drivers/dax/super.c

   1 /*
   2  * Copyright(c) 2017 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  */
  13 #include <linux/pagemap.h>
  14 #include <linux/module.h>
  15 #include <linux/mount.h>
  16 #include <linux/magic.h>
  17 #include <linux/genhd.h>
  18 #include <linux/cdev.h>
  19 #include <linux/hash.h>
  20 #include <linux/slab.h>
  21 #include <linux/uio.h>
  22 #include <linux/dax.h>
  23 #include <linux/fs.h>
  24
  25 static dev_t dax_devt;
  26 DEFINE_STATIC_SRCU(dax_srcu);
  27 static struct vfsmount *dax_mnt;
  28 static DEFINE_IDA(dax_minor_ida);
  29 static struct kmem_cache *dax_cache __read_mostly;
  30 static struct super_block *dax_superblock __read_mostly;
  31
  32 #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
  33 static struct hlist_head dax_host_list[DAX_HASH_SIZE];
  34 static DEFINE_SPINLOCK(dax_host_lock);
  35
  36 int dax_read_lock(void)
  37 {
  38         return srcu_read_lock(&dax_srcu);
  39 }
  40 EXPORT_SYMBOL_GPL(dax_read_lock);
  41
  42 void dax_read_unlock(int id)
  43 {
  44         srcu_read_unlock(&dax_srcu, id);
  45 }
  46 EXPORT_SYMBOL_GPL(dax_read_unlock);
  47
  48 #ifdef CONFIG_BLOCK
  49 #include <linux/blkdev.h>
  50
  51 int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
  52                 pgoff_t *pgoff)
  53 {
  54         phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512;
  55
  56         if (pgoff)
  57                 *pgoff = PHYS_PFN(phys_off);
  58         if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
  59                 return -EINVAL;
  60         return 0;
  61 }
  62 EXPORT_SYMBOL(bdev_dax_pgoff);
  63
  64 #if IS_ENABLED(CONFIG_FS_DAX)
  65 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
  66 {
  67         if (!blk_queue_dax(bdev->bd_queue))
  68                 return NULL;
  69         return fs_dax_get_by_host(bdev->bd_disk->disk_name);
  70 }
  71 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
  72 #endif
  73
  74 /**
  75  * __bdev_dax_supported() - Check if the device supports dax for filesystem
  76  * @sb: The superblock of the device
  77  * @blocksize: The block size of the device
  78  *
  79  * This is a library function for filesystems to check if the block device
  80  * can be mounted with dax option.
  81  *
  82  * Return: negative errno if unsupported, 0 if supported.
  83  */
  84 int __bdev_dax_supported(struct super_block *sb, int blocksize)
  85 {
  86         struct block_device *bdev = sb->s_bdev;
  87         struct dax_device *dax_dev;
  88         pgoff_t pgoff;
  89         int err, id;
  90         void *kaddr;
  91         pfn_t pfn;
  92         long len;
  93
  94         if (blocksize != PAGE_SIZE) {
  95                 pr_debug("VFS (%s): error: unsupported blocksize for dax\n",
  96                                 sb->s_id);
  97                 return -EINVAL;
  98         }
  99
 100         err = bdev_dax_pgoff(bdev, 0, PAGE_SIZE, &pgoff);
 101         if (err) {
 102                 pr_debug("VFS (%s): error: unaligned partition for dax\n",
 103                                 sb->s_id);
 104                 return err;
 105         }
 106
 107         dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
 108         if (!dax_dev) {
 109                 pr_debug("VFS (%s): error: device does not support dax\n",
 110                                 sb->s_id);
 111                 return -EOPNOTSUPP;
 112         }
 113
 114         id = dax_read_lock();
 115         len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
 116         dax_read_unlock(id);
 117
 118         put_dax(dax_dev);
 119
 120         if (len < 1) {
 121                 pr_debug("VFS (%s): error: dax access failed (%ld)\n",
 122                                 sb->s_id, len);
 123                 return len < 0 ? len : -EIO;
 124         }
 125
 126         return 0;
 127 }
 128 EXPORT_SYMBOL_GPL(__bdev_dax_supported);
