsigned long flags;
 869         int ret;
 870
 871         read_lock_irqsave(&sfp->rq_list_lock, flags);
 872         ret = srp->done;
 873         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 874         return ret;
 875 }
 876
 877 static int max_sectors_bytes(struct request_queue *q)
 878 {
 879         unsigned int max_sectors = queue_max_sectors(q);
 880
 881         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 882
 883         return max_sectors << 9;
 884 }
 885
 886 static void
 887 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 888 {
 889         Sg_request *srp;
 890         int val;
 891         unsigned int ms;
 892
 893         val = 0;
 894         list_for_each_entry(srp, &sfp->rq_list, entry) {
 895                 if (val >= SG_MAX_QUEUE)
 896                         break;
 897                 rinfo[val].req_state = srp->done + 1;
 898                 rinfo[val].problem =
 899                         srp->header.masked_status &
 900                         srp->header.host_status &
 901                         srp->header.driver_status;
 902                 if (srp->done)
 903                         rinfo[val].duration =
 904                                 srp->header.duration;
 905                 else {
 906                         ms = jiffies_to_msecs(jiffies);
 907                         rinfo[val].duration =
 908                                 (ms > srp->header.duration) ?
 909                                 (ms - srp->header.duration) : 0;
 910                 }
 911                 rinfo[val].orphan = srp->orphan;
 912                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 913                 rinfo[val].pack_id = srp->header.pack_id;
 914                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 915                 val++;
 916         }
 917 }
 918
 919 static long
 920 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
 921 {
 922         void __user *p = (void __user *)arg;
 923         int __user *ip = p;
 924         int result, val, read_only;
 925         Sg_device *sdp;
 926         Sg_fd *sfp;
 927         Sg_request *srp;
 928         unsigned long iflags;
 929
 930         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 931                 return -ENXIO;
 932
 933         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 934                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 935         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 936
 937         switch (cmd_in) {
 938         case SG_IO:
 939                 if (atomic_read(&sdp->detaching))
 940                         return -ENODEV;
 941                 if (!scsi_block_when_processing_errors(sdp->device))
 942                         return -ENXIO;
 943                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
 944                         return -EFAULT;
 945                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 946                                  1, read_only, 1, &srp);
 947                 if (result < 0)
 948                         return result;
 949                 result = wait_event_interruptible(sfp->read_wait,
 950                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 951                 if (atomic_read(&sdp->detaching))
 952                         return -ENODEV;
 953                 write_lock_irq(&sfp->rq_list_lock);
 954                 if (srp->done) {
 955                         srp->done = 2;
 956                         write_unlock_irq(&sfp->rq_list_lock);
 957                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 958                         return (result < 0) ? result : 0;
 959                 }
 960                 srp->orphan = 1;
 961                 write_unlock_irq(&sfp->rq_list_lock);
 962                 return result;  /* -ERESTARTSYS because signal hit process */
 963         case SG_SET_TIMEOUT:
 964                 result = get_user(val, ip);
 965                 if (result)
 966                         return result;
 967                 if (val < 0)
 968                         return -EIO;
 969                 if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
 970                     val = MULDIV (INT_MAX, USER_HZ, HZ);
 971                 sfp->timeout_user = val;
 972                 sfp->timeout = MULDIV (val, HZ, USER_HZ);
 973
 974                 return 0;
 975         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 976                                 /* strange ..., for backward compatibility */
 977                 return sfp->timeout_user;
 978         case SG_SET_FORCE_LOW_DMA:
 979                 /*
 980                  * N.B. This ioctl never worked properly, but failed to
 981                  * return an error value. So returning '0' to keep compability
 982                  * with legacy applications.
 983                  */
 984                 return 0;
 985         case SG_GET_LOW_DMA:
 986                 return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
 987         case SG_GET_SCSI_ID:
 988                 if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
 989                         return -EFAULT;
 990                 else {
 991                         sg_scsi_id_t __user *sg_idp = p;
 992
 993                         if (atomic_read(&sdp->detaching))
 994                                 return -ENODEV;
 995                         __put_user((int) sdp->device->host->host_no,
 996                                    &sg_idp->host_no);
 997                         __put_user((int) sdp->device->channel,
 998                                    &sg_idp->channel);
 999                         __put_user((int) sdp->device->id, &sg_idp->scsi_id);
1000                         __put_user((int) sdp->device->lun, &sg_idp->lun);
1001                         __put_user((int) sdp->device->type, &sg_idp->scsi_type);
1002                         __put_user((short) sdp->device->host->cmd_per_lun,
1003                                    &sg_idp->h_cmd_per_lun);
1004                         __put_user((short) sdp->device->queue_depth,
1005                                    &sg_idp->d_queue_depth);
1006                         __put_user(0, &sg_idp->unused[0]);
1007                         __put_user(0, &sg_idp->unused[1]);
1008                         return 0;
1009                 }
1010         case SG_SET_FORCE_PACK_ID:
1011                 result = get_user(val, ip);
1012                 if (result)
1013                         return result;
1014                 sfp->force_packid = val ? 1 : 0;
1015                 return 0;
1016         case SG_GET_PACK_ID:
1017                 if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
1018                         return -EFAULT;
1019                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1020                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1021                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1022                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1023                                                        iflags);
1024                                 __put_user(srp->header.pack_id, ip);
1025                                 return 0;
1026                         }
1027                 }
1028                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1029                 __put_user(-1, ip);
1030                 return 0;
1031         case SG_GET_NUM_WAITING:
1032                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1033                 val = 0;
1034                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1035                         if ((1 == srp->done) && (!srp->sg_io_owned))
1036                                 ++val;
1037                 }
1038                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1039                 return put_user(val, ip);
1040         case SG_GET_SG_TABLESIZE:
1041                 return put_user(sdp->sg_tablesize, ip);
1042         case SG_SET_RESERVED_SIZE:
1043                 result = get_user(val, ip);
1044                 if (result)
1045                         return result;
1046                 if (val < 0)
1047                         return -EINVAL;
1048                 val = min_t(int, val,
1049                             max_sectors_bytes(sdp->device->request_queue));
1050                 mutex_lock(&sfp->f_mutex);
1051                 if (val != sfp->reserve.bufflen) {
1052                         if (sfp->mmap_called ||
1053                             sfp->res_in_use) {
1054                                 mutex_unlock(&sfp->f_mutex);
1055                                 return -EBUSY;
1056                         }
1057
1058                         sg_remove_scat(sfp, &sfp->reserve);
1059                         sg_build_reserve(sfp, val);
1060                 }
1061                 mutex_unlock(&sfp->f_mutex);
1062                 return 0;
1063         case SG_GET_RESERVED_SIZE:
1064                 val = min_t(int, sfp->reserve.bufflen,
1065                             max_sectors_bytes(sdp->device->request_queue));
1066                 return put_user(val, ip);
1067         case SG_SET_COMMAND_Q:
1068                 result = get_user(val, ip);
1069                 if (result)
1070                         return result;
1071                 sfp->cmd_q = val ? 1 : 0;
1072                 return 0;
1073         case SG_GET_COMMAND_Q:
1074                 return put_user((int) sfp->cmd_q, ip);
1075         case SG_SET_KEEP_ORPHAN:
1076                 result = get_user(val, ip);
1077                 if (result)
1078                         return result;
1079                 sfp->keep_orphan = val;
1080                 return 0;
1081         case SG_GET_KEEP_ORPHAN:
1082                 return put_user((int) sfp->keep_orphan, ip);
1083         case SG_NEXT_CMD_LEN:
1084                 result = get_user(val, ip);
1085                 if (result)
1086                         return result;
1087                 if (val > SG_MAX_CDB_SIZE)
1088                         return -ENOMEM;
1089                 sfp->next_cmd_len = (val > 0) ? val : 0;
