include/linux/blkdev_compat.h

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (C) 2011 Lawrence Livermore National Security, LLC.
  24  * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
  25  * Written by Brian Behlendorf <behlendorf1@llnl.gov>.
  26  * LLNL-CODE-403049.
  27  */
  28
  29 #ifndef _ZFS_BLKDEV_H
  30 #define _ZFS_BLKDEV_H
  31
  32 #include <linux/blkdev.h>
  33 #include <linux/elevator.h>
  34
  35 #ifndef HAVE_FMODE_T
  36 typedef unsigned __bitwise__ fmode_t;
  37 #endif /* HAVE_FMODE_T */
  38
  39 /*
  40  * 4.7 - 4.x API,
  41  * The blk_queue_write_cache() interface has replaced blk_queue_flush()
  42  * interface.  However, the new interface is GPL-only thus we implement
  43  * our own trivial wrapper when the GPL-only version is detected.
  44  *
  45  * 2.6.36 - 4.6 API,
  46  * The blk_queue_flush() interface has replaced blk_queue_ordered()
  47  * interface.  However, while the old interface was available to all the
  48  * new one is GPL-only.   Thus if the GPL-only version is detected we
  49  * implement our own trivial helper.
  50  *
  51  * 2.6.x - 2.6.35
  52  * Legacy blk_queue_ordered() interface.
  53  */
  54 static inline void
  55 blk_queue_set_write_cache(struct request_queue *q, bool wc, bool fua)
  56 {
  57 #if defined(HAVE_BLK_QUEUE_WRITE_CACHE_GPL_ONLY)
  58         spin_lock_irq(q->queue_lock);
  59         if (wc)
  60                 queue_flag_set(QUEUE_FLAG_WC, q);
  61         else
  62                 queue_flag_clear(QUEUE_FLAG_WC, q);
  63         if (fua)
  64                 queue_flag_set(QUEUE_FLAG_FUA, q);
  65         else
  66                 queue_flag_clear(QUEUE_FLAG_FUA, q);
  67         spin_unlock_irq(q->queue_lock);
  68 #elif defined(HAVE_BLK_QUEUE_WRITE_CACHE)
  69         blk_queue_write_cache(q, wc, fua);
  70 #elif defined(HAVE_BLK_QUEUE_FLUSH_GPL_ONLY)
  71         if (wc)
  72                 q->flush_flags |= REQ_FLUSH;
  73         if (fua)
  74                 q->flush_flags |= REQ_FUA;
  75 #elif defined(HAVE_BLK_QUEUE_FLUSH)
  76         blk_queue_flush(q, (wc ? REQ_FLUSH : 0) | (fua ? REQ_FUA : 0));
  77 #else
  78         blk_queue_ordered(q, QUEUE_ORDERED_DRAIN, NULL);
  79 #endif
  80 }
  81
  82 /*
  83  * Most of the blk_* macros were removed in 2.6.36.  Ostensibly this was
  84  * done to improve readability and allow easier grepping.  However, from
  85  * a portability stand point the macros are helpful.  Therefore the needed
  86  * macros are redefined here if they are missing from the kernel.
  87  */
  88 #ifndef blk_fs_request
  89 #define blk_fs_request(rq)      ((rq)->cmd_type == REQ_TYPE_FS)
  90 #endif
  91
  92 /*
  93  * 2.6.27 API change,
  94  * The blk_queue_stackable() queue flag was added in 2.6.27 to handle dm
  95  * stacking drivers.  Prior to this request stacking drivers were detected
  96  * by checking (q->request_fn == NULL), for earlier kernels we revert to
  97  * this legacy behavior.
  98  */
  99 #ifndef blk_queue_stackable
 100 #define blk_queue_stackable(q)  ((q)->request_fn == NULL)
 101 #endif
 102
 103 /*
 104  * 2.6.34 API change,
 105  * The blk_queue_max_hw_sectors() function replaces blk_queue_max_sectors().
 106  */
 107 #ifndef HAVE_BLK_QUEUE_MAX_HW_SECTORS
 108 #define blk_queue_max_hw_sectors __blk_queue_max_hw_sectors
 109 static inline void
 110 __blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
 111 {
 112         blk_queue_max_sectors(q, max_hw_sectors);
 113 }
 114 #endif
 115
 116 /*
 117  * 2.6.34 API change,
 118  * The blk_queue_max_segments() function consolidates
 119  * blk_queue_max_hw_segments() and blk_queue_max_phys_segments().