 129 #endif
 130
 131 enum dax_device_flags {
 132         /* !alive + rcu grace period == no new operations / mappings */
 133         DAXDEV_ALIVE,
 134         /* gate whether dax_flush() calls the low level flush routine */
 135         DAXDEV_WRITE_CACHE,
 136 };
 137
 138 /**
 139  * struct dax_device - anchor object for dax services
 140  * @inode: core vfs
 141  * @cdev: optional character interface for "device dax"
 142  * @host: optional name for lookups where the device path is not available
 143  * @private: dax driver private data
 144  * @flags: state and boolean properties
 145  */
 146 struct dax_device {
 147         struct hlist_node list;
 148         struct inode inode;
 149         struct cdev cdev;
 150         const char *host;
 151         void *private;
 152         unsigned long flags;
 153         const struct dax_operations *ops;
 154 };
 155
 156 static ssize_t write_cache_show(struct device *dev,
 157                 struct device_attribute *attr, char *buf)
 158 {
 159         struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
 160         ssize_t rc;
 161
 162         WARN_ON_ONCE(!dax_dev);
 163         if (!dax_dev)
 164                 return -ENXIO;
 165
 166         rc = sprintf(buf, "%d\n", !!test_bit(DAXDEV_WRITE_CACHE,
 167                                 &dax_dev->flags));
 168         put_dax(dax_dev);
 169         return rc;
 170 }
 171
 172 static ssize_t write_cache_store(struct device *dev,
 173                 struct device_attribute *attr, const char *buf, size_t len)
 174 {
 175         bool write_cache;
 176         int rc = strtobool(buf, &write_cache);
 177         struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
 178
 179         WARN_ON_ONCE(!dax_dev);
 180         if (!dax_dev)
 181                 return -ENXIO;
 182
 183         if (rc)
 184                 len = rc;
 185         else if (write_cache)
 186                 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 187         else
 188                 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 189
 190         put_dax(dax_dev);
 191         return len;
 192 }
 193 static DEVICE_ATTR_RW(write_cache);
 194
 195 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
 196 {
 197         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 198         struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
 199
 200         WARN_ON_ONCE(!dax_dev);
 201         if (!dax_dev)
 202                 return 0;
 203
 204 #ifndef CONFIG_ARCH_HAS_PMEM_API
 205         if (a == &dev_attr_write_cache.attr)
 206                 return 0;
 207 #endif
 208         return a->mode;
 209 }
 210
 211 static struct attribute *dax_attributes[] = {
 212         &dev_attr_write_cache.attr,
 213         NULL,
 214 };
 215
 216 struct attribute_group dax_attribute_group = {
 217         .name = "dax",
 218         .attrs = dax_attributes,
 219         .is_visible = dax_visible,
 220 };
 221 EXPORT_SYMBOL_GPL(dax_attribute_group);
 222
 223 /**
 224  * dax_direct_access() - translate a device pgoff to an absolute pfn
 225  * @dax_dev: a dax_device instance representing the logical memory range
 226  * @pgoff: offset in pages from the start of the device to translate
 227  * @nr_pages: number of consecutive pages caller can handle relative to @pfn
 228  * @kaddr: output parameter that returns a virtual address mapping of pfn
 229  * @pfn: output parameter that returns an absolute pfn translation of @pgoff
 230  *
 231  * Return: negative errno if an error occurs, otherwise the number of
 232  * pages accessible at the device relative @pgoff.
 233  */
 234 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
 235                 void **kaddr, pfn_t *pfn)
 236 {
 237         long avail;
 238
 239         /*
 240          * The device driver is allowed to sleep, in order to make the
 241          * memory directly accessible.
