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/pfn_t.h>
  19 #include <linux/cdev.h>
  20 #include <linux/hash.h>
  21 #include <linux/slab.h>
  22 #include <linux/uio.h>
  23 #include <linux/dax.h>
  24 #include <linux/fs.h>
  25
  26 static dev_t dax_devt;
  27 DEFINE_STATIC_SRCU(dax_srcu);
  28 static struct vfsmount *dax_mnt;
  29 static DEFINE_IDA(dax_minor_ida);
  30 static struct kmem_cache *dax_cache __read_mostly;
  31 static struct super_block *dax_superblock __read_mostly;
  32
  33 #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head))
  34 static struct hlist_head dax_host_list[DAX_HASH_SIZE];
  35 static DEFINE_SPINLOCK(dax_host_lock);
  36
  37 int dax_read_lock(void)
  38 {
  39         return srcu_read_lock(&dax_srcu);
  40 }
  41 EXPORT_SYMBOL_GPL(dax_read_lock);
  42
  43 void dax_read_unlock(int id)
  44 {
  45         srcu_read_unlock(&dax_srcu, id);
  46 }
  47 EXPORT_SYMBOL_GPL(dax_read_unlock);
  48
  49 #ifdef CONFIG_BLOCK
  50 #include <linux/blkdev.h>
  51
  52 int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size,
  53                 pgoff_t *pgoff)
  54 {
  55         phys_addr_t phys_off = (get_start_sect(bdev) + sector) * 512;
  56
  57         if (pgoff)
  58                 *pgoff = PHYS_PFN(phys_off);
  59         if (phys_off % PAGE_SIZE || size % PAGE_SIZE)
  60                 return -EINVAL;
  61         return 0;
  62 }
  63 EXPORT_SYMBOL(bdev_dax_pgoff);
  64
  65 #if IS_ENABLED(CONFIG_FS_DAX)
  66 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev)
  67 {
  68         if (!blk_queue_dax(bdev->bd_queue))
  69                 return NULL;
  70         return fs_dax_get_by_host(bdev->bd_disk->disk_name);
  71 }
  72 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev);
  73 #endif
  74
  75 /**
  76  * __bdev_dax_supported() - Check if the device supports dax for filesystem
  77  * @sb: The superblock of the device
  78  * @blocksize: The block size of the device
  79  *
  80  * This is a library function for filesystems to check if the block device
  81  * can be mounted with dax option.
  82  *
  83  * Return: negative errno if unsupported, 0 if supported.
  84  */
  85 int __bdev_dax_supported(struct super_block *sb, int blocksize)
  86 {
  87         struct block_device *bdev = sb->s_bdev;
  88         struct dax_device *dax_dev;
  89         pgoff_t pgoff;
  90         int err, id;
  91         void *kaddr;
  92         pfn_t pfn;
  93         long len;
  94
  95         if (blocksize != PAGE_SIZE) {
  96                 pr_debug("VFS (%s): error: unsupported blocksize for dax\n",
  97                                 sb->s_id);
  98                 return -EINVAL;
  99         }
 100
 101         err = bdev_dax_pgoff(bdev, 0, PAGE_SIZE, &pgoff);
 102         if (err) {
 103                 pr_debug("VFS (%s): error: unaligned partition for dax\n",
 104                                 sb->s_id);
 105                 return err;
 106         }
 107
 108         dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
 109         if (!dax_dev) {
 110                 pr_debug("VFS (%s): error: device does not support dax\n",
 111                                 sb->s_id);
 112                 return -EOPNOTSUPP;
 113         }
 114
 115         id = dax_read_lock();
 116         len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn);
 117         dax_read_unlock(id);
 118
 119         put_dax(dax_dev);
 120
 121         if (len < 1) {
 122                 pr_debug("VFS (%s): error: dax access failed (%ld)\n",
 123                                 sb->s_id, len);
 124                 return len < 0 ? len : -EIO;
 125         }
 126
 127         if ((IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn))
 128                         || pfn_t_devmap(pfn))
 129                 /* pass */;
 130         else {
