1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright(c) 2017 Intel Corporation. All rights reserved. 4 */ 5 #include <linux/pagemap.h> 6 #include <linux/module.h> 7 #include <linux/mount.h> 8 #include <linux/pseudo_fs.h> 9 #include <linux/magic.h> 10 #include <linux/pfn_t.h> 11 #include <linux/cdev.h> 12 #include <linux/slab.h> 13 #include <linux/uio.h> 14 #include <linux/dax.h> 15 #include <linux/fs.h> 16 #include "dax-private.h" 17 18 /** 19 * struct dax_device - anchor object for dax services 20 * @inode: core vfs 21 * @cdev: optional character interface for "device dax" 22 * @private: dax driver private data 23 * @flags: state and boolean properties 24 */ 25 struct dax_device { 26 struct inode inode; 27 struct cdev cdev; 28 void *private; 29 unsigned long flags; 30 const struct dax_operations *ops; 31 }; 32 33 static dev_t dax_devt; 34 DEFINE_STATIC_SRCU(dax_srcu); 35 static struct vfsmount *dax_mnt; 36 static DEFINE_IDA(dax_minor_ida); 37 static struct kmem_cache *dax_cache __read_mostly; 38 static struct super_block *dax_superblock __read_mostly; 39 40 int dax_read_lock(void) 41 { 42 return srcu_read_lock(&dax_srcu); 43 } 44 EXPORT_SYMBOL_GPL(dax_read_lock); 45 46 void dax_read_unlock(int id) 47 { 48 srcu_read_unlock(&dax_srcu, id); 49 } 50 EXPORT_SYMBOL_GPL(dax_read_unlock); 51 52 #if defined(CONFIG_BLOCK) && defined(CONFIG_FS_DAX) 53 #include <linux/blkdev.h> 54 55 static DEFINE_XARRAY(dax_hosts); 56 57 int dax_add_host(struct dax_device *dax_dev, struct gendisk *disk) 58 { 59 return xa_insert(&dax_hosts, (unsigned long)disk, dax_dev, GFP_KERNEL); 60 } 61 EXPORT_SYMBOL_GPL(dax_add_host); 62 63 void dax_remove_host(struct gendisk *disk) 64 { 65 xa_erase(&dax_hosts, (unsigned long)disk); 66 } 67 EXPORT_SYMBOL_GPL(dax_remove_host); 68 69 /** 70 * fs_dax_get_by_bdev() - temporary lookup mechanism for filesystem-dax 71 * @bdev: block device to find a dax_device for 72 * @start_off: returns the byte offset into the dax_device that @bdev starts 73 */ 74 struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev, u64 *start_off) 75 { 76 struct dax_device *dax_dev; 77 u64 part_size; 78 int id; 79 80 if (!blk_queue_dax(bdev->bd_disk->queue)) 81 return NULL; 82 83 *start_off = get_start_sect(bdev) * SECTOR_SIZE; 84 part_size = bdev_nr_sectors(bdev) * SECTOR_SIZE; 85 if (*start_off % PAGE_SIZE || part_size % PAGE_SIZE) { 86 pr_info("%pg: error: unaligned partition for dax\n", bdev); 87 return NULL; 88 } 89 90 id = dax_read_lock(); 91 dax_dev = xa_load(&dax_hosts, (unsigned long)bdev->bd_disk); 92 if (!dax_dev || !dax_alive(dax_dev) || !igrab(&dax_dev->inode)) 93 dax_dev = NULL; 94 dax_read_unlock(id); 95 96 return dax_dev; 97 } 98 EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev); 99 #endif /* CONFIG_BLOCK && CONFIG_FS_DAX */ 100 101 enum dax_device_flags { 102 /* !alive + rcu grace period == no new operations / mappings */ 103 DAXDEV_ALIVE, 104 /* gate whether dax_flush() calls the low level flush routine */ 105 DAXDEV_WRITE_CACHE, 106 /* flag to check if device supports synchronous flush */ 107 DAXDEV_SYNC, 108 }; 109 110 /** 111 * dax_direct_access() - translate a device pgoff to an absolute pfn 112 * @dax_dev: a dax_device instance representing the logical memory range 113 * @pgoff: offset in pages from the start of the device to translate 114 * @nr_pages: number of consecutive pages caller can handle relative to @pfn 115 * @kaddr: output parameter that returns a virtual address mapping of pfn 116 * @pfn: output parameter that returns an absolute pfn translation of @pgoff 117 * 118 * Return: negative errno if an error occurs, otherwise the number of 119 * pages accessible at the device relative @pgoff. 