15b497af4SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 27b6be844SDan Williams /* 37b6be844SDan Williams * Copyright(c) 2017 Intel Corporation. All rights reserved. 47b6be844SDan Williams */ 57b6be844SDan Williams #include <linux/pagemap.h> 67b6be844SDan Williams #include <linux/module.h> 77b6be844SDan Williams #include <linux/mount.h> 875d4e06fSDavid Howells #include <linux/pseudo_fs.h> 97b6be844SDan Williams #include <linux/magic.h> 10ef510424SDan Williams #include <linux/genhd.h> 11569d0365SDan Williams #include <linux/pfn_t.h> 127b6be844SDan Williams #include <linux/cdev.h> 137b6be844SDan Williams #include <linux/hash.h> 147b6be844SDan Williams #include <linux/slab.h> 157e026c8cSDan Williams #include <linux/uio.h> 166568b08bSDan Williams #include <linux/dax.h> 177b6be844SDan Williams #include <linux/fs.h> 1851cf784cSDan Williams #include "dax-private.h" 197b6be844SDan Williams 201b764601SChristoph Hellwig /** 211b764601SChristoph Hellwig * struct dax_device - anchor object for dax services 221b764601SChristoph Hellwig * @inode: core vfs 231b764601SChristoph Hellwig * @cdev: optional character interface for "device dax" 241b764601SChristoph Hellwig * @host: optional name for lookups where the device path is not available 251b764601SChristoph Hellwig * @private: dax driver private data 261b764601SChristoph Hellwig * @flags: state and boolean properties 271b764601SChristoph Hellwig */ 281b764601SChristoph Hellwig struct dax_device { 291b764601SChristoph Hellwig struct hlist_node list; 301b764601SChristoph Hellwig struct inode inode; 311b764601SChristoph Hellwig struct cdev cdev; 321b764601SChristoph Hellwig const char *host; 331b764601SChristoph Hellwig void *private; 341b764601SChristoph Hellwig unsigned long flags; 351b764601SChristoph Hellwig const struct dax_operations *ops; 361b764601SChristoph Hellwig }; 371b764601SChristoph Hellwig 387b6be844SDan Williams static dev_t dax_devt; 397b6be844SDan Williams DEFINE_STATIC_SRCU(dax_srcu); 407b6be844SDan Williams static struct vfsmount *dax_mnt; 417b6be844SDan Williams static DEFINE_IDA(dax_minor_ida); 427b6be844SDan Williams static struct kmem_cache *dax_cache __read_mostly; 437b6be844SDan Williams static struct super_block *dax_superblock __read_mostly; 447b6be844SDan Williams 4572058005SDan Williams #define DAX_HASH_SIZE (PAGE_SIZE / sizeof(struct hlist_head)) 4672058005SDan Williams static struct hlist_head dax_host_list[DAX_HASH_SIZE]; 4772058005SDan Williams static DEFINE_SPINLOCK(dax_host_lock); 4872058005SDan Williams 497b6be844SDan Williams int dax_read_lock(void) 507b6be844SDan Williams { 517b6be844SDan Williams return srcu_read_lock(&dax_srcu); 527b6be844SDan Williams } 537b6be844SDan Williams EXPORT_SYMBOL_GPL(dax_read_lock); 547b6be844SDan Williams 557b6be844SDan Williams void dax_read_unlock(int id) 567b6be844SDan Williams { 577b6be844SDan Williams srcu_read_unlock(&dax_srcu, id); 587b6be844SDan Williams } 597b6be844SDan Williams EXPORT_SYMBOL_GPL(dax_read_unlock); 607b6be844SDan Williams 611b764601SChristoph Hellwig static int dax_host_hash(const char *host) 621b764601SChristoph Hellwig { 631b764601SChristoph Hellwig return hashlen_hash(hashlen_string("DAX", host)) % DAX_HASH_SIZE; 641b764601SChristoph Hellwig } 651b764601SChristoph Hellwig 661b764601SChristoph Hellwig /** 671b764601SChristoph Hellwig * dax_get_by_host() - temporary lookup mechanism for filesystem-dax 681b764601SChristoph Hellwig * @host: alternate name for the device registered by a dax driver 691b764601SChristoph Hellwig */ 701b764601SChristoph Hellwig static struct dax_device *dax_get_by_host(const char *host) 711b764601SChristoph Hellwig { 721b764601SChristoph Hellwig struct dax_device *dax_dev, *found = NULL; 731b764601SChristoph Hellwig int hash, id; 741b764601SChristoph Hellwig 751b764601SChristoph Hellwig if (!host) 761b764601SChristoph Hellwig return NULL; 771b764601SChristoph Hellwig 781b764601SChristoph Hellwig hash = dax_host_hash(host); 791b764601SChristoph Hellwig 801b764601SChristoph Hellwig id = dax_read_lock(); 811b764601SChristoph Hellwig spin_lock(&dax_host_lock); 821b764601SChristoph Hellwig hlist_for_each_entry(dax_dev, &dax_host_list[hash], list) { 831b764601SChristoph Hellwig if (!dax_alive(dax_dev) 841b764601SChristoph Hellwig || strcmp(host, dax_dev->host) != 0) 851b764601SChristoph Hellwig continue; 861b764601SChristoph Hellwig 871b764601SChristoph Hellwig if (igrab(&dax_dev->inode)) 881b764601SChristoph Hellwig found = dax_dev; 891b764601SChristoph Hellwig break; 