 120  */
 121 #ifndef HAVE_BLK_QUEUE_MAX_SEGMENTS
 122 #define blk_queue_max_segments __blk_queue_max_segments
 123 static inline void
 124 __blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
 125 {
 126         blk_queue_max_phys_segments(q, max_segments);
 127         blk_queue_max_hw_segments(q, max_segments);
 128 }
 129 #endif
 130
 131 #ifndef HAVE_GET_DISK_RO
 132 static inline int
 133 get_disk_ro(struct gendisk *disk)
 134 {
 135         int policy = 0;
 136
 137         if (disk->part[0])
 138                 policy = disk->part[0]->policy;
 139
 140         return (policy);
 141 }
 142 #endif /* HAVE_GET_DISK_RO */
 143
 144 #ifdef HAVE_BIO_BVEC_ITER
 145 #define BIO_BI_SECTOR(bio)      (bio)->bi_iter.bi_sector
 146 #define BIO_BI_SIZE(bio)        (bio)->bi_iter.bi_size
 147 #define BIO_BI_IDX(bio)         (bio)->bi_iter.bi_idx
 148 #define BIO_BI_SKIP(bio)        (bio)->bi_iter.bi_bvec_done
 149 #define bio_for_each_segment4(bv, bvp, b, i)    \
 150         bio_for_each_segment((bv), (b), (i))
 151 typedef struct bvec_iter bvec_iterator_t;
 152 #else
 153 #define BIO_BI_SECTOR(bio)      (bio)->bi_sector
 154 #define BIO_BI_SIZE(bio)        (bio)->bi_size
 155 #define BIO_BI_IDX(bio)         (bio)->bi_idx
 156 #define BIO_BI_SKIP(bio)        (0)
 157 #define bio_for_each_segment4(bv, bvp, b, i)    \
 158         bio_for_each_segment((bvp), (b), (i))
 159 typedef int bvec_iterator_t;
 160 #endif
 161
 162 /*
 163  * Portable helper for correctly setting the FAILFAST flags.  The
 164  * correct usage has changed 3 times from 2.6.12 to 2.6.38.
 165  */
 166 static inline void
 167 bio_set_flags_failfast(struct block_device *bdev, int *flags)
 168 {
 169 #ifdef CONFIG_BUG
 170         /*
 171          * Disable FAILFAST for loopback devices because of the
 172          * following incorrect BUG_ON() in loop_make_request().
 173          * This support is also disabled for md devices because the
 174          * test suite layers md devices on top of loopback devices.
 175          * This may be removed when the loopback driver is fixed.
 176          *
 177          *   BUG_ON(!lo || (rw != READ && rw != WRITE));
 178          */
 179         if ((MAJOR(bdev->bd_dev) == LOOP_MAJOR) ||
 180             (MAJOR(bdev->bd_dev) == MD_MAJOR))
 181                 return;
 182
 183 #ifdef BLOCK_EXT_MAJOR
 184         if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
 185                 return;
 186 #endif /* BLOCK_EXT_MAJOR */
 187 #endif /* CONFIG_BUG */
 188
 189 #if defined(HAVE_BIO_RW_FAILFAST_DTD)
 190         /* BIO_RW_FAILFAST_* preferred interface from 2.6.28 - 2.6.35 */
 191         *flags |= (
 192             (1 << BIO_RW_FAILFAST_DEV) |
 193             (1 << BIO_RW_FAILFAST_TRANSPORT) |
 194             (1 << BIO_RW_FAILFAST_DRIVER));
 195 #elif defined(HAVE_REQ_FAILFAST_MASK)
 196         /*
 197          * REQ_FAILFAST_* preferred interface from 2.6.36 - 2.6.xx,
 198          * the BIO_* and REQ_* flags were unified under REQ_* flags.
 199          */
 200         *flags |= REQ_FAILFAST_MASK;
 201 #else
 202 #error "Undefined block IO FAILFAST interface."
 203 #endif
 204 }
 205
 206 /*
 207  * Maximum disk label length, it may be undefined for some kernels.
 208  */
 209 #ifndef DISK_NAME_LEN
 210 #define DISK_NAME_LEN   32
 211 #endif /* DISK_NAME_LEN */
 212
 213 /*
 214  * 4.3 API change
 215  * The bio_endio() prototype changed slightly.  These are helper
 216  * macro's to ensure the prototype and invocation are handled.