1090                 return 0;
1091         case SG_GET_VERSION_NUM:
1092                 return put_user(sg_version_num, ip);
1093         case SG_GET_ACCESS_COUNT:
1094                 /* faked - we don't have a real access count anymore */
1095                 val = (sdp->device ? 1 : 0);
1096                 return put_user(val, ip);
1097         case SG_GET_REQUEST_TABLE:
1098                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1099                         return -EFAULT;
1100                 else {
1101                         sg_req_info_t *rinfo;
1102
1103                         rinfo = kzalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
1104                                         GFP_KERNEL);
1105                         if (!rinfo)
1106                                 return -ENOMEM;
1107                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1108                         sg_fill_request_table(sfp, rinfo);
1109                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1110                         result = __copy_to_user(p, rinfo,
1111                                                 SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1112                         result = result ? -EFAULT : 0;
1113                         kfree(rinfo);
1114                         return result;
1115                 }
1116         case SG_EMULATED_HOST:
1117                 if (atomic_read(&sdp->detaching))
1118                         return -ENODEV;
1119                 return put_user(sdp->device->host->hostt->emulated, ip);
1120         case SCSI_IOCTL_SEND_COMMAND:
1121                 if (atomic_read(&sdp->detaching))
1122                         return -ENODEV;
1123                 if (read_only) {
1124                         unsigned char opcode = WRITE_6;
1125                         Scsi_Ioctl_Command __user *siocp = p;
1126
1127                         if (copy_from_user(&opcode, siocp->data, 1))
1128                                 return -EFAULT;
1129                         if (sg_allow_access(filp, &opcode))
1130                                 return -EPERM;
1131                 }
1132                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1133         case SG_SET_DEBUG:
1134                 result = get_user(val, ip);
1135                 if (result)
1136                         return result;
1137                 sdp->sgdebug = (char) val;
1138                 return 0;
1139         case BLKSECTGET:
1140                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1141                                 ip);
1142         case BLKTRACESETUP:
1143                 return blk_trace_setup(sdp->device->request_queue,
1144                                        sdp->disk->disk_name,
1145                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1146                                        NULL,
1147                                        (char *)arg);
1148         case BLKTRACESTART:
1149                 return blk_trace_startstop(sdp->device->request_queue, 1);
1150         case BLKTRACESTOP:
1151                 return blk_trace_startstop(sdp->device->request_queue, 0);
1152         case BLKTRACETEARDOWN:
1153                 return blk_trace_remove(sdp->device->request_queue);
1154         case SCSI_IOCTL_GET_IDLUN:
1155         case SCSI_IOCTL_GET_BUS_NUMBER:
1156         case SCSI_IOCTL_PROBE_HOST:
1157         case SG_GET_TRANSFORM:
1158         case SG_SCSI_RESET:
1159                 if (atomic_read(&sdp->detaching))
1160                         return -ENODEV;
1161                 break;
1162         default:
1163                 if (read_only)
1164                         return -EPERM;  /* don't know so take safe approach */
1165                 break;
1166         }
1167
1168         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1169                         cmd_in, filp->f_flags & O_NDELAY);
1170         if (result)
1171                 return result;
1172         return scsi_ioctl(sdp->device, cmd_in, p);
1173 }
1174
1175 #ifdef CONFIG_COMPAT
1176 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1177 {
1178         Sg_device *sdp;
1179         Sg_fd *sfp;
1180         struct scsi_device *sdev;
1181
1182         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1183                 return -ENXIO;
1184
1185         sdev = sdp->device;
1186         if (sdev->host->hostt->compat_ioctl) {
1187                 int ret;
1188
1189                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1190
1191                 return ret;
1192         }
1193
1194         return -ENOIOCTLCMD;
1195 }
1196 #endif
1197
1198 static unsigned int
1199 sg_poll(struct file *filp, poll_table * wait)
1200 {
1201         unsigned int res = 0;
1202         Sg_device *sdp;
1203         Sg_fd *sfp;
1204         Sg_request *srp;
1205         int count = 0;
1206         unsigned long iflags;
1207
1208         sfp = filp->private_data;
1209         if (!sfp)
1210                 return POLLERR;
1211         sdp = sfp->parentdp;
1212         if (!sdp)
1213                 return POLLERR;
1214         poll_wait(filp, &sfp->read_wait, wait);
1215         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1216         list_for_each_entry(srp, &sfp->rq_list, entry) {
1217                 /* if any read waiting, flag it */
1218                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1219                         res = POLLIN | POLLRDNORM;
1220                 ++count;
1221         }
1222         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1223
1224         if (atomic_read(&sdp->detaching))
1225                 res |= POLLHUP;
1226         else if (!sfp->cmd_q) {
1227                 if (0 == count)
1228                         res |= POLLOUT | POLLWRNORM;
1229         } else if (count < SG_MAX_QUEUE)
1230                 res |= POLLOUT | POLLWRNORM;
1231         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1232                                       "sg_poll: res=0x%x\n", (int) res));
1233         return res;
1234 }
1235
1236 static int
1237 sg_fasync(int fd, struct file *filp, int mode)
1238 {
1239         Sg_device *sdp;
1240         Sg_fd *sfp;
1241
1242         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1243                 return -ENXIO;
1244         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1245                                       "sg_fasync: mode=%d\n", mode));
1246
1247         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1248 }
1249
1250 static int
1251 sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1252 {
1253         Sg_fd *sfp;
1254         unsigned long offset, len, sa;
1255         Sg_scatter_hold *rsv_schp;
1256         int k, length;
1257
1258         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1259                 return VM_FAULT_SIGBUS;
1260         rsv_schp = &sfp->reserve;
1261         offset = vmf->pgoff << PAGE_SHIFT;
1262         if (offset >= rsv_schp->bufflen)
1263                 return VM_FAULT_SIGBUS;
1264         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1265                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1266                                       offset, rsv_schp->k_use_sg));
1267         sa = vma->vm_start;
1268         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1269         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1270                 len = vma->vm_end - sa;
1271                 len = (len < length) ? len : length;
1272                 if (offset < len) {
1273                         struct page *page = nth_page(rsv_schp->pages[k],
1274                                                      offset >> PAGE_SHIFT);
1275                         get_page(page); /* increment page count */
1276                         vmf->page = page;
1277                         return 0; /* success */
1278                 }
1279                 sa += len;
1280                 offset -= len;
1281         }
1282
1283         return VM_FAULT_SIGBUS;
1284 }
1285
1286 static const struct vm_operations_struct sg_mmap_vm_ops = {
1287         .fault = sg_vma_fault,
1288 };
1289
1290 static int
1291 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1292 {
1293         Sg_fd *sfp;
1294         unsigned long req_sz, len, sa;
1295         Sg_scatter_hold *rsv_schp;
1296         int k, length;
1297         int ret = 0;
1298
1299         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1300                 return -ENXIO;
1301         req_sz = vma->vm_end - vma->vm_start;
1302         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1303                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1304                                       (void *) vma->vm_start, (int) req_sz));
1305         if (vma->vm_pgoff)
1306                 return -EINVAL; /* want no offset */
1307         rsv_schp = &sfp->reserve;
1308         mutex_lock(&sfp->f_mutex);
1309         if (req_sz > rsv_schp->bufflen) {
1310                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1311                 goto out;
1312         }
1313
1314         sa = vma->vm_start;
1315         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1316         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1317                 len = vma->vm_end - sa;
1318                 len = (len < length) ? len : length;
1319                 sa += len;
1320         }
1321
1322         sfp->mmap_called = 1;
1323         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1324         vma->vm_private_data = sfp;
1325         vma->vm_ops = &sg_mmap_vm_ops;
1326 out:
1327         mutex_unlock(&sfp->f_mutex);
1328         return ret;
1329 }
1330
1331 static void
1332 sg_rq_end_io_usercontext(struct work_struct *work)
1333 {
1334         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1335         struct sg_fd *sfp = srp->parentfp;
1336
1337         sg_finish_rem_req(srp);
1338         sg_remove_request(sfp, srp);
1339         kref_put(&sfp->f_ref, sg_remove_sfp);
1340 }
1341
1342 /*
1343  * This function is a "bottom half" handler that is called by the mid
1344  * level when a command is completed (or has failed).