901b764601SChristoph Hellwig } 911b764601SChristoph Hellwig spin_unlock(&dax_host_lock); 921b764601SChristoph Hellwig dax_read_unlock(id); 931b764601SChristoph Hellwig 941b764601SChristoph Hellwig return found; 951b764601SChristoph Hellwig } 961b764601SChristoph Hellwig 979d109081SDan Williams #ifdef CONFIG_BLOCK 9878f35473SDan Williams #include <linux/blkdev.h> 9978f35473SDan Williams 100ef510424SDan Williams int bdev_dax_pgoff(struct block_device *bdev, sector_t sector, size_t size, 101ef510424SDan Williams pgoff_t *pgoff) 102ef510424SDan Williams { 1031a9d5d40SVivek Goyal sector_t start_sect = bdev ? get_start_sect(bdev) : 0; 1041a9d5d40SVivek Goyal phys_addr_t phys_off = (start_sect + sector) * 512; 105ef510424SDan Williams 106ef510424SDan Williams if (pgoff) 107ef510424SDan Williams *pgoff = PHYS_PFN(phys_off); 108ef510424SDan Williams if (phys_off % PAGE_SIZE || size % PAGE_SIZE) 109ef510424SDan Williams return -EINVAL; 110ef510424SDan Williams return 0; 111ef510424SDan Williams } 112ef510424SDan Williams EXPORT_SYMBOL(bdev_dax_pgoff); 113ef510424SDan Williams 11426f2f4deSDan Williams #if IS_ENABLED(CONFIG_FS_DAX) 11578f35473SDan Williams struct dax_device *fs_dax_get_by_bdev(struct block_device *bdev) 11678f35473SDan Williams { 117e556f6baSChristoph Hellwig if (!blk_queue_dax(bdev->bd_disk->queue)) 11878f35473SDan Williams return NULL; 119f01b16a8SVivek Goyal return dax_get_by_host(bdev->bd_disk->disk_name); 12078f35473SDan Williams } 12178f35473SDan Williams EXPORT_SYMBOL_GPL(fs_dax_get_by_bdev); 12226f2f4deSDan Williams #endif 12378f35473SDan Williams 1247bf7eac8SDan Williams bool __generic_fsdax_supported(struct dax_device *dax_dev, 1257bf7eac8SDan Williams struct block_device *bdev, int blocksize, sector_t start, 1267bf7eac8SDan Williams sector_t sectors) 127ef510424SDan Williams { 128e7638488SDan Williams bool dax_enabled = false; 129ad428cdbSDan Williams pgoff_t pgoff, pgoff_end; 130ad428cdbSDan Williams void *kaddr, *end_kaddr; 131ad428cdbSDan Williams pfn_t pfn, end_pfn; 132ad428cdbSDan Williams sector_t last_page; 133ad428cdbSDan Williams long len, len2; 134ad428cdbSDan Williams int err, id; 135ef510424SDan Williams 136ef510424SDan Williams if (blocksize != PAGE_SIZE) { 13739b6389aSChristoph Hellwig pr_info("%pg: error: unsupported blocksize for dax\n", bdev); 13880660f20SDave Jiang return false; 139ef510424SDan Williams } 140ef510424SDan Williams 141d4c5da50SAdrian Huang if (!dax_dev) { 14239b6389aSChristoph Hellwig pr_debug("%pg: error: dax unsupported by block device\n", bdev); 143d4c5da50SAdrian Huang return false; 144d4c5da50SAdrian Huang } 145d4c5da50SAdrian Huang 1467bf7eac8SDan Williams err = bdev_dax_pgoff(bdev, start, PAGE_SIZE, &pgoff); 147ef510424SDan Williams if (err) { 14839b6389aSChristoph Hellwig pr_info("%pg: error: unaligned partition for dax\n", bdev); 14980660f20SDave Jiang return false; 150ef510424SDan Williams } 151ef510424SDan Williams 1527bf7eac8SDan Williams last_page = PFN_DOWN((start + sectors - 1) * 512) * PAGE_SIZE / 512; 153ad428cdbSDan Williams err = bdev_dax_pgoff(bdev, last_page, PAGE_SIZE, &pgoff_end); 154ad428cdbSDan Williams if (err) { 15539b6389aSChristoph Hellwig pr_info("%pg: error: unaligned partition for dax\n", bdev); 156ad428cdbSDan Williams return false; 157ad428cdbSDan Williams } 158ad428cdbSDan Williams 159ef510424SDan Williams id = dax_read_lock(); 160ad428cdbSDan Williams len = dax_direct_access(dax_dev, pgoff, 1, &kaddr, &pfn); 161ad428cdbSDan Williams len2 = dax_direct_access(dax_dev, pgoff_end, 1, &end_kaddr, &end_pfn); 162ef510424SDan Williams 163ad428cdbSDan Williams if (len < 1 || len2 < 1) { 16439b6389aSChristoph Hellwig pr_info("%pg: error: dax access failed (%ld)\n", 16539b6389aSChristoph Hellwig bdev, len < 1 ? len : len2); 166eedfd73dSIra Weiny dax_read_unlock(id); 16780660f20SDave Jiang return false; 168ef510424SDan Williams } 169ef510424SDan Williams 1703fe0791cSDan Williams if (IS_ENABLED(CONFIG_FS_DAX_LIMITED) && pfn_t_special(pfn)) { 1713fe0791cSDan Williams /* 1723fe0791cSDan Williams * An arch that has enabled the pmem api should also 1733fe0791cSDan Williams * have its drivers support pfn_t_devmap() 1743fe0791cSDan Williams * 1753fe0791cSDan Williams * This is a developer warning and should not trigger in 1763fe0791cSDan Williams * production. dax_flush() will crash since it depends 1773fe0791cSDan Williams * on being able to do (page_address(pfn_to_page())). 