 217  */
 218 #ifdef HAVE_1ARG_BIO_END_IO_T
 219 #define BIO_END_IO_PROTO(fn, x, z)      static void fn(struct bio *x)
 220 #define BIO_END_IO(bio, error)          bio->bi_error = error; bio_endio(bio);
 221 #else
 222 #define BIO_END_IO_PROTO(fn, x, z)      static void fn(struct bio *x, int z)
 223 #define BIO_END_IO(bio, error)          bio_endio(bio, error);
 224 #endif /* HAVE_1ARG_BIO_END_IO_T */
 225
 226 /*
 227  * 2.6.38 - 2.6.x API,
 228  *   blkdev_get_by_path()
 229  *   blkdev_put()
 230  *
 231  * 2.6.28 - 2.6.37 API,
 232  *   open_bdev_exclusive()
 233  *   close_bdev_exclusive()
 234  *
 235  * 2.6.12 - 2.6.27 API,
 236  *   open_bdev_excl()
 237  *   close_bdev_excl()
 238  *
 239  * Used to exclusively open a block device from within the kernel.
 240  */
 241 #if defined(HAVE_BLKDEV_GET_BY_PATH)
 242 #define vdev_bdev_open(path, md, hld)   blkdev_get_by_path(path, \
 243                                             (md) | FMODE_EXCL, hld)
 244 #define vdev_bdev_close(bdev, md)       blkdev_put(bdev, (md) | FMODE_EXCL)
 245 #elif defined(HAVE_OPEN_BDEV_EXCLUSIVE)
 246 #define vdev_bdev_open(path, md, hld)   open_bdev_exclusive(path, md, hld)
 247 #define vdev_bdev_close(bdev, md)       close_bdev_exclusive(bdev, md)
 248 #else
 249 #define vdev_bdev_open(path, md, hld)   open_bdev_excl(path, md, hld)
 250 #define vdev_bdev_close(bdev, md)       close_bdev_excl(bdev)
 251 #endif /* HAVE_BLKDEV_GET_BY_PATH | HAVE_OPEN_BDEV_EXCLUSIVE */
 252
 253 /*
 254  * 2.6.22 API change
 255  * The function invalidate_bdev() lost it's second argument because
 256  * it was unused.
 257  */
 258 #ifdef HAVE_1ARG_INVALIDATE_BDEV
 259 #define vdev_bdev_invalidate(bdev)      invalidate_bdev(bdev)
 260 #else
 261 #define vdev_bdev_invalidate(bdev)      invalidate_bdev(bdev, 1)
 262 #endif /* HAVE_1ARG_INVALIDATE_BDEV */
 263
 264 /*
 265  * 2.6.27 API change
 266  * The function was exported for use, prior to this it existed but the
 267  * symbol was not exported.
 268  *
 269  * 4.4.0-6.21 API change for Ubuntu
 270  * lookup_bdev() gained a second argument, FMODE_*, to check inode permissions.
 271  */
 272 #ifdef HAVE_1ARG_LOOKUP_BDEV
 273 #define vdev_lookup_bdev(path)  lookup_bdev(path)
 274 #else
 275 #ifdef HAVE_2ARGS_LOOKUP_BDEV
 276 #define vdev_lookup_bdev(path)  lookup_bdev(path, 0)
 277 #else
 278 #define vdev_lookup_bdev(path)  ERR_PTR(-ENOTSUP)
 279 #endif /* HAVE_2ARGS_LOOKUP_BDEV */
 280 #endif /* HAVE_1ARG_LOOKUP_BDEV */
 281
 282 /*
 283  * 2.6.30 API change
 284  * To ensure good performance preferentially use the physical block size
 285  * for proper alignment.  The physical size is supposed to be the internal
 286  * sector size used by the device.  This is often 4096 byte for AF devices,
 287  * while a smaller 512 byte logical size is supported for compatibility.
 288  *
 289  * Unfortunately, many drives still misreport their physical sector size.
 290  * For devices which are known to lie you may need to manually set this
 291  * at pool creation time with 'zpool create -o ashift=12 ...'.
 292  *
 293  * When the physical block size interface isn't available, we fall back to
 294  * the logical block size interface and then the older hard sector size.
 295  */
 296 #ifdef HAVE_BDEV_PHYSICAL_BLOCK_SIZE
 297 #define vdev_bdev_block_size(bdev)      bdev_physical_block_size(bdev)
 298 #else
 299 #ifdef HAVE_BDEV_LOGICAL_BLOCK_SIZE
 300 #define vdev_bdev_block_size(bdev)      bdev_logical_block_size(bdev)
 301 #else
 302 #define vdev_bdev_block_size(bdev)      bdev_hardsect_size(bdev)
 303 #endif /* HAVE_BDEV_LOGICAL_BLOCK_SIZE */
 304 #endif /* HAVE_BDEV_PHYSICAL_BLOCK_SIZE */
 305
 306 #ifndef HAVE_BIO_SET_OP_ATTRS
 307 /*
 308  * Kernels without bio_set_op_attrs use bi_rw for the bio flags.