1345  */
1346 static void
1347 sg_rq_end_io(struct request *rq, int uptodate)
1348 {
1349         struct sg_request *srp = rq->end_io_data;
1350         Sg_device *sdp;
1351         Sg_fd *sfp;
1352         unsigned long iflags;
1353         unsigned int ms;
1354         char *sense;
1355         int result, resid, done = 1;
1356
1357         if (WARN_ON(srp->done != 0))
1358                 return;
1359
1360         sfp = srp->parentfp;
1361         if (WARN_ON(sfp == NULL))
1362                 return;
1363
1364         sdp = sfp->parentdp;
1365         if (unlikely(atomic_read(&sdp->detaching)))
1366                 pr_info("%s: device detaching\n", __func__);
1367
1368         sense = rq->sense;
1369         result = rq->errors;
1370         resid = rq->resid_len;
1371
1372         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1373                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1374                                       srp->header.pack_id, result));
1375         srp->header.resid = resid;
1376         ms = jiffies_to_msecs(jiffies);
1377         srp->header.duration = (ms > srp->header.duration) ?
1378                                 (ms - srp->header.duration) : 0;
1379         if (0 != result) {
1380                 struct scsi_sense_hdr sshdr;
1381
1382                 srp->header.status = 0xff & result;
1383                 srp->header.masked_status = status_byte(result);
1384                 srp->header.msg_status = msg_byte(result);
1385                 srp->header.host_status = host_byte(result);
1386                 srp->header.driver_status = driver_byte(result);
1387                 if ((sdp->sgdebug > 0) &&
1388                     ((CHECK_CONDITION == srp->header.masked_status) ||
1389                      (COMMAND_TERMINATED == srp->header.masked_status)))
1390                         __scsi_print_sense(sdp->device, __func__, sense,
1391                                            SCSI_SENSE_BUFFERSIZE);
1392
1393                 /* Following if statement is a patch supplied by Eric Youngdale */
1394                 if (driver_byte(result) != 0
1395                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1396                     && !scsi_sense_is_deferred(&sshdr)
1397                     && sshdr.sense_key == UNIT_ATTENTION
1398                     && sdp->device->removable) {
1399                         /* Detected possible disc change. Set the bit - this */
1400                         /* may be used if there are filesystems using this device */
1401                         sdp->device->changed = 1;
1402                 }
1403         }
1404         /* Rely on write phase to clean out srp status values, so no "else" */
1405
1406         /*
1407          * Free the request as soon as it is complete so that its resources
1408          * can be reused without waiting for userspace to read() the
1409          * result.  But keep the associated bio (if any) around until
1410          * blk_rq_unmap_user() can be called from user context.
1411          */
1412         srp->rq = NULL;
1413         if (rq->cmd != rq->__cmd)
1414                 kfree(rq->cmd);
1415         __blk_put_request(rq->q, rq);
1416
1417         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1418         if (unlikely(srp->orphan)) {
1419                 if (sfp->keep_orphan)
1420                         srp->sg_io_owned = 0;
1421                 else
1422                         done = 0;
1423         }
1424         srp->done = done;
1425         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1426
1427         if (likely(done)) {
1428                 /* Now wake up any sg_read() that is waiting for this
1429                  * packet.
1430                  */
1431                 wake_up_interruptible(&sfp->read_wait);
1432                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1433                 kref_put(&sfp->f_ref, sg_remove_sfp);
1434         } else {
1435                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1436                 schedule_work(&srp->ew.work);
1437         }
1438 }
1439
1440 static const struct file_operations sg_fops = {
1441         .owner = THIS_MODULE,
1442         .read = sg_read,
1443         .write = sg_write,
1444         .poll = sg_poll,
1445         .unlocked_ioctl = sg_ioctl,
1446 #ifdef CONFIG_COMPAT
1447         .compat_ioctl = sg_compat_ioctl,
1448 #endif
1449         .open = sg_open,
1450         .mmap = sg_mmap,
1451         .release = sg_release,
1452         .fasync = sg_fasync,
1453         .llseek = no_llseek,
1454 };
1455
1456 static struct class *sg_sysfs_class;
1457
1458 static int sg_sysfs_valid = 0;
1459
1460 static Sg_device *
1461 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1462 {
1463         struct request_queue *q = scsidp->request_queue;
1464         Sg_device *sdp;
1465         unsigned long iflags;
1466         int error;
1467         u32 k;
1468
1469         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1470         if (!sdp) {
1471                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1472                             "failure\n", __func__);
1473                 return ERR_PTR(-ENOMEM);
1474         }
1475
1476         idr_preload(GFP_KERNEL);
1477         write_lock_irqsave(&sg_index_lock, iflags);
1478
1479         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1480         if (error < 0) {
1481                 if (error == -ENOSPC) {
1482                         sdev_printk(KERN_WARNING, scsidp,
1483                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1484                                     scsidp->type, SG_MAX_DEVS - 1);
1485                         error = -ENODEV;
1486                 } else {
1487                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1488                                     "allocation Sg_device failure: %d\n",
1489                                     __func__, error);
1490                 }
1491                 goto out_unlock;
1492         }
1493         k = error;
1494
1495         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1496                                         "sg_alloc: dev=%d \n", k));
1497         sprintf(disk->disk_name, "sg%d", k);
1498         disk->first_minor = k;
1499         sdp->disk = disk;
1500         sdp->device = scsidp;
1501         mutex_init(&sdp->open_rel_lock);
1502         INIT_LIST_HEAD(&sdp->sfds);
1503         init_waitqueue_head(&sdp->open_wait);
1504         atomic_set(&sdp->detaching, 0);
1505         rwlock_init(&sdp->sfd_lock);
1506         sdp->sg_tablesize = queue_max_segments(q);
1507         sdp->index = k;
1508         kref_init(&sdp->d_ref);
1509         error = 0;
1510
1511 out_unlock:
1512         write_unlock_irqrestore(&sg_index_lock, iflags);
1513         idr_preload_end();
1514
1515         if (error) {
1516                 kfree(sdp);
1517                 return ERR_PTR(error);
1518         }
1519         return sdp;
1520 }
1521
1522 static int
1523 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1524 {
1525         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1526         struct gendisk *disk;
1527         Sg_device *sdp = NULL;
1528         struct cdev * cdev = NULL;
1529         int error;
1530         unsigned long iflags;
1531
1532         disk = alloc_disk(1);
1533         if (!disk) {
1534                 pr_warn("%s: alloc_disk failed\n", __func__);
1535                 return -ENOMEM;
1536         }
1537         disk->major = SCSI_GENERIC_MAJOR;
1538
1539         error = -ENOMEM;
1540         cdev = cdev_alloc();
1541         if (!cdev) {
1542                 pr_warn("%s: cdev_alloc failed\n", __func__);
1543                 goto out;
1544         }
1545         cdev->owner = THIS_MODULE;
1546         cdev->ops = &sg_fops;
1547
1548         sdp = sg_alloc(disk, scsidp);
1549         if (IS_ERR(sdp)) {
1550                 pr_warn("%s: sg_alloc failed\n", __func__);
1551                 error = PTR_ERR(sdp);
1552                 goto out;
1553         }
1554
1555         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1556         if (error)