 131                 pr_debug("VFS (%s): error: dax support not enabled\n",
 132                                 sb->s_id);
 133                 return -EOPNOTSUPP;
 134         }
 135
 136         return 0;
 137 }
 138 EXPORT_SYMBOL_GPL(__bdev_dax_supported);
 139 #endif
 140
 141 enum dax_device_flags {
 142         /* !alive + rcu grace period == no new operations / mappings */
 143         DAXDEV_ALIVE,
 144         /* gate whether dax_flush() calls the low level flush routine */
 145         DAXDEV_WRITE_CACHE,
 146 };
 147
 148 /**
 149  * struct dax_device - anchor object for dax services
 150  * @inode: core vfs
 151  * @cdev: optional character interface for "device dax"
 152  * @host: optional name for lookups where the device path is not available
 153  * @private: dax driver private data
 154  * @flags: state and boolean properties
 155  */
 156 struct dax_device {
 157         struct hlist_node list;
 158         struct inode inode;
 159         struct cdev cdev;
 160         const char *host;
 161         void *private;
 162         unsigned long flags;
 163         const struct dax_operations *ops;
 164 };
 165
 166 static ssize_t write_cache_show(struct device *dev,
 167                 struct device_attribute *attr, char *buf)
 168 {
 169         struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
 170         ssize_t rc;
 171
 172         WARN_ON_ONCE(!dax_dev);
 173         if (!dax_dev)
 174                 return -ENXIO;
 175
 176         rc = sprintf(buf, "%d\n", !!test_bit(DAXDEV_WRITE_CACHE,
 177                                 &dax_dev->flags));
 178         put_dax(dax_dev);
 179         return rc;
 180 }
 181
 182 static ssize_t write_cache_store(struct device *dev,
 183                 struct device_attribute *attr, const char *buf, size_t len)
 184 {
 185         bool write_cache;
 186         int rc = strtobool(buf, &write_cache);
 187         struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
 188
 189         WARN_ON_ONCE(!dax_dev);
 190         if (!dax_dev)
 191                 return -ENXIO;
 192
 193         if (rc)
 194                 len = rc;
 195         else if (write_cache)
 196                 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 197         else
 198                 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 199
 200         put_dax(dax_dev);
 201         return len;
 202 }
 203 static DEVICE_ATTR_RW(write_cache);
 204
 205 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
 206 {
 207         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 208         struct dax_device *dax_dev = dax_get_by_host(dev_name(dev));
 209
 210         WARN_ON_ONCE(!dax_dev);
 211         if (!dax_dev)
 212                 return 0;
 213
 214 #ifndef CONFIG_ARCH_HAS_PMEM_API
 215         if (a == &dev_attr_write_cache.attr)
 216                 return 0;
 217 #endif
 218         return a->mode;
 219 }
 220
 221 static struct attribute *dax_attributes[] = {
 222         &dev_attr_write_cache.attr,
 223         NULL,
 224 };
 225
 226 struct attribute_group dax_attribute_group = {
 227         .name = "dax",
 228         .attrs = dax_attributes,
 229         .is_visible = dax_visible,
 230 };
 231 EXPORT_SYMBOL_GPL(dax_attribute_group);
 232
 233 /**
 234  * dax_direct_access() - translate a device pgoff to an absolute pfn
 235  * @dax_dev: a dax_device instance representing the logical memory range
 236  * @pgoff: offset in pages from the start of the device to translate
 237  * @nr_pages: number of consecutive pages caller can handle relative to @pfn
 238  * @kaddr: output parameter that returns a virtual address mapping of pfn
 239  * @pfn: output parameter that returns an absolute pfn translation of @pgoff
 240  *
 241  * Return: negative errno if an error occurs, otherwise the number of
 242  * pages accessible at the device relative @pgoff.