1783fe0791cSDan Williams */ 1793fe0791cSDan Williams WARN_ON(IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API)); 180e7638488SDan Williams dax_enabled = true; 181ad428cdbSDan Williams } else if (pfn_t_devmap(pfn) && pfn_t_devmap(end_pfn)) { 182ad428cdbSDan Williams struct dev_pagemap *pgmap, *end_pgmap; 183e7638488SDan Williams 184e7638488SDan Williams pgmap = get_dev_pagemap(pfn_t_to_pfn(pfn), NULL); 185ad428cdbSDan Williams end_pgmap = get_dev_pagemap(pfn_t_to_pfn(end_pfn), NULL); 186ad428cdbSDan Williams if (pgmap && pgmap == end_pgmap && pgmap->type == MEMORY_DEVICE_FS_DAX 187ad428cdbSDan Williams && pfn_t_to_page(pfn)->pgmap == pgmap 188ad428cdbSDan Williams && pfn_t_to_page(end_pfn)->pgmap == pgmap 189ad428cdbSDan Williams && pfn_t_to_pfn(pfn) == PHYS_PFN(__pa(kaddr)) 190ad428cdbSDan Williams && pfn_t_to_pfn(end_pfn) == PHYS_PFN(__pa(end_kaddr))) 191e7638488SDan Williams dax_enabled = true; 192e7638488SDan Williams put_dev_pagemap(pgmap); 193ad428cdbSDan Williams put_dev_pagemap(end_pgmap); 194ad428cdbSDan Williams 195e7638488SDan Williams } 196eedfd73dSIra Weiny dax_read_unlock(id); 197e7638488SDan Williams 198e7638488SDan Williams if (!dax_enabled) { 19939b6389aSChristoph Hellwig pr_info("%pg: error: dax support not enabled\n", bdev); 20080660f20SDave Jiang return false; 201569d0365SDan Williams } 20280660f20SDave Jiang return true; 203ef510424SDan Williams } 2047bf7eac8SDan Williams EXPORT_SYMBOL_GPL(__generic_fsdax_supported); 2057bf7eac8SDan Williams 2067bf7eac8SDan Williams /** 2077bf7eac8SDan Williams * __bdev_dax_supported() - Check if the device supports dax for filesystem 2087bf7eac8SDan Williams * @bdev: block device to check 2097bf7eac8SDan Williams * @blocksize: The block size of the device 2107bf7eac8SDan Williams * 2117bf7eac8SDan Williams * This is a library function for filesystems to check if the block device 2127bf7eac8SDan Williams * can be mounted with dax option. 2137bf7eac8SDan Williams * 2147bf7eac8SDan Williams * Return: true if supported, false if unsupported 2157bf7eac8SDan Williams */ 2167bf7eac8SDan Williams bool __bdev_dax_supported(struct block_device *bdev, int blocksize) 2177bf7eac8SDan Williams { 2187bf7eac8SDan Williams struct dax_device *dax_dev; 2197bf7eac8SDan Williams struct request_queue *q; 2207bf7eac8SDan Williams char buf[BDEVNAME_SIZE]; 2217bf7eac8SDan Williams bool ret; 2227bf7eac8SDan Williams 2237bf7eac8SDan Williams q = bdev_get_queue(bdev); 2247bf7eac8SDan Williams if (!q || !blk_queue_dax(q)) { 2257bf7eac8SDan Williams pr_debug("%s: error: request queue doesn't support dax\n", 2267bf7eac8SDan Williams bdevname(bdev, buf)); 2277bf7eac8SDan Williams return false; 2287bf7eac8SDan Williams } 2297bf7eac8SDan Williams 2307bf7eac8SDan Williams dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); 2317bf7eac8SDan Williams if (!dax_dev) { 2327bf7eac8SDan Williams pr_debug("%s: error: device does not support dax\n", 2337bf7eac8SDan Williams bdevname(bdev, buf)); 2347bf7eac8SDan Williams return false; 2357bf7eac8SDan Williams } 2367bf7eac8SDan Williams 2377bf7eac8SDan Williams ret = dax_supported(dax_dev, bdev, blocksize, 0, 2387bf7eac8SDan Williams i_size_read(bdev->bd_inode) / 512); 2397bf7eac8SDan Williams 2407bf7eac8SDan Williams put_dax(dax_dev); 2417bf7eac8SDan Williams 2427bf7eac8SDan Williams return ret; 2437bf7eac8SDan Williams } 244ef510424SDan Williams EXPORT_SYMBOL_GPL(__bdev_dax_supported); 2459d109081SDan Williams #endif 246ef510424SDan Williams 2479a60c3efSDan Williams enum dax_device_flags { 2489a60c3efSDan Williams /* !alive + rcu grace period == no new operations / mappings */ 2499a60c3efSDan Williams DAXDEV_ALIVE, 2506e0c90d6SDan Williams /* gate whether dax_flush() calls the low level flush routine */ 2516e0c90d6SDan Williams DAXDEV_WRITE_CACHE, 252fefc1d97SPankaj Gupta /* flag to check if device supports synchronous flush */ 253fefc1d97SPankaj Gupta DAXDEV_SYNC, 2549a60c3efSDan Williams }; 2559a60c3efSDan Williams 2566e0c90d6SDan Williams static ssize_t write_cache_show(struct device *dev, 2576e0c90d6SDan Williams struct device_attribute *attr, char *buf) 2586e0c90d6SDan Williams { 2596e0c90d6SDan Williams struct dax_device *dax_dev = dax_get_by_host(dev_name(dev)); 2606e0c90d6SDan Williams ssize_t rc; 2616e0c90d6SDan Williams 2626e0c90d6SDan Williams WARN_ON_ONCE(!dax_dev); 2636e0c90d6SDan Williams if (!dax_dev) 2646e0c90d6SDan Williams return -ENXIO; 2656e0c90d6SDan Williams 266808c340bSRoss Zwisler rc = sprintf(buf, "%d\n", !!dax_write_cache_enabled(dax_dev)); 2676e0c90d6SDan Williams put_dax(dax_dev); 2686e0c90d6SDan Williams return rc; 2696e0c90d6SDan Williams } 