1557                 goto cdev_add_err;
1558
1559         sdp->cdev = cdev;
1560         if (sg_sysfs_valid) {
1561                 struct device *sg_class_member;
1562
1563                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1564                                                 MKDEV(SCSI_GENERIC_MAJOR,
1565                                                       sdp->index),
1566                                                 sdp, "%s", disk->disk_name);
1567                 if (IS_ERR(sg_class_member)) {
1568                         pr_err("%s: device_create failed\n", __func__);
1569                         error = PTR_ERR(sg_class_member);
1570                         goto cdev_add_err;
1571                 }
1572                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1573                                           &sg_class_member->kobj, "generic");
1574                 if (error)
1575                         pr_err("%s: unable to make symlink 'generic' back "
1576                                "to sg%d\n", __func__, sdp->index);
1577         } else
1578                 pr_warn("%s: sg_sys Invalid\n", __func__);
1579
1580         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1581                     "type %d\n", sdp->index, scsidp->type);
1582
1583         dev_set_drvdata(cl_dev, sdp);
1584
1585         return 0;
1586
1587 cdev_add_err:
1588         write_lock_irqsave(&sg_index_lock, iflags);
1589         idr_remove(&sg_index_idr, sdp->index);
1590         write_unlock_irqrestore(&sg_index_lock, iflags);
1591         kfree(sdp);
1592
1593 out:
1594         put_disk(disk);
1595         if (cdev)
1596                 cdev_del(cdev);
1597         return error;
1598 }
1599
1600 static void
1601 sg_device_destroy(struct kref *kref)
1602 {
1603         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1604         unsigned long flags;
1605
1606         /* CAUTION!  Note that the device can still be found via idr_find()
1607          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1608          * any other cleanup.
1609          */
1610
1611         write_lock_irqsave(&sg_index_lock, flags);
1612         idr_remove(&sg_index_idr, sdp->index);
1613         write_unlock_irqrestore(&sg_index_lock, flags);
1614
1615         SCSI_LOG_TIMEOUT(3,
1616                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1617
1618         put_disk(sdp->disk);
1619         kfree(sdp);
1620 }
1621
1622 static void
1623 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1624 {
1625         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1626         Sg_device *sdp = dev_get_drvdata(cl_dev);
1627         unsigned long iflags;
1628         Sg_fd *sfp;
1629         int val;
1630
1631         if (!sdp)
1632                 return;
1633         /* want sdp->detaching non-zero as soon as possible */
1634         val = atomic_inc_return(&sdp->detaching);
1635         if (val > 1)
1636                 return; /* only want to do following once per device */
1637
1638         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1639                                       "%s\n", __func__));
1640
1641         read_lock_irqsave(&sdp->sfd_lock, iflags);
1642         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1643                 wake_up_interruptible_all(&sfp->read_wait);
1644                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1645         }
1646         wake_up_interruptible_all(&sdp->open_wait);
1647         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1648
1649         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1650         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1651         cdev_del(sdp->cdev);
1652         sdp->cdev = NULL;
1653
1654         kref_put(&sdp->d_ref, sg_device_destroy);
1655 }
1656
1657 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1658 module_param_named(def_reserved_size, def_reserved_size, int,
1659                    S_IRUGO | S_IWUSR);
1660 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1661
1662 MODULE_AUTHOR("Douglas Gilbert");
1663 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1664 MODULE_LICENSE("GPL");
1665 MODULE_VERSION(SG_VERSION_STR);
1666 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1667
1668 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1669                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1670 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1671 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1672
1673 static int __init
1674 init_sg(void)
1675 {
1676         int rc;
1677
1678         if (scatter_elem_sz < PAGE_SIZE) {
1679                 scatter_elem_sz = PAGE_SIZE;
1680                 scatter_elem_sz_prev = scatter_elem_sz;
1681         }
1682         if (def_reserved_size >= 0)
1683                 sg_big_buff = def_reserved_size;
1684         else
1685                 def_reserved_size = sg_big_buff;
1686
1687         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1688                                     SG_MAX_DEVS, "sg");
1689         if (rc)
1690                 return rc;
1691         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1692         if ( IS_ERR(sg_sysfs_class) ) {
1693                 rc = PTR_ERR(sg_sysfs_class);
1694                 goto err_out;
1695         }
1696         sg_sysfs_valid = 1;
1697         rc = scsi_register_interface(&sg_interface);
1698         if (0 == rc) {
1699 #ifdef CONFIG_SCSI_PROC_FS
1700                 sg_proc_init();
1701 #endif                          /* CONFIG_SCSI_PROC_FS */
1702                 return 0;
1703         }
1704         class_destroy(sg_sysfs_class);
1705 err_out:
1706         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1707         return rc;
1708 }
1709
1710 static void __exit
1711 exit_sg(void)
1712 {
1713 #ifdef CONFIG_SCSI_PROC_FS
1714         sg_proc_cleanup();
1715 #endif                          /* CONFIG_SCSI_PROC_FS */
1716         scsi_unregister_interface(&sg_interface);
1717         class_destroy(sg_sysfs_class);
1718         sg_sysfs_valid = 0;
1719         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1720                                  SG_MAX_DEVS);
1721         idr_destroy(&sg_index_idr);
1722 }
1723
1724 static int
1725 sg_start_req(Sg_request *srp, unsigned char *cmd)
1726 {
1727         int res;
1728         struct request *rq;
1729         Sg_fd *sfp = srp->parentfp;
1730         sg_io_hdr_t *hp = &srp->header;
1731         int dxfer_len = (int) hp->dxfer_len;
1732         int dxfer_dir = hp->dxfer_direction;
1733         unsigned int iov_count = hp->iovec_count;
1734         Sg_scatter_hold *req_schp = &srp->data;
1735         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1736         struct request_queue *q = sfp->parentdp->device->request_queue;
1737         struct rq_map_data *md, map_data;
1738         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1739         unsigned char *long_cmdp = NULL;
1740
1741         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1742                                       "sg_start_req: dxfer_len=%d\n",
1743                                       dxfer_len));
1744
1745         if (hp->cmd_len > BLK_MAX_CDB) {
1746                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1747                 if (!long_cmdp)
1748                         return -ENOMEM;
1749         }
1750
1751         /*
1752          * NOTE
1753          *
1754          * With scsi-mq enabled, there are a fixed number of preallocated
1755          * requests equal in number to shost->can_queue.  If all of the
1756          * preallocated requests are already in use, then using GFP_ATOMIC with
1757          * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1758          * will cause blk_get_request() to sleep until an active command
1759          * completes, freeing up a request.  Neither option is ideal, but
1760          * GFP_KERNEL is the better choice to prevent userspace from getting an
1761          * unexpected EWOULDBLOCK.