 242          */
 243         might_sleep();
 244
 245         if (!dax_dev)
 246                 return -EOPNOTSUPP;
 247
 248         if (!dax_alive(dax_dev))
 249                 return -ENXIO;
 250
 251         if (nr_pages < 0)
 252                 return nr_pages;
 253
 254         avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
 255                         kaddr, pfn);
 256         if (!avail)
 257                 return -ERANGE;
 258         return min(avail, nr_pages);
 259 }
 260 EXPORT_SYMBOL_GPL(dax_direct_access);
 261
 262 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
 263                 size_t bytes, struct iov_iter *i)
 264 {
 265         if (!dax_alive(dax_dev))
 266                 return 0;
 267
 268         return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
 269 }
 270 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
 271
 272 #ifdef CONFIG_ARCH_HAS_PMEM_API
 273 void arch_wb_cache_pmem(void *addr, size_t size);
 274 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
 275 {
 276         if (unlikely(!test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags)))
 277                 return;
 278
 279         arch_wb_cache_pmem(addr, size);
 280 }
 281 #else
 282 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
 283 {
 284 }
 285 #endif
 286 EXPORT_SYMBOL_GPL(dax_flush);
 287
 288 void dax_write_cache(struct dax_device *dax_dev, bool wc)
 289 {
 290         if (wc)
 291                 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 292         else
 293                 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 294 }
 295 EXPORT_SYMBOL_GPL(dax_write_cache);
 296
 297 bool dax_write_cache_enabled(struct dax_device *dax_dev)
 298 {
 299         return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 300 }
 301 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
 302
 303 bool dax_alive(struct dax_device *dax_dev)
 304 {
 305         lockdep_assert_held(&dax_srcu);
 306         return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
 307 }
 308 EXPORT_SYMBOL_GPL(dax_alive);
 309
 310 static int dax_host_hash(const char *host)
 311 {
 312         return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
 313 }
 314
 315 /*
 316  * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
 317  * that any fault handlers or operations that might have seen
 318  * dax_alive(), have completed.  Any operations that start after
 319  * synchronize_srcu() has run will abort upon seeing !dax_alive().
 320  */
 321 void kill_dax(struct dax_device *dax_dev)
 322 {
 323         if (!dax_dev)
 324                 return;
 325
 326         clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
 327
 328         synchronize_srcu(&dax_srcu);
 329
 330         spin_lock(&dax_host_lock);
 331         hlist_del_init(&dax_dev->list);
 332         spin_unlock(&dax_host_lock);
 333
 334         dax_dev->private = NULL;
 335 }
 336 EXPORT_SYMBOL_GPL(kill_dax);
 337
 338 static struct inode *dax_alloc_inode(struct super_block *sb)
 339 {
 340         struct dax_device *dax_dev;
 341         struct inode *inode;
 342
 343         dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
 344         if (!dax_dev)
 345                 return NULL;
 346
 347         inode = &dax_dev->inode;
 348         inode->i_rdev = 0;
 349         return inode;
 350 }
 351
 352 static struct dax_device *to_dax_dev(struct inode *inode)
 353 {
 354         return container_of(inode, struct dax_device, inode);
 355 }
 356
 357 static void dax_i_callback(struct rcu_head *head)
 358 {
 359         struct inode *inode = container_of(head, struct inode, i_rcu);
 360         struct dax_device *dax_dev = to_dax_dev(inode);
 361
 362         kfree(dax_dev->host);
 363         dax_dev->host = NULL;
 364         if (inode->i_rdev)
 365                 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
 366         kmem_cache_free(dax_cache, dax_dev);
 367 }
 368
 369 static void dax_destroy_inode(struct inode *inode)
 370 {
 371         struct dax_device *dax_dev = to_dax_dev(inode);
 372
 373         WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
 374                         "kill_dax() must be called before final iput()\n");
 375         call_rcu(&inode->i_rcu, dax_i_callback);
 376 }
 377
 378 static const struct super_operations dax_sops = {
 379         .statfs = simple_statfs,
 380         .alloc_inode = dax_alloc_inode,
 381         .destroy_inode = dax_destroy_inode,
 382         .drop_inode = generic_delete_inode,
 383 };
 384
 385 static struct dentry *dax_mount(struct file_system_type *fs_type,
