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         if (!dax_dev)
 250                 return -EOPNOTSUPP;
 251
 252         if (!dax_alive(dax_dev))
 253                 return -ENXIO;
 254
 255         if (nr_pages < 0)
 256                 return nr_pages;
 257
 258         avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages,
 259                         kaddr, pfn);
 260         if (!avail)
 261                 return -ERANGE;
 262         return min(avail, nr_pages);
 263 }
 264 EXPORT_SYMBOL_GPL(dax_direct_access);
 265
 266 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr,
 267                 size_t bytes, struct iov_iter *i)
 268 {
 269         if (!dax_alive(dax_dev))
 270                 return 0;
 271
 272         return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i);
 273 }
 274 EXPORT_SYMBOL_GPL(dax_copy_from_iter);
 275
 276 #ifdef CONFIG_ARCH_HAS_PMEM_API
 277 void arch_wb_cache_pmem(void *addr, size_t size);
 278 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
 279 {
 280         if (unlikely(!test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags)))
 281                 return;
 282
 283         arch_wb_cache_pmem(addr, size);
 284 }
 285 #else
 286 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size)
 287 {
 288 }
 289 #endif
 290 EXPORT_SYMBOL_GPL(dax_flush);
 291
 292 void dax_write_cache(struct dax_device *dax_dev, bool wc)
 293 {
 294         if (wc)
 295                 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 296         else
 297                 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 298 }
 299 EXPORT_SYMBOL_GPL(dax_write_cache);
 300
 301 bool dax_write_cache_enabled(struct dax_device *dax_dev)
 302 {
 303         return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags);
 304 }
 305 EXPORT_SYMBOL_GPL(dax_write_cache_enabled);
 306
 307 bool dax_alive(struct dax_device *dax_dev)
 308 {
 309         lockdep_assert_held(&dax_srcu);
 310         return test_bit(DAXDEV_ALIVE, &dax_dev->flags);
 311 }
 312 EXPORT_SYMBOL_GPL(dax_alive);
 313
 314 static int dax_host_hash(const char *host)
 315 {
 316         return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE;
 317 }
 318
 319 /*
 320  * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring
 321  * that any fault handlers or operations that might have seen
 322  * dax_alive(), have completed.  Any operations that start after
 323  * synchronize_srcu() has run will abort upon seeing !dax_alive().
 324  */
 325 void kill_dax(struct dax_device *dax_dev)
 326 {
 327         if (!dax_dev)
 328                 return;
 329
 330         clear_bit(DAXDEV_ALIVE, &dax_dev->flags);
 331
 332         synchronize_srcu(&dax_srcu);
 333
 334         spin_lock(&dax_host_lock);
 335         hlist_del_init(&dax_dev->list);
 336         spin_unlock(&dax_host_lock);
 337
 338         dax_dev->private = NULL;
 339 }
 340 EXPORT_SYMBOL_GPL(kill_dax);
 341
 342 static struct inode *dax_alloc_inode(struct super_block *sb)
 343 {
 344         struct dax_device *dax_dev;
 345         struct inode *inode;
 346
 347         dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL);
 348         if (!dax_dev)
 349                 return NULL;
 350
 351         inode = &dax_dev->inode;
 352         inode->i_rdev = 0;
 353         return inode;
 354 }
 355
 356 static struct dax_device *to_dax_dev(struct inode *inode)
 357 {
 358         return container_of(inode, struct dax_device, inode);
 359 }
 360
 361 static void dax_i_callback(struct rcu_head *head)
 362 {
 363         struct inode *inode = container_of(head, struct inode, i_rcu);
 364         struct dax_device *dax_dev = to_dax_dev(inode);
 365
 366         kfree(dax_dev->host);
 367         dax_dev->host = NULL;
 368         if (inode->i_rdev)
 369                 ida_simple_remove(&dax_minor_ida, MINOR(inode->i_rdev));
 370         kmem_cache_free(dax_cache, dax_dev);
 371 }
 372
 373 static void dax_destroy_inode(struct inode *inode)
 374 {
 375         struct dax_device *dax_dev = to_dax_dev(inode);
 376
 377         WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags),
 378                         "kill_dax() must be called before final iput()\n");
 379         call_rcu(&inode->i_rcu, dax_i_callback);
 380 }
 381
 382 static const struct super_operations dax_sops = {
 383         .statfs = simple_statfs,
 384         .alloc_inode = dax_alloc_inode,
 385         .destroy_inode = dax_destroy_inode,
 386         .drop_inode = generic_delete_inode,
 387 };
 388
 389 static struct dentry *dax_mount(struct file_system_type *fs_type,