1762          *
1763          * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1764          * does not sleep except under memory pressure.
1765          */
1766         rq = blk_get_request(q, rw, GFP_KERNEL);
1767         if (IS_ERR(rq)) {
1768                 kfree(long_cmdp);
1769                 return PTR_ERR(rq);
1770         }
1771
1772         blk_rq_set_block_pc(rq);
1773
1774         if (hp->cmd_len > BLK_MAX_CDB)
1775                 rq->cmd = long_cmdp;
1776         memcpy(rq->cmd, cmd, hp->cmd_len);
1777         rq->cmd_len = hp->cmd_len;
1778
1779         srp->rq = rq;
1780         rq->end_io_data = srp;
1781         rq->sense = srp->sense_b;
1782         rq->retries = SG_DEFAULT_RETRIES;
1783
1784         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1785                 return 0;
1786
1787         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1788             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1789             !sfp->parentdp->device->host->unchecked_isa_dma &&
1790             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1791                 md = NULL;
1792         else
1793                 md = &map_data;
1794
1795         if (md) {
1796                 mutex_lock(&sfp->f_mutex);
1797                 if (dxfer_len <= rsv_schp->bufflen &&
1798                     !sfp->res_in_use) {
1799                         sfp->res_in_use = 1;
1800                         sg_link_reserve(sfp, srp, dxfer_len);
1801                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1802                         res = -EBUSY; /* sfp->res_in_use == 1 */
1803                         if (dxfer_len > rsv_schp->bufflen)
1804                                 res = -ENOMEM;
1805                         mutex_unlock(&sfp->f_mutex);
1806                         return res;
1807                 } else {
1808                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1809                         if (res) {
1810                                 mutex_unlock(&sfp->f_mutex);
1811                                 return res;
1812                         }
1813                 }
1814                 mutex_unlock(&sfp->f_mutex);
1815
1816                 md->pages = req_schp->pages;
1817                 md->page_order = req_schp->page_order;
1818                 md->nr_entries = req_schp->k_use_sg;
1819                 md->offset = 0;
1820                 md->null_mapped = hp->dxferp ? 0 : 1;
1821                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1822                         md->from_user = 1;
1823                 else
1824                         md->from_user = 0;
1825         }
1826
1827         if (iov_count) {
1828                 struct iovec *iov = NULL;
1829                 struct iov_iter i;
1830
1831                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1832                 if (res < 0)
1833                         return res;
1834
1835                 iov_iter_truncate(&i, hp->dxfer_len);
1836                 if (!iov_iter_count(&i)) {
1837                         kfree(iov);
1838                         return -EINVAL;
1839                 }
1840
1841                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1842                 kfree(iov);
1843         } else
1844                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1845                                       hp->dxfer_len, GFP_ATOMIC);
1846
1847         if (!res) {
1848                 srp->bio = rq->bio;
1849
1850                 if (!md) {
1851                         req_schp->dio_in_use = 1;
1852                         hp->info |= SG_INFO_DIRECT_IO;
1853                 }
1854         }
1855         return res;
1856 }
1857
1858 static int
1859 sg_finish_rem_req(Sg_request *srp)
1860 {
1861         int ret = 0;
1862
1863         Sg_fd *sfp = srp->parentfp;
1864         Sg_scatter_hold *req_schp = &srp->data;
1865
1866         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1867                                       "sg_finish_rem_req: res_used=%d\n",
1868                                       (int) srp->res_used));
1869         if (srp->bio)
1870                 ret = blk_rq_unmap_user(srp->bio);
1871
1872         if (srp->rq) {
1873                 if (srp->rq->cmd != srp->rq->__cmd)
1874                         kfree(srp->rq->cmd);
1875                 blk_put_request(srp->rq);
1876         }
1877
1878         if (srp->res_used)
1879                 sg_unlink_reserve(sfp, srp);
1880         else
1881                 sg_remove_scat(sfp, req_schp);
1882
1883         return ret;
1884 }
1885
1886 static int
1887 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1888 {
1889         int sg_bufflen = tablesize * sizeof(struct page *);
1890         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1891
1892         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1893         if (!schp->pages)
1894                 return -ENOMEM;
1895         schp->sglist_len = sg_bufflen;
1896         return tablesize;       /* number of scat_gath elements allocated */
1897 }
1898
1899 static int
1900 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1901 {
1902         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1903         int sg_tablesize = sfp->parentdp->sg_tablesize;
1904         int blk_size = buff_size, order;
1905         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1906         struct sg_device *sdp = sfp->parentdp;
1907
1908         if (blk_size < 0)
1909                 return -EFAULT;
1910         if (0 == blk_size)
1911                 ++blk_size;     /* don't know why */
1912         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1913         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1914         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1915                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1916                 buff_size, blk_size));
1917
1918         /* N.B. ret_sz carried into this block ... */
1919         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1920         if (mx_sc_elems < 0)
1921                 return mx_sc_elems;     /* most likely -ENOMEM */
1922
1923         num = scatter_elem_sz;
1924         if (unlikely(num != scatter_elem_sz_prev)) {
1925                 if (num < PAGE_SIZE) {
1926                         scatter_elem_sz = PAGE_SIZE;
1927                         scatter_elem_sz_prev = PAGE_SIZE;
1928                 } else
1929                         scatter_elem_sz_prev = num;
1930         }
1931
1932         if (sdp->device->host->unchecked_isa_dma)
1933                 gfp_mask |= GFP_DMA;
1934
1935         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1936                 gfp_mask |= __GFP_ZERO;
1937
1938         order = get_order(num);
1939 retry:
1940         ret_sz = 1 << (PAGE_SHIFT + order);
1941
1942         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1943              k++, rem_sz -= ret_sz) {
1944
1945                 num = (rem_sz > scatter_elem_sz_prev) ?