2706e0c90d6SDan Williams 2716e0c90d6SDan Williams static ssize_t write_cache_store(struct device *dev, 2726e0c90d6SDan Williams struct device_attribute *attr, const char *buf, size_t len) 2736e0c90d6SDan Williams { 2746e0c90d6SDan Williams bool write_cache; 2756e0c90d6SDan Williams int rc = strtobool(buf, &write_cache); 2766e0c90d6SDan Williams struct dax_device *dax_dev = dax_get_by_host(dev_name(dev)); 2776e0c90d6SDan Williams 2786e0c90d6SDan Williams WARN_ON_ONCE(!dax_dev); 2796e0c90d6SDan Williams if (!dax_dev) 2806e0c90d6SDan Williams return -ENXIO; 2816e0c90d6SDan Williams 2826e0c90d6SDan Williams if (rc) 2836e0c90d6SDan Williams len = rc; 2846e0c90d6SDan Williams else 285808c340bSRoss Zwisler dax_write_cache(dax_dev, write_cache); 2866e0c90d6SDan Williams 2876e0c90d6SDan Williams put_dax(dax_dev); 2886e0c90d6SDan Williams return len; 2896e0c90d6SDan Williams } 2906e0c90d6SDan Williams static DEVICE_ATTR_RW(write_cache); 2916e0c90d6SDan Williams 2926e0c90d6SDan Williams static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n) 2936e0c90d6SDan Williams { 2946e0c90d6SDan Williams struct device *dev = container_of(kobj, typeof(*dev), kobj); 2956e0c90d6SDan Williams struct dax_device *dax_dev = dax_get_by_host(dev_name(dev)); 2966e0c90d6SDan Williams 2976e0c90d6SDan Williams WARN_ON_ONCE(!dax_dev); 2986e0c90d6SDan Williams if (!dax_dev) 2996e0c90d6SDan Williams return 0; 3006e0c90d6SDan Williams 301c3ca015fSMikulas Patocka #ifndef CONFIG_ARCH_HAS_PMEM_API 302c3ca015fSMikulas Patocka if (a == &dev_attr_write_cache.attr) 3036e0c90d6SDan Williams return 0; 304c3ca015fSMikulas Patocka #endif 3056e0c90d6SDan Williams return a->mode; 3066e0c90d6SDan Williams } 3076e0c90d6SDan Williams 3086e0c90d6SDan Williams static struct attribute *dax_attributes[] = { 3096e0c90d6SDan Williams &dev_attr_write_cache.attr, 3106e0c90d6SDan Williams NULL, 3116e0c90d6SDan Williams }; 3126e0c90d6SDan Williams 3136e0c90d6SDan Williams struct attribute_group dax_attribute_group = { 3146e0c90d6SDan Williams .name = "dax", 3156e0c90d6SDan Williams .attrs = dax_attributes, 3166e0c90d6SDan Williams .is_visible = dax_visible, 3176e0c90d6SDan Williams }; 3186e0c90d6SDan Williams EXPORT_SYMBOL_GPL(dax_attribute_group); 3196e0c90d6SDan Williams 320b0686260SDan Williams /** 321b0686260SDan Williams * dax_direct_access() - translate a device pgoff to an absolute pfn 322b0686260SDan Williams * @dax_dev: a dax_device instance representing the logical memory range 323b0686260SDan Williams * @pgoff: offset in pages from the start of the device to translate 324b0686260SDan Williams * @nr_pages: number of consecutive pages caller can handle relative to @pfn 325b0686260SDan Williams * @kaddr: output parameter that returns a virtual address mapping of pfn 326b0686260SDan Williams * @pfn: output parameter that returns an absolute pfn translation of @pgoff 327b0686260SDan Williams * 328b0686260SDan Williams * Return: negative errno if an error occurs, otherwise the number of 329b0686260SDan Williams * pages accessible at the device relative @pgoff. 330b0686260SDan Williams */ 331b0686260SDan Williams long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, 332b0686260SDan Williams void **kaddr, pfn_t *pfn) 333b0686260SDan Williams { 334b0686260SDan Williams long avail; 335b0686260SDan Williams 336b0686260SDan Williams if (!dax_dev) 337b0686260SDan Williams return -EOPNOTSUPP; 338b0686260SDan Williams 339b0686260SDan Williams if (!dax_alive(dax_dev)) 340b0686260SDan Williams return -ENXIO; 341b0686260SDan Williams 342b0686260SDan Williams if (nr_pages < 0) 343b05d4c57SIra Weiny return -EINVAL; 344b0686260SDan Williams 345b0686260SDan Williams avail = dax_dev->ops->direct_access(dax_dev, pgoff, nr_pages, 346b0686260SDan Williams kaddr, pfn); 347b0686260SDan Williams if (!avail) 348b0686260SDan Williams return -ERANGE; 349b0686260SDan Williams return min(avail, nr_pages); 350b0686260SDan Williams } 351b0686260SDan Williams EXPORT_SYMBOL_GPL(dax_direct_access); 352b0686260SDan Williams 3537bf7eac8SDan Williams bool dax_supported(struct dax_device *dax_dev, struct block_device *bdev, 3547bf7eac8SDan Williams int blocksize, sector_t start, sector_t len) 3557bf7eac8SDan Williams { 356*673a0658SChristoph Hellwig bool ret = false; 357*673a0658SChristoph Hellwig int id; 358*673a0658SChristoph Hellwig 359e2ec5128SJan Kara if (!dax_dev) 360e2ec5128SJan Kara return false; 361e2ec5128SJan Kara 362*673a0658SChristoph Hellwig id = dax_read_lock(); 363*673a0658SChristoph Hellwig if (dax_alive(dax_dev)) 