 243  */
 244 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages,
 245                 void **kaddr, pfn_t *pfn)
 246 {
 247         long avail;
 248
 249         /*
 250          * The device driver is allowed to sleep, in order to make the
 251          * memory directly accessible.
 252          */
 253         might_sleep();
 254
 255         if (!dax_dev)
 256                 return -EOPNOTSUPP;
 257
 258         if (!dax_alive(dax_dev))
 259                 return -ENXIO;
 260
 261         if (nr_pages < 0)
 262                 return nr_pages;
 263
 264         avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
 265                         kaddr, pfn);
 266         if (!avail)
 267                 return -ERANGE;
 268         return min(avail, nr_pages);
 269 }
 270 EXPORT_SYMBOL_GPL(dax_direct_access);
 271
 272 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
 273                 size_t bytes, struct iov_iter *i)
 274 {
 275         if (!dax_alive(dax_dev))
 276                 return 0;
 277
 278         return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
 279 }
 280 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
 281
 282 #ifdef CONFIG_ARCH_HAS_PMEM_API
 283 void arch_wb_cache_pmem(void *addr, size_t size);
 284 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
 285 {
 286         if (unlikely(!test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags)))
 287                 return;
 288
 289         arch_wb_cache_pmem(addr, size);
 290 }
 291 #else
 292 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
 293 {
 294 }
 295 #endif
 296 EXPORT_SYMBOL_GPL(dax_flush);
 297
 298 void dax_write_cache(struct dax_device *dax_dev, bool wc)
 299 {
 300         if (wc)
 301                 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 302         else
 303                 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 304 }
 305 EXPORT_SYMBOL_GPL(dax_write_cache);
 306
 307 bool dax_write_cache_enabled(struct dax_device *dax_dev)
 308 {
 309         return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 310 }
 311 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
 312
 313 bool dax_alive(struct dax_device *dax_dev)
 314 {
 315         lockdep_assert_held(&dax_srcu);
 316         return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
 317 }
 318 EXPORT_SYMBOL_GPL(dax_alive);
 319
 320 static int dax_host_hash(const char *host)
 321 {
 322         return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
 323 }
 324
 325 /*
 326  * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
 327  * that any fault handlers or operations that might have seen
 328  * dax_alive(), have completed.  Any operations that start after
 329  * synchronize_srcu() has run will abort upon seeing !dax_alive().
 330  */
 331 void kill_dax(struct dax_device *dax_dev)
 332 {
 333         if (!dax_dev)
 334                 return;
 335
 336         clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
 337
 338         synchronize_srcu(&dax_srcu);
 339
 340         spin_lock(&dax_host_lock);
 341         hlist_del_init(&dax_dev->list);
 342         spin_unlock(&dax_host_lock);
 343
 344         dax_dev->private = NULL;
 345 }
 346 EXPORT_SYMBOL_GPL(kill_dax);
 347
 348 static struct inode *dax_alloc_inode(struct super_block *sb)
 349 {
 350         struct dax_device *dax_dev;
 351         struct inode *inode;
 352
 353         dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
 354         if (!dax_dev)
 355                 return NULL;
 356
 357         inode = &dax_dev->inode;
 358         inode->i_rdev = 0;
 359         return inode;
 360 }
 361
 362 static struct dax_device *to_dax_dev(struct inode *inode)
 363 {
 364         return container_of(inode, struct dax_device, inode);
 365 }
 366
 367 static void dax_i_callback(struct rcu_head *head)
 368 {
 369         struct inode *inode = container_of(head, struct inode, i_rcu);
 370         struct dax_device *dax_dev = to_dax_dev(inode);
 371
 372         kfree(dax_dev->host);
 373         dax_dev->host = NULL;
 374         if (inode->i_rdev)
 375                 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
 376         kmem_cache_free(dax_cache, dax_dev);
 377 }
 378
 379 static void dax_destroy_inode(struct inode *inode)
 380 {
 381         struct dax_device *dax_dev = to_dax_dev(inode);
 382
 383         WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
 384                         "kill_dax() must be called before final iput()\n");
 385         call_rcu(&inode->i_rcu, dax_i_callback);
 386 }
 387
 388 static const struct super_operations dax_sops = {
 389         .statfs = simple_statfs,
 390         .alloc_inode = dax_alloc_inode,