 390                 int flags, const char *dev_name, void *data)
 391 {
 392         return mount_pseudo(fs_type, "dax:", &dax_sops, NULL, DAXFS_MAGIC);
 393 }
 394
 395 static struct file_system_type dax_fs_type = {
 396         .name = "dax",
 397         .mount = dax_mount,
 398         .kill_sb = kill_anon_super,
 399 };
 400
 401 static int dax_test(struct inode *inode, void *data)
 402 {
 403         dev_t devt = *(dev_t *) data;
 404
 405         return inode->i_rdev == devt;
 406 }
 407
 408 static int dax_set(struct inode *inode, void *data)
 409 {
 410         dev_t devt = *(dev_t *) data;
 411
 412         inode->i_rdev = devt;
 413         return 0;
 414 }
 415
 416 static struct dax_device *dax_dev_get(dev_t devt)
 417 {
 418         struct dax_device *dax_dev;
 419         struct inode *inode;
 420
 421         inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
 422                         dax_test, dax_set, &devt);
 423
 424         if (!inode)
 425                 return NULL;
 426
 427         dax_dev = to_dax_dev(inode);
 428         if (inode->i_state & I_NEW) {
 429                 set_bit(DAXDEV_ALIVE, &dax_dev->flags);
 430                 inode->i_cdev = &dax_dev->cdev;
 431                 inode->i_mode = S_IFCHR;
 432                 inode->i_flags = S_DAX;
 433                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 434                 unlock_new_inode(inode);
 435         }
 436
 437         return dax_dev;
 438 }
 439
 440 static void dax_add_host(struct dax_device *dax_dev, const char *host)
 441 {
 442         int hash;
 443
 444         /*
 445          * Unconditionally init dax_dev since it's coming from a
 446          * non-zeroed slab cache
 447          */
 448         INIT_HLIST_NODE(&dax_dev->list);
 449         dax_dev->host = host;
 450         if (!host)
 451                 return;
 452
 453         hash = dax_host_hash(host);
 454         spin_lock(&dax_host_lock);
 455         hlist_add_head(&dax_dev->list, &dax_host_list[hash]);
 456         spin_unlock(&dax_host_lock);
 457 }
 458
 459 struct dax_device *alloc_dax(void *private, const char *__host,
 460                 const struct dax_operations *ops)
 461 {
 462         struct dax_device *dax_dev;
 463         const char *host;
 464         dev_t devt;
 465         int minor;
 466
 467         host = kstrdup(__host, GFP_KERNEL);
 468         if (__host && !host)
 469                 return NULL;
 470
 471         minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL);
 472         if (minor < 0)
 473                 goto err_minor;
 474
 475         devt = MKDEV(MAJOR(dax_devt), minor);
 476         dax_dev = dax_dev_get(devt);
 477         if (!dax_dev)
 478                 goto err_dev;
 479
 480         dax_add_host(dax_dev, host);
 481         dax_dev->ops = ops;
 482         dax_dev->private = private;
 483         return dax_dev;
 484
 485  err_dev:
 486         ida_simple_remove(&dax_minor_ida, minor);
 487  err_minor:
 488         kfree(host);
 489         return NULL;
 490 }
 491 EXPORT_SYMBOL_GPL(alloc_dax);
 492
 493 void put_dax(struct dax_device *dax_dev)
 494 {
 495         if (!dax_dev)
 496                 return;
 497         iput(&dax_dev->inode);
 498 }
 499 EXPORT_SYMBOL_GPL(put_dax);
 500
 501 /**
 502  * dax_get_by_host() - temporary lookup mechanism for filesystem-dax
 503  * @host: alternate name for the device registered by a dax driver
 504  */
 505 struct dax_device *dax_get_by_host(const char *host)
 506 {
 507         struct dax_device *dax_dev, *found = NULL;
 508         int hash, id;
 509
 510         if (!host)
 511                 return NULL;
 512
 513         hash = dax_host_hash(host);
 514
 515         id = dax_read_lock();
 516         spin_lock(&dax_host_lock);
 517         hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) {
 518                 if (!dax_alive(dax_dev)
 519                                 || strcmp(host, dax_dev->host) != 0)
 520                         continue;
 521
 522                 if (igrab(&dax_dev->inode))
 523                         found = dax_dev;
 524                 break;
 525         }
 526         spin_unlock(&dax_host_lock);
 527         dax_read_unlock(id);
 528
 529         return found;
 530 }
 531 EXPORT_SYMBOL_GPL(dax_get_by_host);
 532
 533 /**
 534  * inode_dax: convert a public inode into its dax_dev
 535  * @inode: An inode with i_cdev pointing to a dax_dev
 536  *
 537  * Note this is not equivalent to to_dax_dev() which is for private
 538  * internal use where we know the inode filesystem type == dax_fs_type.