1946                         scatter_elem_sz_prev : rem_sz;
1947
1948                 schp->pages[k] = alloc_pages(gfp_mask | __GFP_ZERO, order);
1949                 if (!schp->pages[k])
1950                         goto out;
1951
1952                 if (num == scatter_elem_sz_prev) {
1953                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1954                                 scatter_elem_sz = ret_sz;
1955                                 scatter_elem_sz_prev = ret_sz;
1956                         }
1957                 }
1958
1959                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1960                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1961                                  k, num, ret_sz));
1962         }               /* end of for loop */
1963
1964         schp->page_order = order;
1965         schp->k_use_sg = k;
1966         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1967                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1968                          k, rem_sz));
1969
1970         schp->bufflen = blk_size;
1971         if (rem_sz > 0) /* must have failed */
1972                 return -ENOMEM;
1973         return 0;
1974 out:
1975         for (i = 0; i < k; i++)
1976                 __free_pages(schp->pages[i], order);
1977
1978         if (--order >= 0)
1979                 goto retry;
1980
1981         return -ENOMEM;
1982 }
1983
1984 static void
1985 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1986 {
1987         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1988                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1989         if (schp->pages && schp->sglist_len > 0) {
1990                 if (!schp->dio_in_use) {
1991                         int k;
1992
1993                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1994                                 SCSI_LOG_TIMEOUT(5,
1995                                         sg_printk(KERN_INFO, sfp->parentdp,
1996                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1997                                         k, schp->pages[k]));
1998                                 __free_pages(schp->pages[k], schp->page_order);
1999                         }
2000
2001                         kfree(schp->pages);
2002                 }
2003         }
2004         memset(schp, 0, sizeof (*schp));
2005 }
2006
2007 static int
2008 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
2009 {
2010         Sg_scatter_hold *schp = &srp->data;
2011         int k, num;
2012
2013         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2014                          "sg_read_oxfer: num_read_xfer=%d\n",
2015                          num_read_xfer));
2016         if ((!outp) || (num_read_xfer <= 0))
2017                 return 0;
2018
2019         num = 1 << (PAGE_SHIFT + schp->page_order);
2020         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2021                 if (num > num_read_xfer) {
2022                         if (__copy_to_user(outp, page_address(schp->pages[k]),
2023                                            num_read_xfer))
2024                                 return -EFAULT;
2025                         break;
2026                 } else {
2027                         if (__copy_to_user(outp, page_address(schp->pages[k]),
2028                                            num))
2029                                 return -EFAULT;
2030                         num_read_xfer -= num;
2031                         if (num_read_xfer <= 0)
2032                                 break;
2033                         outp += num;
2034                 }
2035         }
2036
2037         return 0;
2038 }
2039
2040 static void
2041 sg_build_reserve(Sg_fd * sfp, int req_size)
2042 {
2043         Sg_scatter_hold *schp = &sfp->reserve;
2044
2045         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2046                          "sg_build_reserve: req_size=%d\n", req_size));
2047         do {
2048                 if (req_size < PAGE_SIZE)
2049                         req_size = PAGE_SIZE;
2050                 if (0 == sg_build_indirect(schp, sfp, req_size))
2051                         return;
2052                 else
2053                         sg_remove_scat(sfp, schp);
2054                 req_size >>= 1; /* divide by 2 */
2055         } while (req_size > (PAGE_SIZE / 2));
2056 }
2057
2058 static void
2059 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2060 {
2061         Sg_scatter_hold *req_schp = &srp->data;
2062         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2063         int k, num, rem;
2064
2065         srp->res_used = 1;
2066         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2067                          "sg_link_reserve: size=%d\n", size));
2068         rem = size;
2069
2070         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2071         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2072                 if (rem <= num) {
2073                         req_schp->k_use_sg = k + 1;
2074                         req_schp->sglist_len = rsv_schp->sglist_len;
2075                         req_schp->pages = rsv_schp->pages;
2076
2077                         req_schp->bufflen = size;
2078                         req_schp->page_order = rsv_schp->page_order;
2079                         break;
2080                 } else
2081                         rem -= num;
2082         }
2083
2084         if (k >= rsv_schp->k_use_sg)
2085                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2086                                  "sg_link_reserve: BAD size\n"));
2087 }
2088
2089 static void
2090 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2091 {
2092         Sg_scatter_hold *req_schp = &srp->data;
2093
2094         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2095                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2096                                       (int) req_schp->k_use_sg));
2097         req_schp->k_use_sg = 0;
2098         req_schp->bufflen = 0;
2099         req_schp->pages = NULL;
2100         req_schp->page_order = 0;
2101         req_schp->sglist_len = 0;
2102         srp->res_used = 0;
2103         /* Called without mutex lock to avoid deadlock */
2104         sfp->res_in_use = 0;
2105 }
2106
2107 static Sg_request *
2108 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2109 {
2110         Sg_request *resp;
2111         unsigned long iflags;
2112
2113         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2114         list_for_each_entry(resp, &sfp->rq_list, entry) {
2115                 /* look for requests that are ready + not SG_IO owned */
2116                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2117                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2118                         resp->done = 2; /* guard against other readers */
2119                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2120                         return resp;
2121                 }
2122         }
2123         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2124         return NULL;
2125 }
2126
2127 /* always adds to end of list */
2128 static Sg_request *
2129 sg_add_request(Sg_fd * sfp)
2130 {
2131         int k;
2132         unsigned long iflags;
2133         Sg_request *rp = sfp->req_arr;
2134
2135         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2136         if (!list_empty(&sfp->rq_list)) {
2137                 if (!sfp->cmd_q)
2138                         goto out_unlock;
2139
2140                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2141                         if (!rp->parentfp)
2142                                 break;
2143                 }
2144                 if (k >= SG_MAX_QUEUE)
2145                         goto out_unlock;
2146         }
2147         memset(rp, 0, sizeof (Sg_request));
2148         rp->parentfp = sfp;
2149         rp->header.duration = jiffies_to_msecs(jiffies);
2150         list_add_tail(&rp->entry, &sfp->rq_list);
2151         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2152         return rp;
2153 out_unlock:
2154         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2155         return NULL;
2156 }
2157
2158 /* Return of 1 for found; 0 for not found */
2159 static int
2160 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2161 {
2162         unsigned long iflags;
2163         int res = 0;
2164
2165         if (!sfp || !srp || list_empty(&sfp->rq_list))
2166                 return res;
2167         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2168         if (!list_empty(&srp->entry)) {
2169                 list_del(&srp->entry);
2170                 srp->parentfp = NULL;
2171                 res = 1;
2172         }
2173         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2174         return res;
2175 }
2176
2177 static Sg_fd *
2178 sg_add_sfp(Sg_device * sdp)
2179 {
2180         Sg_fd *sfp;
2181         unsigned long iflags;
2182         int bufflen;
2183
2184         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2185         if (!sfp)
2186                 return ERR_PTR(-ENOMEM);
2187
2188         init_waitqueue_head(&sfp->read_wait);
2189         rwlock_init(&sfp->rq_list_lock);
2190         INIT_LIST_HEAD(&sfp->rq_list);
2191         kref_init(&sfp->f_ref);
2192         mutex_init(&sfp->f_mutex);
2193         sfp->timeout = SG_DEFAULT_TIMEOUT;
2194         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2195         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2196         sfp->cmd_q = SG_DEF_COMMAND_Q;
2197         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2198         sfp->parentdp = sdp;
2199         write_lock_irqsave(&sdp->sfd_lock, iflags);
2200         if (atomic_read(&sdp->detaching)) {
2201                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2202                 kfree(sfp);
2203                 return ERR_PTR(-ENODEV);
2204         }