 309  */
 310 static inline void
 311 bio_set_op_attrs(struct bio *bio, unsigned rw, unsigned flags)
 312 {
 313         bio->bi_rw |= rw | flags;
 314 }
 315 #endif
 316
 317 /*
 318  * bio_set_flush - Set the appropriate flags in a bio to guarantee
 319  * data are on non-volatile media on completion.
 320  *
 321  * 2.6.X - 2.6.36 API,
 322  *   WRITE_BARRIER - Tells the block layer to commit all previously submitted
 323  *   writes to stable storage before this one is started and that the current
 324  *   write is on stable storage upon completion.  Also prevents reordering
 325  *   on both sides of the current operation.
 326  *
 327  * 2.6.37 - 4.8 API,
 328  *   Introduce  WRITE_FLUSH, WRITE_FUA, and WRITE_FLUSH_FUA flags as a
 329  *   replacement for WRITE_BARRIER to allow expressing richer semantics
 330  *   to the block layer.  It's up to the block layer to implement the
 331  *   semantics correctly. Use the WRITE_FLUSH_FUA flag combination.
 332  *
 333  * 4.8 - 4.9 API,
 334  *   REQ_FLUSH was renamed to REQ_PREFLUSH.  For consistency with previous
 335  *   ZoL releases, prefer the WRITE_FLUSH_FUA flag set if it's available.
 336  *
 337  * 4.10 API,
 338  *   The read/write flags and their modifiers, including WRITE_FLUSH,
 339  *   WRITE_FUA and WRITE_FLUSH_FUA were removed from fs.h in
 340  *   torvalds/linux@70fd7614 and replaced by direct flag modification
 341  *   of the REQ_ flags in bio->bi_opf.  Use REQ_PREFLUSH.
 342  */
 343 static inline void
 344 bio_set_flush(struct bio *bio)
 345 {
 346 #if defined(REQ_PREFLUSH)       /* >= 4.10 */
 347         bio_set_op_attrs(bio, 0, REQ_PREFLUSH);
 348 #elif defined(WRITE_FLUSH_FUA)  /* >= 2.6.37 and <= 4.9 */
 349         bio_set_op_attrs(bio, 0, WRITE_FLUSH_FUA);
 350 #elif defined(WRITE_BARRIER)    /* < 2.6.37 */
 351         bio_set_op_attrs(bio, 0, WRITE_BARRIER);
 352 #else
 353 #error  "Allowing the build will cause bio_set_flush requests to be ignored."
 354 #endif
 355 }
 356
 357 /*
 358  * 4.8 - 4.x API,
 359  *   REQ_OP_FLUSH
 360  *
 361  * 4.8-rc0 - 4.8-rc1,
 362  *   REQ_PREFLUSH
 363  *
 364  * 2.6.36 - 4.7 API,
 365  *   REQ_FLUSH
 366  *
 367  * 2.6.x - 2.6.35 API,
 368  *   HAVE_BIO_RW_BARRIER
 369  *
 370  * Used to determine if a cache flush has been requested.  This check has
 371  * been left intentionally broad in order to cover both a legacy flush
 372  * and the new preflush behavior introduced in Linux 4.8.  This is correct
 373  * in all cases but may have a performance impact for some kernels.  It
 374  * has the advantage of minimizing kernel specific changes in the zvol code.
 375  *
 376  */
 377 static inline boolean_t
 378 bio_is_flush(struct bio *bio)
 379 {
 380 #if defined(HAVE_REQ_OP_FLUSH) && defined(HAVE_BIO_BI_OPF)
 381         return ((bio_op(bio) == REQ_OP_FLUSH) || (bio->bi_opf & REQ_PREFLUSH));
 382 #elif defined(REQ_PREFLUSH) && defined(HAVE_BIO_BI_OPF)
 383         return (bio->bi_opf & REQ_PREFLUSH);
 384 #elif defined(REQ_PREFLUSH) && !defined(HAVE_BIO_BI_OPF)
 385         return (bio->bi_rw & REQ_PREFLUSH);
 386 #elif defined(REQ_FLUSH)
 387         return (bio->bi_rw & REQ_FLUSH);
 388 #elif defined(HAVE_BIO_RW_BARRIER)
 389         return (bio->bi_rw & (1 << BIO_RW_BARRIER));
 390 #else
 391 #error  "Allowing the build will cause flush requests to be ignored."