 391         .destroy_inode = dax_destroy_inode,
 392         .drop_inode = generic_delete_inode,
 393 };
 394
 395 static struct dentry *dax_mount(struct file_system_type *fs_type,
 396                 int flags, const char *dev_name, void *data)
 397 {
 398         return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
 399 }
 400
 401 static struct file_system_type dax_fs_type = {
 402         .name = "dax",
 403         .mount = dax_mount,
 404         .kill_sb = kill_anon_super,
 405 };
 406
 407 static int dax_test(struct inode *inode, void *data)
 408 {
 409         dev_t devt = *(dev_t *) data;
 410
 411         return inode->i_rdev == devt;
 412 }
 413
 414 static int dax_set(struct inode *inode, void *data)
 415 {
 416         dev_t devt = *(dev_t *) data;
 417
 418         inode->i_rdev = devt;
 419         return 0;
 420 }
 421
 422 static struct dax_device *dax_dev_get(dev_t devt)
 423 {
 424         struct dax_device *dax_dev;
 425         struct inode *inode;
 426
 427         inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
 428                         dax_test, dax_set, &devt);
 429
 430         if (!inode)
 431                 return NULL;
 432
 433         dax_dev = to_dax_dev(inode);
 434         if (inode->i_state & I_NEW) {
 435                 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
 436                 inode->i_cdev = &dax_dev->cdev;
 437                 inode->i_mode = S_IFCHR;
 438                 inode->i_flags = S_DAX;
 439                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 440                 unlock_new_inode(inode);
 441         }
 442
 443         return dax_dev;
 444 }
 445
 446 static void dax_add_host(struct dax_device *dax_dev, const char *host)
 447 {
 448         int hash;
 449
 450         /*
 451          * Unconditionally init dax_dev since it's coming from a
 452          * non-zeroed slab cache
 453          */
 454         INIT_HLIST_NODE(&dax_dev->list);
 455         dax_dev->host = host;
 456         if (!host)
 457                 return;
 458
 459         hash = dax_host_hash(host);
 460         spin_lock(&dax_host_lock);
 461         hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
 462         spin_unlock(&dax_host_lock);
 463 }
 464
 465 struct dax_device *alloc_dax(void *private, const char *__host,
 466                 const struct dax_operations *ops)
 467 {
 468         struct dax_device *dax_dev;
 469         const char *host;
 470         dev_t devt;
 471         int minor;
 472
 473         host = kstrdup(__host, GFP_KERNEL);
 474         if (__host && !host)
 475                 return NULL;
 476
 477         minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
 478         if (minor < 0)
 479                 goto err_minor;
 480
 481         devt = MKDEV(MAJOR(dax_devt), minor);
 482         dax_dev = dax_dev_get(devt);
 483         if (!dax_dev)
 484                 goto err_dev;
 485
 486         dax_add_host(dax_dev, host);
 487         dax_dev->ops = ops;
 488         dax_dev->private = private;
 489         return dax_dev;
 490
 491  err_dev:
 492         ida_simple_remove(&dax_minor_ida, minor);
 493  err_minor:
 494         kfree(host);
 495         return NULL;
 496 }
 497 EXPORT_SYMBOL_GPL(alloc_dax);
 498
 499 void put_dax(struct dax_device *dax_dev)
 500 {
 501         if (!dax_dev)
 502                 return;
 503         iput(&dax_dev->inode);
 504 }
 505 EXPORT_SYMBOL_GPL(put_dax);
 506
 507 /**
 508  * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
 509  * @host: alternate name for the device registered by a dax driver
 510  */
 511 struct dax_device *dax_get_by_host(const char *host)
 512 {
 513         struct dax_device *dax_dev, *found = NULL;
 514         int hash, id;
 515
 516         if (!host)
 517                 return NULL;
 518
 519         hash = dax_host_hash(host);
 520
 521         id = dax_read_lock();
 522         spin_lock(&dax_host_lock);
 523         hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
 524                 if (!dax_alive(dax_dev)
 525                                 || strcmp(host, dax_dev->host) != 0)
 526                         continue;
 527
 528                 if (igrab(&dax_dev->inode))
 529                         found = dax_dev;
 530                 break;
 531         }
 532         spin_unlock(&dax_host_lock);
 533         dax_read_unlock(id);
 534
 535         return found;
 536 }
 537 EXPORT_SYMBOL_GPL(dax_get_by_host);
 538
 539 /**
 540  * inode_dax: convert a public inode into its dax_dev
 541  * @inode: An inode with i_cdev pointing to a dax_dev
 542  *
 543  * Note this is not equivalent to to_dax_dev() which is for private
 544  * internal use where we know the inode filesystem type == dax_fs_type.