120 */ 121 long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, 122 void **kaddr, pfn_t *pfn) 123 { 124 long avail; 125 126 if (!dax_dev) 127 return -EOPNOTSUPP; 128 129 if (!dax_alive(dax_dev)) 130 return -ENXIO; 131 132 if (nr_pages < 0) 133 return -EINVAL; 134 135 avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages, 136 kaddr, pfn); 137 if (!avail) 138 return -ERANGE; 139 return min(avail, nr_pages); 140 } 141 EXPORT_SYMBOL_GPL(dax_direct_access); 142 143 size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, 144 size_t bytes, struct iov_iter *i) 145 { 146 if (!dax_alive(dax_dev)) 147 return 0; 148 149 return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i); 150 } 151 EXPORT_SYMBOL_GPL(dax_copy_from_iter); 152 153 size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, 154 size_t bytes, struct iov_iter *i) 155 { 156 if (!dax_alive(dax_dev)) 157 return 0; 158 159 return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i); 160 } 161 EXPORT_SYMBOL_GPL(dax_copy_to_iter); 162 163 int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, 164 size_t nr_pages) 165 { 166 if (!dax_alive(dax_dev)) 167 return -ENXIO; 168 /* 169 * There are no callers that want to zero more than one page as of now. 170 * Once users are there, this check can be removed after the 171 * device mapper code has been updated to split ranges across targets. 172 */ 173 if (nr_pages != 1) 174 return -EIO; 175 176 return dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages); 177 } 178 EXPORT_SYMBOL_GPL(dax_zero_page_range); 179 180 #ifdef CONFIG_ARCH_HAS_PMEM_API 181 void arch_wb_cache_pmem(void *addr, size_t size); 182 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size) 183 { 184 if (unlikely(!dax_write_cache_enabled(dax_dev))) 185 return; 186 187 arch_wb_cache_pmem(addr, size); 188 } 189 #else 190 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size) 191 { 192 } 193 #endif 194 EXPORT_SYMBOL_GPL(dax_flush); 195 196 void dax_write_cache(struct dax_device *dax_dev, bool wc) 197 { 198 if (wc) 199 set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); 200 else 201 clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); 202 } 203 EXPORT_SYMBOL_GPL(dax_write_cache); 204 205 bool dax_write_cache_enabled(struct dax_device *dax_dev) 206 { 207 return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); 208 } 209 EXPORT_SYMBOL_GPL(dax_write_cache_enabled); 210 211 bool dax_synchronous(struct dax_device *dax_dev) 212 { 213 return test_bit(DAXDEV_SYNC, &dax_dev->flags); 214 } 215 EXPORT_SYMBOL_GPL(dax_synchronous); 216 217 void set_dax_synchronous(struct dax_device *dax_dev) 218 { 219 set_bit(DAXDEV_SYNC, &dax_dev->flags); 220 } 221 EXPORT_SYMBOL_GPL(set_dax_synchronous); 222 223 bool dax_alive(struct dax_device *dax_dev) 224 { 225 lockdep_assert_held(&dax_srcu); 226 return test_bit(DAXDEV_ALIVE, &dax_dev->flags); 227 } 228 EXPORT_SYMBOL_GPL(dax_alive); 229 230 /* 231 * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring 232 * that any fault handlers or operations that might have seen 233 * dax_alive(), have completed. Any operations that start after 234 * synchronize_srcu() has run will abort upon seeing !dax_alive(). 