364*673a0658SChristoph Hellwig ret = dax_dev->ops->dax_supported(dax_dev, bdev, blocksize, 365*673a0658SChristoph Hellwig start, len); 366*673a0658SChristoph Hellwig dax_read_unlock(id); 367*673a0658SChristoph Hellwig return ret; 3687bf7eac8SDan Williams } 369e2ec5128SJan Kara EXPORT_SYMBOL_GPL(dax_supported); 3707bf7eac8SDan Williams 3717e026c8cSDan Williams size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, 3727e026c8cSDan Williams size_t bytes, struct iov_iter *i) 3737e026c8cSDan Williams { 3747e026c8cSDan Williams if (!dax_alive(dax_dev)) 3757e026c8cSDan Williams return 0; 3767e026c8cSDan Williams 3777e026c8cSDan Williams return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i); 3787e026c8cSDan Williams } 3797e026c8cSDan Williams EXPORT_SYMBOL_GPL(dax_copy_from_iter); 3807e026c8cSDan Williams 381b3a9a0c3SDan Williams size_t dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, 382b3a9a0c3SDan Williams size_t bytes, struct iov_iter *i) 383b3a9a0c3SDan Williams { 384b3a9a0c3SDan Williams if (!dax_alive(dax_dev)) 385b3a9a0c3SDan Williams return 0; 386b3a9a0c3SDan Williams 387b3a9a0c3SDan Williams return dax_dev->ops->copy_to_iter(dax_dev, pgoff, addr, bytes, i); 388b3a9a0c3SDan Williams } 389b3a9a0c3SDan Williams EXPORT_SYMBOL_GPL(dax_copy_to_iter); 390b3a9a0c3SDan Williams 391f605a263SVivek Goyal int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, 392f605a263SVivek Goyal size_t nr_pages) 393f605a263SVivek Goyal { 394f605a263SVivek Goyal if (!dax_alive(dax_dev)) 395f605a263SVivek Goyal return -ENXIO; 396f605a263SVivek Goyal /* 397f605a263SVivek Goyal * There are no callers that want to zero more than one page as of now. 398f605a263SVivek Goyal * Once users are there, this check can be removed after the 399f605a263SVivek Goyal * device mapper code has been updated to split ranges across targets. 400f605a263SVivek Goyal */ 401f605a263SVivek Goyal if (nr_pages != 1) 402f605a263SVivek Goyal return -EIO; 403f605a263SVivek Goyal 404f605a263SVivek Goyal return dax_dev->ops->zero_page_range(dax_dev, pgoff, nr_pages); 405f605a263SVivek Goyal } 406f605a263SVivek Goyal EXPORT_SYMBOL_GPL(dax_zero_page_range); 407f605a263SVivek Goyal 408c3ca015fSMikulas Patocka #ifdef CONFIG_ARCH_HAS_PMEM_API 409c3ca015fSMikulas Patocka void arch_wb_cache_pmem(void *addr, size_t size); 410c3ca015fSMikulas Patocka void dax_flush(struct dax_device *dax_dev, void *addr, size_t size) 411abebfbe2SDan Williams { 412808c340bSRoss Zwisler if (unlikely(!dax_write_cache_enabled(dax_dev))) 4136e0c90d6SDan Williams return; 4146e0c90d6SDan Williams 415c3ca015fSMikulas Patocka arch_wb_cache_pmem(addr, size); 416abebfbe2SDan Williams } 417c3ca015fSMikulas Patocka #else 418c3ca015fSMikulas Patocka void dax_flush(struct dax_device *dax_dev, void *addr, size_t size) 419c3ca015fSMikulas Patocka { 420c3ca015fSMikulas Patocka } 421c3ca015fSMikulas Patocka #endif 422abebfbe2SDan Williams EXPORT_SYMBOL_GPL(dax_flush); 423abebfbe2SDan Williams 4246e0c90d6SDan Williams void dax_write_cache(struct dax_device *dax_dev, bool wc) 4256e0c90d6SDan Williams { 4266e0c90d6SDan Williams if (wc) 4276e0c90d6SDan Williams set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); 4286e0c90d6SDan Williams else 4296e0c90d6SDan Williams clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); 4306e0c90d6SDan Williams } 4316e0c90d6SDan Williams EXPORT_SYMBOL_GPL(dax_write_cache); 4326e0c90d6SDan Williams 433273752c9SVivek Goyal bool dax_write_cache_enabled(struct dax_device *dax_dev) 434273752c9SVivek Goyal { 435273752c9SVivek Goyal return test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); 436273752c9SVivek Goyal } 437273752c9SVivek Goyal EXPORT_SYMBOL_GPL(dax_write_cache_enabled); 438273752c9SVivek Goyal 439fefc1d97SPankaj Gupta bool __dax_synchronous(struct dax_device *dax_dev) 440fefc1d97SPankaj Gupta { 441fefc1d97SPankaj Gupta return test_bit(DAXDEV_SYNC, &dax_dev->flags); 442fefc1d97SPankaj Gupta } 443fefc1d97SPankaj Gupta EXPORT_SYMBOL_GPL(__dax_synchronous); 444fefc1d97SPankaj Gupta 445fefc1d97SPankaj Gupta void __set_dax_synchronous(struct dax_device *dax_dev) 446fefc1d97SPankaj Gupta { 447fefc1d97SPankaj Gupta set_bit(DAXDEV_SYNC, &dax_dev->flags); 448fefc1d97SPankaj Gupta } 449fefc1d97SPankaj Gupta EXPORT_SYMBOL_GPL(__set_dax_synchronous); 450fefc1d97SPankaj Gupta 4517b6be844SDan Williams bool dax_alive(struct dax_device *dax_dev) 4527b6be844SDan Williams { 4537b6be844SDan Williams lockdep_assert_held(&dax_srcu); 4549a60c3efSDan Williams return test_bit(DAXDEV_ALIVE, &dax_dev->flags); 4557b6be844SDan Williams } 4567b6be844SDan Williams EXPORT_SYMBOL_GPL(dax_alive); 4577b6be844SDan Williams 4587b6be844SDan Williams /* 4597b6be844SDan Williams * Note, rcu is not protecting the liveness of dax_dev, rcu is ensuring 4607b6be844SDan Williams * that any fault handlers or operations that might have seen 4617b6be844SDan Williams * dax_alive(), have completed. Any operations that start after 4627b6be844SDan Williams * synchronize_srcu() has run will abort upon seeing !dax_alive(). 