 545  */
 546 struct dax_device *inode_dax(struct inode *inode)
 547 {
 548         struct cdev *cdev = inode->i_cdev;
 549
 550         return container_of(cdev, struct dax_device, cdev);
 551 }
 552 EXPORT_SYMBOL_GPL(inode_dax);
 553
 554 struct inode *dax_inode(struct dax_device *dax_dev)
 555 {
 556         return &dax_dev->inode;
 557 }
 558 EXPORT_SYMBOL_GPL(dax_inode);
 559
 560 void *dax_get_private(struct dax_device *dax_dev)
 561 {
 562         return dax_dev->private;
 563 }
 564 EXPORT_SYMBOL_GPL(dax_get_private);
 565
 566 static void init_once(void *_dax_dev)
 567 {
 568         struct dax_device *dax_dev = _dax_dev;
 569         struct inode *inode = &dax_dev->inode;
 570
 571         memset(dax_dev, 0, sizeof(*dax_dev));
 572         inode_init_once(inode);
 573 }
 574
 575 static int __dax_fs_init(void)
 576 {
 577         int rc;
 578
 579         dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
 580                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 581                          SLAB_MEM_SPREAD|SLAB_ACCOUNT),
 582                         init_once);
 583         if (!dax_cache)
 584                 return -ENOMEM;
 585
 586         rc = register_filesystem(&dax_fs_type);
 587         if (rc)
 588                 goto err_register_fs;
 589
 590         dax_mnt = kern_mount(&dax_fs_type);
 591         if (IS_ERR(dax_mnt)) {
 592                 rc = PTR_ERR(dax_mnt);
 593                 goto err_mount;
 594         }
 595         dax_superblock = dax_mnt->mnt_sb;
 596
 597         return 0;
 598
 599  err_mount:
 600         unregister_filesystem(&dax_fs_type);
 601  err_register_fs:
 602         kmem_cache_destroy(dax_cache);
 603
 604         return rc;
 605 }
 606
 607 static void __dax_fs_exit(void)
 608 {
 609         kern_unmount(dax_mnt);
 610         unregister_filesystem(&dax_fs_type);
 611         kmem_cache_destroy(dax_cache);
 612 }
 613
 614 static int __init dax_fs_init(void)
 615 {
 616         int rc;
 617
 618         rc = __dax_fs_init();
 619         if (rc)
 620                 return rc;
 621
 622         rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
 623         if (rc)
 624                 __dax_fs_exit();
 625         return rc;
 626 }
 627
 628 static void __exit dax_fs_exit(void)
 629 {
 630         unregister_chrdev_region(dax_devt, MINORMASK+1);
 631         ida_destroy(&dax_minor_ida);
 632         __dax_fs_exit();
 633 }
 634
 635 MODULE_AUTHOR("Intel Corporation");
 636 MODULE_LICENSE("GPL v2");
 637 subsys_initcall(dax_fs_init);
 638 module_exit(dax_fs_exit);