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