4637b6be844SDan Williams */ 4647b6be844SDan Williams void kill_dax(struct dax_device *dax_dev) 4657b6be844SDan Williams { 4667b6be844SDan Williams if (!dax_dev) 4677b6be844SDan Williams return; 4687b6be844SDan Williams 4699a60c3efSDan Williams clear_bit(DAXDEV_ALIVE, &dax_dev->flags); 47072058005SDan Williams 4717b6be844SDan Williams synchronize_srcu(&dax_srcu); 47272058005SDan Williams 47372058005SDan Williams spin_lock(&dax_host_lock); 47472058005SDan Williams hlist_del_init(&dax_dev->list); 47572058005SDan Williams spin_unlock(&dax_host_lock); 4767b6be844SDan Williams } 4777b6be844SDan Williams EXPORT_SYMBOL_GPL(kill_dax); 4787b6be844SDan Williams 4799567da0bSDan Williams void run_dax(struct dax_device *dax_dev) 4809567da0bSDan Williams { 4819567da0bSDan Williams set_bit(DAXDEV_ALIVE, &dax_dev->flags); 4829567da0bSDan Williams } 4839567da0bSDan Williams EXPORT_SYMBOL_GPL(run_dax); 4849567da0bSDan Williams 4857b6be844SDan Williams static struct inode *dax_alloc_inode(struct super_block *sb) 4867b6be844SDan Williams { 4877b6be844SDan Williams struct dax_device *dax_dev; 488b9d39d17SDan Williams struct inode *inode; 4897b6be844SDan Williams 4907b6be844SDan Williams dax_dev = kmem_cache_alloc(dax_cache, GFP_KERNEL); 4919f586fffSMikulas Patocka if (!dax_dev) 4929f586fffSMikulas Patocka return NULL; 4939f586fffSMikulas Patocka 494b9d39d17SDan Williams inode = &dax_dev->inode; 495b9d39d17SDan Williams inode->i_rdev = 0; 496b9d39d17SDan Williams return inode; 4977b6be844SDan Williams } 4987b6be844SDan Williams 4997b6be844SDan Williams static struct dax_device *to_dax_dev(struct inode *inode) 5007b6be844SDan Williams { 5017b6be844SDan Williams return container_of(inode, struct dax_device, inode); 5027b6be844SDan Williams } 5037b6be844SDan Williams 50453e22829SAl Viro static void dax_free_inode(struct inode *inode) 5057b6be844SDan Williams { 5067b6be844SDan Williams struct dax_device *dax_dev = to_dax_dev(inode); 50772058005SDan Williams kfree(dax_dev->host); 50872058005SDan Williams dax_dev->host = NULL; 509b9d39d17SDan Williams if (inode->i_rdev) 5106f24784fSAl Viro ida_simple_remove(&dax_minor_ida, iminor(inode)); 5117b6be844SDan Williams kmem_cache_free(dax_cache, dax_dev); 5127b6be844SDan Williams } 5137b6be844SDan Williams 5147b6be844SDan Williams static void dax_destroy_inode(struct inode *inode) 5157b6be844SDan Williams { 5167b6be844SDan Williams struct dax_device *dax_dev = to_dax_dev(inode); 5179a60c3efSDan Williams WARN_ONCE(test_bit(DAXDEV_ALIVE, &dax_dev->flags), 5187b6be844SDan Williams "kill_dax() must be called before final iput()\n"); 5197b6be844SDan Williams } 5207b6be844SDan Williams 5217b6be844SDan Williams static const struct super_operations dax_sops = { 5227b6be844SDan Williams .statfs = simple_statfs, 5237b6be844SDan Williams .alloc_inode = dax_alloc_inode, 5247b6be844SDan Williams .destroy_inode = dax_destroy_inode, 52553e22829SAl Viro .free_inode = dax_free_inode, 5267b6be844SDan Williams .drop_inode = generic_delete_inode, 5277b6be844SDan Williams }; 5287b6be844SDan Williams 52975d4e06fSDavid Howells static int dax_init_fs_context(struct fs_context *fc) 5307b6be844SDan Williams { 53175d4e06fSDavid Howells struct pseudo_fs_context *ctx = init_pseudo(fc, DAXFS_MAGIC); 53275d4e06fSDavid Howells if (!ctx) 53375d4e06fSDavid Howells return -ENOMEM; 53475d4e06fSDavid Howells ctx->ops = &dax_sops; 53575d4e06fSDavid Howells return 0; 5367b6be844SDan Williams } 5377b6be844SDan Williams 5387b6be844SDan Williams static struct file_system_type dax_fs_type = { 5397b6be844SDan Williams .name = "dax", 54075d4e06fSDavid Howells .init_fs_context = dax_init_fs_context, 5417b6be844SDan Williams .kill_sb = kill_anon_super, 5427b6be844SDan Williams }; 5437b6be844SDan Williams 5447b6be844SDan Williams static int dax_test(struct inode *inode, void *data) 5457b6be844SDan Williams { 5467b6be844SDan Williams dev_t devt = *(dev_t *) data; 5477b6be844SDan Williams 5487b6be844SDan Williams return inode->i_rdev == devt; 5497b6be844SDan