2205         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2206         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2207         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2208                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2209         if (unlikely(sg_big_buff != def_reserved_size))
2210                 sg_big_buff = def_reserved_size;
2211
2212         bufflen = min_t(int, sg_big_buff,
2213                         max_sectors_bytes(sdp->device->request_queue));
2214         sg_build_reserve(sfp, bufflen);
2215         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2216                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2217                                       sfp->reserve.bufflen,
2218                                       sfp->reserve.k_use_sg));
2219
2220         kref_get(&sdp->d_ref);
2221         __module_get(THIS_MODULE);
2222         return sfp;
2223 }
2224
2225 static void
2226 sg_remove_sfp_usercontext(struct work_struct *work)
2227 {
2228         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2229         struct sg_device *sdp = sfp->parentdp;
2230         Sg_request *srp;
2231         unsigned long iflags;
2232
2233         /* Cleanup any responses which were never read(). */
2234         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2235         while (!list_empty(&sfp->rq_list)) {
2236                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2237                 sg_finish_rem_req(srp);
2238                 list_del(&srp->entry);
2239                 srp->parentfp = NULL;
2240         }
2241         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2242
2243         if (sfp->reserve.bufflen > 0) {
2244                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2245                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2246                                 (int) sfp->reserve.bufflen,
2247                                 (int) sfp->reserve.k_use_sg));
2248                 sg_remove_scat(sfp, &sfp->reserve);
2249         }
2250
2251         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2252                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2253         kfree(sfp);
2254
2255         scsi_device_put(sdp->device);
2256         kref_put(&sdp->d_ref, sg_device_destroy);
2257         module_put(THIS_MODULE);
2258 }
2259
2260 static void
2261 sg_remove_sfp(struct kref *kref)
2262 {
2263         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2264         struct sg_device *sdp = sfp->parentdp;
2265         unsigned long iflags;
2266
2267         write_lock_irqsave(&sdp->sfd_lock, iflags);
2268         list_del(&sfp->sfd_siblings);
2269         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2270
2271         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2272         schedule_work(&sfp->ew.work);
2273 }
2274
2275 #ifdef CONFIG_SCSI_PROC_FS
2276 static int
2277 sg_idr_max_id(int id, void *p, void *data)
2278 {
2279         int *k = data;
2280
2281         if (*k < id)
2282                 *k = id;
2283
2284         return 0;
2285 }
2286
2287 static int
2288 sg_last_dev(void)
2289 {
2290         int k = -1;
2291         unsigned long iflags;
2292
2293         read_lock_irqsave(&sg_index_lock, iflags);
2294         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2295         read_unlock_irqrestore(&sg_index_lock, iflags);
2296         return k + 1;           /* origin 1 */
2297 }
2298 #endif
2299
2300 /* must be called with sg_index_lock held */
2301 static Sg_device *sg_lookup_dev(int dev)
2302 {
2303         return idr_find(&sg_index_idr, dev);
2304 }
2305
2306 static Sg_device *
2307 sg_get_dev(int dev)
2308 {
2309         struct sg_device *sdp;
2310         unsigned long flags;
2311
2312         read_lock_irqsave(&sg_index_lock, flags);
2313         sdp = sg_lookup_dev(dev);
2314         if (!sdp)
2315                 sdp = ERR_PTR(-ENXIO);
2316         else if (atomic_read(&sdp->detaching)) {
2317                 /* If sdp->detaching, then the refcount may already be 0, in
2318                  * which case it would be a bug to do kref_get().
2319                  */
2320                 sdp = ERR_PTR(-ENODEV);
2321         } else
2322                 kref_get(&sdp->d_ref);
2323         read_unlock_irqrestore(&sg_index_lock, flags);
2324
2325         return sdp;
2326 }
2327
2328 #ifdef CONFIG_SCSI_PROC_FS
2329
2330 static struct proc_dir_entry *sg_proc_sgp = NULL;
2331
2332 static char sg_proc_sg_dirname[] = "scsi/sg";
2333
2334 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2335
2336 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2337 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2338                                   size_t count, loff_t *off);
2339 static const struct file_operations adio_fops = {
2340         .owner = THIS_MODULE,
2341         .open = sg_proc_single_open_adio,
2342         .read = seq_read,
2343         .llseek = seq_lseek,
2344         .write = sg_proc_write_adio,
2345         .release = single_release,
2346 };
2347
2348 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2349 static ssize_t sg_proc_write_dressz(struct file *filp,
2350                 const char __user *buffer, size_t count, loff_t *off);
2351 static const struct file_operations dressz_fops = {
2352         .owner = THIS_MODULE,
2353         .open = sg_proc_single_open_dressz,
2354         .read = seq_read,
2355         .llseek = seq_lseek,
2356         .write = sg_proc_write_dressz,
2357         .release = single_release,
2358 };
2359
2360 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2361 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2362 static const struct file_operations version_fops = {
2363         .owner = THIS_MODULE,
2364         .open = sg_proc_single_open_version,
2365         .read = seq_read,
2366         .llseek = seq_lseek,
2367         .release = single_release,
2368 };
2369
2370 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2371 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2372 static const struct file_operations devhdr_fops = {
2373         .owner = THIS_MODULE,
2374         .open = sg_proc_single_open_devhdr,
2375         .read = seq_read,
2376         .llseek = seq_lseek,
2377         .release = single_release,
2378 };
2379
2380 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2381 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2382 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2383 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2384 static void dev_seq_stop(struct seq_file *s, void *v);
2385 static const struct file_operations dev_fops = {
2386         .owner = THIS_MODULE,
2387         .open = sg_proc_open_dev,
2388         .read = seq_read,
2389         .llseek = seq_lseek,
2390         .release = seq_release,
2391 };
2392 static const struct seq_operations dev_seq_ops = {
2393         .start = dev_seq_start,
2394         .next  = dev_seq_next,
2395         .stop  = dev_seq_stop,
2396         .show  = sg_proc_seq_show_dev,
2397 };
2398
2399 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2400 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2401 static const struct file_operations devstrs_fops = {
2402         .owner = THIS_MODULE,
2403         .open = sg_proc_open_devstrs,
2404         .read = seq_read,
2405         .llseek = seq_lseek,
2406         .release = seq_release,
2407 };
2408 static const struct seq_operations devstrs_seq_ops = {
2409         .start = dev_seq_start,
2410         .next  = dev_seq_next,
2411         .stop  = dev_seq_stop,
2412         .show  = sg_proc_seq_show_devstrs,
2413 };
2414
2415 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2416 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2417 static const struct file_operations debug_fops = {
2418         .owner = THIS_MODULE,
2419         .open = sg_proc_open_debug,
2420         .read = seq_read,
2421         .llseek = seq_lseek,
2422         .release = seq_release,
2423 };
2424 static const struct seq_operations debug_seq_ops = {
2425         .start = dev_seq_start,
2426         .next  = dev_seq_next,
2427         .stop  = dev_seq_stop,
2428         .show  = sg_proc_seq_show_debug,
2429 };
2430
2431
2432 struct sg_proc_leaf {
2433         const char * name;
2434         const struct file_operations * fops;
2435 };
2436
2437 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2438         {"allow_dio", &adio_fops},
2439         {"debug", &debug_fops},
2440         {"def_reserved_size", &dressz_fops},
2441         {"device_hdr", &devhdr_fops},
2442         {"devices", &dev_fops},
2443         {"device_strs", &devstrs_fops},
2444         {"version", &version_fops}
2445 };
2446
2447 static int
2448 sg_proc_init(void)
2449 {
2450         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2451         int k;
2452
2453         sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2454         if (!sg_proc_sgp)
2455                 return 1;
2456         for (k = 0; k < num_leaves; ++k) {
2457                 const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2458                 umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2459                 proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2460         }
2461         return 0;
2462 }
2463
2464 static void
2465 sg_proc_cleanup(void)
2466 {
2467         int k;
2468         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2469
2470         if (!sg_proc_sgp)
2471                 return;
2472         for (k = 0; k < num_leaves; ++k)
2473                 remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2474         remove_proc_entry(sg_proc_sg_dirname, NULL);
2475 }
2476
2477
2478 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2479 {
2480         seq_printf(s, "%d\n", *((int *)s->private));
2481         return 0;
2482 }
2483
2484 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2485 {
2486         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2487 }
2488
2489 static ssize_t
2490 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2491                    size_t count, loff_t *off)
2492 {