 386                 int flags, const char *dev_name, void *data)
 387 {
 388         return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
 389 }
 390
 391 static struct file_system_type dax_fs_type = {
 392         .name = "dax",
 393         .mount = dax_mount,
 394         .kill_sb = kill_anon_super,
 395 };
 396
 397 static int dax_test(struct inode *inode, void *data)
 398 {
 399         dev_t devt = *(dev_t *) data;
 400
 401         return inode->i_rdev == devt;
 402 }
 403
 404 static int dax_set(struct inode *inode, void *data)
 405 {
 406         dev_t devt = *(dev_t *) data;
 407
 408         inode->i_rdev = devt;
 409         return 0;
 410 }
 411
 412 static struct dax_device *dax_dev_get(dev_t devt)
 413 {
 414         struct dax_device *dax_dev;
 415         struct inode *inode;
 416
 417         inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
 418                         dax_test, dax_set, &devt);
 419
 420         if (!inode)
 421                 return NULL;
 422
 423         dax_dev = to_dax_dev(inode);
 424         if (inode->i_state & I_NEW) {
 425                 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
 426                 inode->i_cdev = &dax_dev->cdev;
 427                 inode->i_mode = S_IFCHR;
 428                 inode->i_flags = S_DAX;
 429                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 430                 unlock_new_inode(inode);
 431         }
 432
 433         return dax_dev;
 434 }
 435
 436 static void dax_add_host(struct dax_device *dax_dev, const char *host)
 437 {
 438         int hash;
 439
 440         /*
 441          * Unconditionally init dax_dev since it's coming from a
 442          * non-zeroed slab cache
 443          */
 444         INIT_HLIST_NODE(&dax_dev->list);
 445         dax_dev->host = host;
 446         if (!host)
 447                 return;
 448
 449         hash = dax_host_hash(host);
 450         spin_lock(&dax_host_lock);
 451         hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
 452         spin_unlock(&dax_host_lock);
 453 }
 454
 455 struct dax_device *alloc_dax(void *private, const char *__host,
 456                 const struct dax_operations *ops)
 457 {
 458         struct dax_device *dax_dev;
 459         const char *host;
 460         dev_t devt;
 461         int minor;
 462
 463         host = kstrdup(__host, GFP_KERNEL);
 464         if (__host && !host)
 465                 return NULL;
 466
 467         minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
 468         if (minor < 0)
 469                 goto err_minor;
 470
 471         devt = MKDEV(MAJOR(dax_devt), minor);
 472         dax_dev = dax_dev_get(devt);
 473         if (!dax_dev)
 474                 goto err_dev;
 475
 476         dax_add_host(dax_dev, host);
 477         dax_dev->ops = ops;
 478         dax_dev->private = private;
 479         return dax_dev;
 480
 481  err_dev:
 482         ida_simple_remove(&dax_minor_ida, minor);
 483  err_minor:
 484         kfree(host);
 485         return NULL;
 486 }
 487 EXPORT_SYMBOL_GPL(alloc_dax);
 488
 489 void put_dax(struct dax_device *dax_dev)
 490 {
 491         if (!dax_dev)
 492                 return;
 493         iput(&dax_dev->inode);
 494 }
 495 EXPORT_SYMBOL_GPL(put_dax);
 496
 497 /**
 498  * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
 499  * @host: alternate name for the device registered by a dax driver
 500  */
 501 struct dax_device *dax_get_by_host(const char *host)
 502 {
 503         struct dax_device *dax_dev, *found = NULL;
 504         int hash, id;
 505
 506         if (!host)
 507                 return NULL;
 508
 509         hash = dax_host_hash(host);
 510
 511         id = dax_read_lock();
 512         spin_lock(&dax_host_lock);
 513         hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
 514                 if (!dax_alive(dax_dev)
 515                                 || strcmp(host, dax_dev->host) != 0)
 516                         continue;
 517
 518                 if (igrab(&dax_dev->inode))
 519                         found = dax_dev;
 520                 break;
 521         }
 522         spin_unlock(&dax_host_lock);
 523         dax_read_unlock(id);
 524
 525         return found;
 526 }
 527 EXPORT_SYMBOL_GPL(dax_get_by_host);
 528
 529 /**
 530  * inode_dax: convert a public inode into its dax_dev
 531  * @inode: An inode with i_cdev pointing to a dax_dev
 532  *
 533  * Note this is not equivalent to to_dax_dev() which is for private
 534  * internal use where we know the inode filesystem type == dax_fs_type.