Williams } 5507b6be844SDan Williams 5517b6be844SDan Williams static int dax_set(struct inode *inode, void *data) 5527b6be844SDan Williams { 5537b6be844SDan Williams dev_t devt = *(dev_t *) data; 5547b6be844SDan Williams 5557b6be844SDan Williams inode->i_rdev = devt; 5567b6be844SDan Williams return 0; 5577b6be844SDan Williams } 5587b6be844SDan Williams 5597b6be844SDan Williams static struct dax_device *dax_dev_get(dev_t devt) 5607b6be844SDan Williams { 5617b6be844SDan Williams struct dax_device *dax_dev; 5627b6be844SDan Williams struct inode *inode; 5637b6be844SDan Williams 5647b6be844SDan Williams inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31), 5657b6be844SDan Williams dax_test, dax_set, &devt); 5667b6be844SDan Williams 5677b6be844SDan Williams if (!inode) 5687b6be844SDan Williams return NULL; 5697b6be844SDan Williams 5707b6be844SDan Williams dax_dev = to_dax_dev(inode); 5717b6be844SDan Williams if (inode->i_state & I_NEW) { 5729a60c3efSDan Williams set_bit(DAXDEV_ALIVE, &dax_dev->flags); 5737b6be844SDan Williams inode->i_cdev = &dax_dev->cdev; 5747b6be844SDan Williams inode->i_mode = S_IFCHR; 5757b6be844SDan Williams inode->i_flags = S_DAX; 5767b6be844SDan Williams mapping_set_gfp_mask(&inode->i_data, GFP_USER); 5777b6be844SDan Williams unlock_new_inode(inode); 5787b6be844SDan Williams } 5797b6be844SDan Williams 5807b6be844SDan Williams return dax_dev; 5817b6be844SDan Williams } 5827b6be844SDan Williams 58372058005SDan Williams static void dax_add_host(struct dax_device *dax_dev, const char *host) 58472058005SDan Williams { 58572058005SDan Williams int hash; 58672058005SDan Williams 58772058005SDan Williams /* 58872058005SDan Williams * Unconditionally init dax_dev since it's coming from a 58972058005SDan Williams * non-zeroed slab cache 59072058005SDan Williams */ 59172058005SDan Williams INIT_HLIST_NODE(&dax_dev->list); 59272058005SDan Williams dax_dev->host = host; 59372058005SDan Williams if (!host) 59472058005SDan Williams return; 59572058005SDan Williams 59672058005SDan Williams hash = dax_host_hash(host); 59772058005SDan Williams spin_lock(&dax_host_lock); 59872058005SDan Williams hlist_add_head(&dax_dev->list, &dax_host_list[hash]); 59972058005SDan Williams spin_unlock(&dax_host_lock); 60072058005SDan Williams } 60172058005SDan Williams 6026568b08bSDan Williams struct dax_device *alloc_dax(void *private, const char *__host, 603fefc1d97SPankaj Gupta const struct dax_operations *ops, unsigned long flags) 6047b6be844SDan Williams { 6057b6be844SDan Williams struct dax_device *dax_dev; 60672058005SDan Williams const char *host; 6077b6be844SDan Williams dev_t devt; 6087b6be844SDan Williams int minor; 6097b6be844SDan Williams 6104e4ced93SVivek Goyal if (ops && !ops->zero_page_range) { 6114e4ced93SVivek Goyal pr_debug("%s: error: device does not provide dax" 6124e4ced93SVivek Goyal " operation zero_page_range()\n", 6134e4ced93SVivek Goyal __host ? __host : "Unknown"); 6144e4ced93SVivek Goyal return ERR_PTR(-EINVAL); 6154e4ced93SVivek Goyal } 6164e4ced93SVivek Goyal 61772058005SDan Williams host = kstrdup(__host, GFP_KERNEL); 61872058005SDan Williams if (__host && !host) 6194e4ced93SVivek Goyal return ERR_PTR(-ENOMEM); 62072058005SDan Williams 621cf1e2289SDan Williams minor = ida_simple_get(&dax_minor_ida, 0, MINORMASK+1, GFP_KERNEL); 6227b6be844SDan Williams if (minor < 0) 62372058005SDan Williams goto err_minor; 6247b6be844SDan Williams 6257b6be844SDan Williams devt = MKDEV(MAJOR(dax_devt), minor); 6267b6be844SDan Williams dax_dev = dax_dev_get(devt); 6277b6be844SDan Williams if (!dax_dev) 62872058005SDan Williams goto err_dev; 6297b6be844SDan Williams 63072058005SDan Williams dax_add_host(dax_dev, host); 6316568b08bSDan Williams dax_dev->ops = ops; 6327b6be844SDan Williams dax_dev->private = private; 633fefc1d97SPankaj Gupta if (flags & DAXDEV_F_SYNC) 634fefc1d97SPankaj Gupta set_dax_synchronous(dax_dev); 635fefc1d97SPankaj Gupta 6367b6be844SDan Williams return dax_dev; 6377b6be844SDan Williams 63872058005SDan Williams err_dev: 6397b6be844SDan Williams ida_simple_remove(&dax_minor_ida, minor); 64072058005SDan Williams err_minor: 64172058005SDan Williams kfree(host); 6424e4ced93SVivek Goyal return ERR_PTR(-ENOMEM); 6437b6be844SDan Williams } 6447b6be844SDan Williams EXPORT_SYMBOL_GPL(alloc_dax); 6457b6be844SDan Williams 6467b6be844SDan Williams void put_dax(struct dax_device *dax_dev) 6477b6be844SDan Williams { 6487b6be844SDan Williams if (!dax_dev) 6497b6be844SDan Williams return; 6507b6be844SDan Williams iput(&dax_dev->inode); 