 392 #endif
 393 }
 394
 395 /*
 396  * 4.8 - 4.x API,
 397  *   REQ_FUA flag moved to bio->bi_opf
 398  *
 399  * 2.6.x - 4.7 API,
 400  *   REQ_FUA
 401  */
 402 static inline boolean_t
 403 bio_is_fua(struct bio *bio)
 404 {
 405 #if defined(HAVE_BIO_BI_OPF)
 406         return (bio->bi_opf & REQ_FUA);
 407 #elif defined(REQ_FUA)
 408         return (bio->bi_rw & REQ_FUA);
 409 #else
 410 #error  "Allowing the build will cause fua requests to be ignored."
 411 #endif
 412 }
 413
 414 /*
 415  * 4.8 - 4.x API,
 416  *   REQ_OP_DISCARD
 417  *
 418  * 2.6.36 - 4.7 API,
 419  *   REQ_DISCARD
 420  *
 421  * 2.6.28 - 2.6.35 API,
 422  *   BIO_RW_DISCARD
 423  *
 424  * In all cases the normal I/O path is used for discards.  The only
 425  * difference is how the kernel tags individual I/Os as discards.
 426  *
 427  * Note that 2.6.32 era kernels provide both BIO_RW_DISCARD and REQ_DISCARD,
 428  * where BIO_RW_DISCARD is the correct interface.  Therefore, it is important
 429  * that the HAVE_BIO_RW_DISCARD check occur before the REQ_DISCARD check.
 430  */
 431 static inline boolean_t
 432 bio_is_discard(struct bio *bio)
 433 {
 434 #if defined(HAVE_REQ_OP_DISCARD)
 435         return (bio_op(bio) == REQ_OP_DISCARD);
 436 #elif defined(HAVE_BIO_RW_DISCARD)
 437         return (bio->bi_rw & (1 << BIO_RW_DISCARD));
 438 #elif defined(REQ_DISCARD)
 439         return (bio->bi_rw & REQ_DISCARD);
 440 #else
 441 /* potentially triggering the DMU_MAX_ACCESS assertion.  */
 442 #error  "Allowing the build will cause discard requests to become writes."
 443 #endif
 444 }
 445
 446 /*
 447  * 4.8 - 4.x API,
 448  *   REQ_OP_SECURE_ERASE
 449  *
 450  * 2.6.36 - 4.7 API,
 451  *   REQ_SECURE
 452  *
 453  * 2.6.x - 2.6.35 API,
 454  *   Unsupported by kernel
 455  */
 456 static inline boolean_t
 457 bio_is_secure_erase(struct bio *bio)
 458 {
 459 #if defined(HAVE_REQ_OP_SECURE_ERASE)
 460         return (bio_op(bio) == REQ_OP_SECURE_ERASE);
 461 #elif defined(REQ_SECURE)
 462         return (bio->bi_rw & REQ_SECURE);
 463 #else
 464         return (0);
 465 #endif
 466 }
 467
 468 /*
 469  * 2.6.33 API change
 470  * Discard granularity and alignment restrictions may now be set.  For
 471  * older kernels which do not support this it is safe to skip it.
 472  */
 473 #ifdef HAVE_DISCARD_GRANULARITY
 474 static inline void
 475 blk_queue_discard_granularity(struct request_queue *q, unsigned int dg)
 476 {
 477         q->limits.discard_granularity = dg;
 478 }
 479 #else
 480 #define blk_queue_discard_granularity(x, dg)    ((void)0)
 481 #endif /* HAVE_DISCARD_GRANULARITY */
 482
 483 /*
 484  * Default Linux IO Scheduler,
 485  * Setting the scheduler to noop will allow the Linux IO scheduler to
 486  * still perform front and back merging, while leaving the request
 487  * ordering and prioritization to the ZFS IO scheduler.
 488  */
 489 #define VDEV_SCHEDULER                  "noop"
 490
 491 /*
 492  * A common holder for vdev_bdev_open() is used to relax the exclusive open
 493  * semantics slightly.  Internal vdev disk callers may pass VDEV_HOLDER to
 494  * allow them to open the device multiple times.  Other kernel callers and
 495  * user space processes which don't pass this value will get EBUSY.  This is
 496  * currently required for the correct operation of hot spares.
 497  */
 498 #define VDEV_HOLDER                     ((void *)0x2401de7)
 499
 500 #ifndef HAVE_GENERIC_IO_ACCT
 501 #define generic_start_io_acct(rw, slen, part)           ((void)0)
 502 #define generic_end_io_acct(rw, part, start_jiffies)    ((void)0)
 503 #endif
 504
 505 #endif /* _ZFS_BLKDEV_H */