235 */ 236 void kill_dax(struct dax_device *dax_dev) 237 { 238 if (!dax_dev) 239 return; 240 241 clear_bit(DAXDEV_ALIVE, &dax_dev->flags); 242 synchronize_srcu(&dax_srcu); 243 } 244 EXPORT_SYMBOL_GPL(kill_dax); 245 246 void run_dax(struct dax_device *dax_dev) 247 { 248 set_bit(DAXDEV_ALIVE, &dax_dev->flags); 249 } 250 EXPORT_SYMBOL_GPL(run_dax); 251 252 static struct inode *dax_alloc_inode(struct super_block *sb) 253 { 254 struct dax_device *dax_dev; 255 struct inode *inode; 256 257 dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL); 258 if (!dax_dev) 259 return NULL; 260 261 inode = &dax_dev->inode; 262 inode->i_rdev = 0; 263 return inode; 264 } 265 266 static struct dax_device *to_dax_dev(struct inode *inode) 267 { 268 return container_of(inode, struct dax_device, inode); 269 } 270 271 static void dax_free_inode(struct inode *inode) 272 { 273 struct dax_device *dax_dev = to_dax_dev(inode); 274 if (inode->i_rdev) 275 ida_simple_remove(&dax_minor_ida, iminor(inode)); 276 kmem_cache_free(dax_cache, dax_dev); 277 } 278 279 static void dax_destroy_inode(struct inode *inode) 280 { 281 struct dax_device *dax_dev = to_dax_dev(inode); 282 WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags), 283 "kill_dax() must be called before final iput()\n"); 284 } 285 286 static const struct super_operations dax_sops = { 287 .statfs = simple_statfs, 288 .alloc_inode = dax_alloc_inode, 289 .destroy_inode = dax_destroy_inode, 290 .free_inode = dax_free_inode, 291 .drop_inode = generic_delete_inode, 292 }; 293 294 static int dax_init_fs_context(struct fs_context *fc) 295 { 296 struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC); 297 if (!ctx) 298 return -ENOMEM; 299 ctx->ops = &dax_sops; 300 return 0; 301 } 302 303 static struct file_system_type dax_fs_type = { 304 .name = "dax", 305 .init_fs_context = dax_init_fs_context, 306 .kill_sb = kill_anon_super, 307 }; 308 309 static int dax_test(struct inode *inode, void *data) 310 { 311 dev_t devt = *(dev_t *) data; 312 313 return inode->i_rdev == devt; 314 } 315 316 static int dax_set(struct inode *inode, void *data) 317 { 318 dev_t devt = *(dev_t *) data; 319 320 inode->i_rdev = devt; 321 return 0; 322 } 323 324 static struct dax_device *dax_dev_get(dev_t devt) 325 { 326 struct dax_device *dax_dev; 327 struct inode *inode; 328 329 inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31), 330 dax_test, dax_set, &devt); 331 332 if (!inode) 333 return NULL; 334 335 dax_dev = to_dax_dev(inode); 336 if (inode->i_state & I_NEW) { 337 set_bit(DAXDEV_ALIVE, &dax_dev->flags); 338 inode->i_cdev = &dax_dev->cdev; 339 inode->i_mode = S_IFCHR; 340 inode->i_flags = S_DAX; 341 mapping_set_gfp_mask(&inode->i_data, GFP_USER); 342 unlock_new_inode(inode); 343 } 344 345 return dax_dev; 346 } 347 348 struct dax_device *alloc_dax(void *private, const struct dax_operations *ops, 349 unsigned long flags) 350 { 