2493         int err;
2494         unsigned long num;
2495
2496         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2497                 return -EACCES;
2498         err = kstrtoul_from_user(buffer, count, 0, &num);
2499         if (err)
2500                 return err;
2501         sg_allow_dio = num ? 1 : 0;
2502         return count;
2503 }
2504
2505 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2506 {
2507         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2508 }
2509
2510 static ssize_t
2511 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2512                      size_t count, loff_t *off)
2513 {
2514         int err;
2515         unsigned long k = ULONG_MAX;
2516
2517         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2518                 return -EACCES;
2519
2520         err = kstrtoul_from_user(buffer, count, 0, &k);
2521         if (err)
2522                 return err;
2523         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2524                 sg_big_buff = k;
2525                 return count;
2526         }
2527         return -ERANGE;
2528 }
2529
2530 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2531 {
2532         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2533                    sg_version_date);
2534         return 0;
2535 }
2536
2537 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2538 {
2539         return single_open(file, sg_proc_seq_show_version, NULL);
2540 }
2541
2542 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2543 {
2544         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2545         return 0;
2546 }
2547
2548 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2549 {
2550         return single_open(file, sg_proc_seq_show_devhdr, NULL);
2551 }
2552
2553 struct sg_proc_deviter {
2554         loff_t  index;
2555         size_t  max;
2556 };
2557
2558 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2559 {
2560         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2561
2562         s->private = it;
2563         if (! it)
2564                 return NULL;
2565
2566         it->index = *pos;
2567         it->max = sg_last_dev();
2568         if (it->index >= it->max)
2569                 return NULL;
2570         return it;
2571 }
2572
2573 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2574 {
2575         struct sg_proc_deviter * it = s->private;
2576
2577         *pos = ++it->index;
2578         return (it->index < it->max) ? it : NULL;
2579 }
2580
2581 static void dev_seq_stop(struct seq_file *s, void *v)
2582 {
2583         kfree(s->private);
2584 }
2585
2586 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2587 {
2588         return seq_open(file, &dev_seq_ops);
2589 }
2590
2591 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2592 {
2593         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2594         Sg_device *sdp;
2595         struct scsi_device *scsidp;
2596         unsigned long iflags;
2597
2598         read_lock_irqsave(&sg_index_lock, iflags);
2599         sdp = it ? sg_lookup_dev(it->index) : NULL;
2600         if ((NULL == sdp) || (NULL == sdp->device) ||
2601             (atomic_read(&sdp->detaching)))
2602                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2603         else {
2604                 scsidp = sdp->device;
2605                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2606                               scsidp->host->host_no, scsidp->channel,
2607                               scsidp->id, scsidp->lun, (int) scsidp->type,
2608                               1,
2609                               (int) scsidp->queue_depth,
2610                               (int) atomic_read(&scsidp->device_busy),
2611                               (int) scsi_device_online(scsidp));
2612         }
2613         read_unlock_irqrestore(&sg_index_lock, iflags);
2614         return 0;
2615 }
2616
2617 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2618 {
2619         return seq_open(file, &devstrs_seq_ops);
2620 }
2621
2622 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2623 {
2624         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2625         Sg_device *sdp;
2626         struct scsi_device *scsidp;
2627         unsigned long iflags;
2628
2629         read_lock_irqsave(&sg_index_lock, iflags);
2630         sdp = it ? sg_lookup_dev(it->index) : NULL;
2631         scsidp = sdp ? sdp->device : NULL;
2632         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2633                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2634                            scsidp->vendor, scsidp->model, scsidp->rev);
2635         else
2636                 seq_puts(s, "<no active device>\n");
2637         read_unlock_irqrestore(&sg_index_lock, iflags);
2638         return 0;
2639 }
2640
2641 /* must be called while holding sg_index_lock */
2642 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2643 {
2644         int k, new_interface, blen, usg;
2645         Sg_request *srp;
2646         Sg_fd *fp;
2647         const sg_io_hdr_t *hp;
2648         const char * cp;
2649         unsigned int ms;
2650
2651         k = 0;
2652         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2653                 k++;
2654                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2655                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2656                            "(res)sgat=%d low_dma=%d\n", k,
2657                            jiffies_to_msecs(fp->timeout),
2658                            fp->reserve.bufflen,
2659                            (int) fp->reserve.k_use_sg,
2660                            (int) sdp->device->host->unchecked_isa_dma);
2661                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2662                            (int) fp->cmd_q, (int) fp->force_packid,
2663                            (int) fp->keep_orphan);
2664                 list_for_each_entry(srp, &fp->rq_list, entry) {
2665                         hp = &srp->header;
2666                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2667                         if (srp->res_used) {
2668                                 if (new_interface &&
2669                                     (SG_FLAG_MMAP_IO & hp->flags))
2670                                         cp = "     mmap>> ";
2671                                 else
2672                                         cp = "     rb>> ";
2673                         } else {
2674                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2675                                         cp = "     dio>> ";
2676                                 else
2677                                         cp = "     ";
2678                         }
2679                         seq_puts(s, cp);
2680                         blen = srp->data.bufflen;
2681                         usg = srp->data.k_use_sg;
2682                         seq_puts(s, srp->done ?
2683                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2684                                   : "act:");
2685                         seq_printf(s, " id=%d blen=%d",
2686                                    srp->header.pack_id, blen);
2687                         if (srp->done)
2688                                 seq_printf(s, " dur=%d", hp->duration);
2689                         else {
2690                                 ms = jiffies_to_msecs(jiffies);
2691                                 seq_printf(s, " t_o/elap=%d/%d",
2692                                         (new_interface ? hp->timeout :
2693                                                   jiffies_to_msecs(fp->timeout)),
2694                                         (ms > hp->duration ? ms - hp->duration : 0));
2695                         }
2696                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2697                                    (int) srp->data.cmd_opcode);
2698                 }
2699                 if (list_empty(&fp->rq_list))
2700                         seq_puts(s, "     No requests active\n");
2701                 read_unlock(&fp->rq_list_lock);
2702         }
2703 }
2704
2705 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2706 {
2707         return seq_open(file, &debug_seq_ops);
2708 }
2709
2710 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2711 {
2712         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2713         Sg_device *sdp;
2714         unsigned long iflags;
2715
2716         if (it && (0 == it->index))
2717                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2718                            (int)it->max, sg_big_buff);
2719
2720         read_lock_irqsave(&sg_index_lock, iflags);
2721         sdp = it ? sg_lookup_dev(it->index) : NULL;
2722         if (NULL == sdp)
2723                 goto skip;
2724         read_lock(&sdp->sfd_lock);
2725         if (!list_empty(&sdp->sfds)) {
2726                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2727                 if (atomic_read(&sdp->detaching))
2728                         seq_puts(s, "detaching pending close ");
2729                 else if (sdp->device) {
2730                         struct scsi_device *scsidp = sdp->device;
2731
2732                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2733                                    scsidp->host->host_no,
2734                                    scsidp->channel, scsidp->id,
2735                                    scsidp->lun,
2736                                    scsidp->host->hostt->emulated);
2737                 }
2738                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2739                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2740                 sg_proc_debug_helper(s, sdp);
2741         }
2742         read_unlock(&sdp->sfd_lock);
2743 skip:
2744         read_unlock_irqrestore(&sg_index_lock, iflags);
2745         return 0;
2746 }
2747
2748 #endif                          /* CONFIG_SCSI_PROC_FS */
2749
2750 module_init(init_sg);
2751 module_exit(exit_sg);