 535  */
 536 struct dax_device *inode_dax(struct inode *inode)
 537 {
 538         struct cdev *cdev = inode->i_cdev;
 539
 540         return container_of(cdev, struct dax_device, cdev);
 541 }
 542 EXPORT_SYMBOL_GPL(inode_dax);
 543
 544 struct inode *dax_inode(struct dax_device *dax_dev)
 545 {
 546         return &dax_dev->inode;
 547 }
 548 EXPORT_SYMBOL_GPL(dax_inode);
 549
 550 void *dax_get_private(struct dax_device *dax_dev)
 551 {
 552         return dax_dev->private;
 553 }
 554 EXPORT_SYMBOL_GPL(dax_get_private);
 555
 556 static void init_once(void *_dax_dev)
 557 {
 558         struct dax_device *dax_dev = _dax_dev;
 559         struct inode *inode = &dax_dev->inode;
 560
 561         memset(dax_dev, 0, sizeof(*dax_dev));
 562         inode_init_once(inode);
 563 }
 564
 565 static int __dax_fs_init(void)
 566 {
 567         int rc;
 568
 569         dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
 570                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 571                          SLAB_MEM_SPREAD|SLAB_ACCOUNT),
 572                         init_once);
 573         if (!dax_cache)
 574                 return -ENOMEM;
 575
 576         rc = register_filesystem(&dax_fs_type);
 577         if (rc)
 578                 goto err_register_fs;
 579
 580         dax_mnt = kern_mount(&dax_fs_type);
 581         if (IS_ERR(dax_mnt)) {
 582                 rc = PTR_ERR(dax_mnt);
 583                 goto err_mount;
 584         }
 585         dax_superblock = dax_mnt->mnt_sb;
 586
 587         return 0;
 588
 589  err_mount:
 590         unregister_filesystem(&dax_fs_type);
 591  err_register_fs:
 592         kmem_cache_destroy(dax_cache);
 593
 594         return rc;
 595 }
 596
 597 static void __dax_fs_exit(void)
 598 {
 599         kern_unmount(dax_mnt);
 600         unregister_filesystem(&dax_fs_type);
 601         kmem_cache_destroy(dax_cache);
 602 }
 603
 604 static int __init dax_fs_init(void)
 605 {
 606         int rc;
 607
 608         rc = __dax_fs_init();
 609         if (rc)
 610                 return rc;
 611
 612         rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
 613         if (rc)
 614                 __dax_fs_exit();
 615         return rc;
 616 }
 617
 618 static void __exit dax_fs_exit(void)
 619 {
 620         unregister_chrdev_region(dax_devt, MINORMASK+1);
 621         ida_destroy(&dax_minor_ida);
 622         __dax_fs_exit();
 623 }
 624
 625 MODULE_AUTHOR("Intel Corporation");
 626 MODULE_LICENSE("GPL v2");
 627 subsys_initcall(dax_fs_init);
 628 module_exit(dax_fs_exit);