 539  */
 540 struct dax_device *inode_dax(struct inode *inode)
 541 {
 542         struct cdev *cdev = inode->i_cdev;
 543
 544         return container_of(cdev, struct dax_device, cdev);
 545 }
 546 EXPORT_SYMBOL_GPL(inode_dax);
 547
 548 struct inode *dax_inode(struct dax_device *dax_dev)
 549 {
 550         return &dax_dev->inode;
 551 }
 552 EXPORT_SYMBOL_GPL(dax_inode);
 553
 554 void *dax_get_private(struct dax_device *dax_dev)
 555 {
 556         return dax_dev->private;
 557 }
 558 EXPORT_SYMBOL_GPL(dax_get_private);
 559
 560 static void init_once(void *_dax_dev)
 561 {
 562         struct dax_device *dax_dev = _dax_dev;
 563         struct inode *inode = &dax_dev->inode;
 564
 565         memset(dax_dev, 0, sizeof(*dax_dev));
 566         inode_init_once(inode);
 567 }
 568
 569 static int __dax_fs_init(void)
 570 {
 571         int rc;
 572
 573         dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0,
 574                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 575                          SLAB_MEM_SPREAD|SLAB_ACCOUNT),
 576                         init_once);
 577         if (!dax_cache)
 578                 return -ENOMEM;
 579
 580         rc = register_filesystem(&dax_fs_type);
 581         if (rc)
 582                 goto err_register_fs;
 583
 584         dax_mnt = kern_mount(&dax_fs_type);
 585         if (IS_ERR(dax_mnt)) {
 586                 rc = PTR_ERR(dax_mnt);
 587                 goto err_mount;
 588         }
 589         dax_superblock = dax_mnt->mnt_sb;
 590
 591         return 0;
 592
 593  err_mount:
 594         unregister_filesystem(&dax_fs_type);
 595  err_register_fs:
 596         kmem_cache_destroy(dax_cache);
 597
 598         return rc;
 599 }
 600
 601 static void __dax_fs_exit(void)
 602 {
 603         kern_unmount(dax_mnt);
 604         unregister_filesystem(&dax_fs_type);
 605         kmem_cache_destroy(dax_cache);
 606 }
 607
 608 static int __init dax_fs_init(void)
 609 {
 610         int rc;
 611
 612         rc = __dax_fs_init();
 613         if (rc)
 614                 return rc;
 615
 616         rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax");
 617         if (rc)
 618                 __dax_fs_exit();
 619         return rc;
 620 }
 621
 622 static void __exit dax_fs_exit(void)
 623 {
 624         unregister_chrdev_region(dax_devt, MINORMASK+1);
 625         ida_destroy(&dax_minor_ida);
 626         __dax_fs_exit();
 627 }
 628
 629 MODULE_AUTHOR("Intel Corporation");
 630 MODULE_LICENSE("GPL v2");
 631 subsys_initcall(dax_fs_init);
 632 module_exit(dax_fs_exit);