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