351 struct dax_device *dax_dev; 352 dev_t devt; 353 int minor; 354 355 if (WARN_ON_ONCE(ops && !ops->zero_page_range)) 356 return ERR_PTR(-EINVAL); 357 358 minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL); 359 if (minor < 0) 360 return ERR_PTR(-ENOMEM); 361 362 devt = MKDEV(MAJOR(dax_devt), minor); 363 dax_dev = dax_dev_get(devt); 364 if (!dax_dev) 365 goto err_dev; 366 367 dax_dev->ops = ops; 368 dax_dev->private = private; 369 if (flags & DAXDEV_F_SYNC) 370 set_dax_synchronous(dax_dev); 371 372 return dax_dev; 373 374 err_dev: 375 ida_simple_remove(&dax_minor_ida, minor); 376 return ERR_PTR(-ENOMEM); 377 } 378 EXPORT_SYMBOL_GPL(alloc_dax); 379 380 void put_dax(struct dax_device *dax_dev) 381 { 382 if (!dax_dev) 383 return; 384 iput(&dax_dev->inode); 385 } 386 EXPORT_SYMBOL_GPL(put_dax); 387 388 /** 389 * inode_dax: convert a public inode into its dax_dev 390 * @inode: An inode with i_cdev pointing to a dax_dev 391 * 392 * Note this is not equivalent to to_dax_dev() which is for private 393 * internal use where we know the inode filesystem type == dax_fs_type. 394 */ 395 struct dax_device *inode_dax(struct inode *inode) 396 { 397 struct cdev *cdev = inode->i_cdev; 398 399 return container_of(cdev, struct dax_device, cdev); 400 } 401 EXPORT_SYMBOL_GPL(inode_dax); 402 403 struct inode *dax_inode(struct dax_device *dax_dev) 404 { 405 return &dax_dev->inode; 406 } 407 EXPORT_SYMBOL_GPL(dax_inode); 408 409 void *dax_get_private(struct dax_device *dax_dev) 410 { 411 if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags)) 412 return NULL; 413 return dax_dev->private; 414 } 415 EXPORT_SYMBOL_GPL(dax_get_private); 416 417 static void init_once(void *_dax_dev) 418 { 419 struct dax_device *dax_dev = _dax_dev; 420 struct inode *inode = &dax_dev->inode; 421 422 memset(dax_dev, 0, sizeof(*dax_dev)); 423 inode_init_once(inode); 424 } 425 426 static int dax_fs_init(void) 427 { 428 int rc; 429 430 dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0, 431 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| 432 SLAB_MEM_SPREAD|SLAB_ACCOUNT), 433 init_once); 434 if (!dax_cache) 435 return -ENOMEM; 436 437 dax_mnt = kern_mount(&dax_fs_type); 438 if (IS_ERR(dax_mnt)) { 439 rc = PTR_ERR(dax_mnt); 440 goto err_mount; 441 } 442 dax_superblock = dax_mnt->mnt_sb; 443 444 return 0; 445 446 err_mount: 447 kmem_cache_destroy(dax_cache); 448 449 return rc; 450 } 451 452 static void dax_fs_exit(void) 453 { 454 kern_unmount(dax_mnt); 455 kmem_cache_destroy(dax_cache); 456 } 457 458 static int __init dax_core_init(void) 459 { 460 int rc; 461 462 rc = dax_fs_init(); 463 if (rc) 464 return rc; 465 466 rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax"); 467 if (rc) 468 goto err_chrdev; 469 470 rc = dax_bus_init(); 471 if (rc) 472 goto err_bus; 473 return 0; 474 475 err_bus: 476 unregister_chrdev_region(dax_devt, MINORMASK+1); 477 err_chrdev: 478 dax_fs_exit(); 479 return 0; 480 } 481 482 static void __exit dax_core_exit(void) 483 { 484 dax_bus_exit(); 485 unregister_chrdev_region(dax_devt, MINORMASK+1); 486 ida_destroy(&dax_minor_ida); 487 dax_fs_exit(); 488 } 489 490 MODULE_AUTHOR("Intel Corporation"); 491 MODULE_LICENSE("GPL v2"); 492 subsys_initcall(dax_core_init); 493 module_exit(dax_core_exit); 494