6517b6be844SDan Williams } 6527b6be844SDan Williams EXPORT_SYMBOL_GPL(put_dax); 6537b6be844SDan Williams 6547b6be844SDan Williams /** 6557b6be844SDan Williams * inode_dax: convert a public inode into its dax_dev 6567b6be844SDan Williams * @inode: An inode with i_cdev pointing to a dax_dev 6577b6be844SDan Williams * 6587b6be844SDan Williams * Note this is not equivalent to to_dax_dev() which is for private 6597b6be844SDan Williams * internal use where we know the inode filesystem type == dax_fs_type. 6607b6be844SDan Williams */ 6617b6be844SDan Williams struct dax_device *inode_dax(struct inode *inode) 6627b6be844SDan Williams { 6637b6be844SDan Williams struct cdev *cdev = inode->i_cdev; 6647b6be844SDan Williams 6657b6be844SDan Williams return container_of(cdev, struct dax_device, cdev); 6667b6be844SDan Williams } 6677b6be844SDan Williams EXPORT_SYMBOL_GPL(inode_dax); 6687b6be844SDan Williams 6697b6be844SDan Williams struct inode *dax_inode(struct dax_device *dax_dev) 6707b6be844SDan Williams { 6717b6be844SDan Williams return &dax_dev->inode; 6727b6be844SDan Williams } 6737b6be844SDan Williams EXPORT_SYMBOL_GPL(dax_inode); 6747b6be844SDan Williams 6757b6be844SDan Williams void *dax_get_private(struct dax_device *dax_dev) 6767b6be844SDan Williams { 6779567da0bSDan Williams if (!test_bit(DAXDEV_ALIVE, &dax_dev->flags)) 6789567da0bSDan Williams return NULL; 6797b6be844SDan Williams return dax_dev->private; 6807b6be844SDan Williams } 6817b6be844SDan Williams EXPORT_SYMBOL_GPL(dax_get_private); 6827b6be844SDan Williams 6837b6be844SDan Williams static void init_once(void *_dax_dev) 6847b6be844SDan Williams { 6857b6be844SDan Williams struct dax_device *dax_dev = _dax_dev; 6867b6be844SDan Williams struct inode *inode = &dax_dev->inode; 6877b6be844SDan Williams 688b9d39d17SDan Williams memset(dax_dev, 0, sizeof(*dax_dev)); 6897b6be844SDan Williams inode_init_once(inode); 6907b6be844SDan Williams } 6917b6be844SDan Williams 6929567da0bSDan Williams static int dax_fs_init(void) 6937b6be844SDan Williams { 6947b6be844SDan Williams int rc; 6957b6be844SDan Williams 6967b6be844SDan Williams dax_cache = kmem_cache_create("dax_cache", sizeof(struct dax_device), 0, 6977b6be844SDan Williams (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| 6987b6be844SDan Williams SLAB_MEM_SPREAD|SLAB_ACCOUNT), 6997b6be844SDan Williams init_once); 7007b6be844SDan Williams if (!dax_cache) 7017b6be844SDan Williams return -ENOMEM; 7027b6be844SDan Williams 7037b6be844SDan Williams dax_mnt = kern_mount(&dax_fs_type); 7047b6be844SDan Williams if (IS_ERR(dax_mnt)) { 7057b6be844SDan Williams rc = PTR_ERR(dax_mnt); 7067b6be844SDan Williams goto err_mount; 7077b6be844SDan Williams } 7087b6be844SDan Williams dax_superblock = dax_mnt->mnt_sb; 7097b6be844SDan Williams 7107b6be844SDan Williams return 0; 7117b6be844SDan Williams 7127b6be844SDan Williams err_mount: 7137b6be844SDan Williams kmem_cache_destroy(dax_cache); 7147b6be844SDan Williams 7157b6be844SDan Williams return rc; 7167b6be844SDan Williams } 7177b6be844SDan Williams 7189567da0bSDan Williams static void dax_fs_exit(void) 7197b6be844SDan Williams { 7207b6be844SDan Williams kern_unmount(dax_mnt); 7217b6be844SDan Williams kmem_cache_destroy(dax_cache); 7227b6be844SDan Williams } 7237b6be844SDan Williams 7249567da0bSDan Williams static int __init dax_core_init(void) 7257b6be844SDan Williams { 7267b6be844SDan Williams int rc; 7277b6be844SDan Williams 7289567da0bSDan Williams rc = dax_fs_init(); 7297b6be844SDan Williams if (rc) 7307b6be844SDan Williams return rc; 7317b6be844SDan Williams 732cf1e2289SDan Williams rc = alloc_chrdev_region(&dax_devt, 0, MINORMASK+1, "dax"); 7337b6be844SDan Williams if (rc) 7349567da0bSDan Williams goto err_chrdev; 7359567da0bSDan Williams 7369567da0bSDan Williams rc = dax_bus_init(); 7379567da0bSDan Williams if (rc) 7389567da0bSDan Williams goto err_bus; 7399567da0bSDan Williams return 0; 7409567da0bSDan Williams 7419567da0bSDan Williams err_bus: 7429567da0bSDan Williams unregister_chrdev_region(dax_devt, MINORMASK+1); 7439567da0bSDan Williams err_chrdev: 7449567da0bSDan Williams dax_fs_exit(); 7459567da0bSDan Williams return 0; 7467b6be844SDan Williams } 7477b6be844SDan Williams 7489567da0bSDan Williams static void __exit dax_core_exit(void) 7497b6be844SDan Williams { 7501aa57431SWang Hai dax_bus_exit(); 751cf1e2289SDan Williams unregister_chrdev_region(dax_devt, MINORMASK+1); 7527b6be844SDan Williams ida_destroy(&dax_minor_ida); 7539567da0bSDan Williams dax_fs_exit(); 7547b6be844SDan Williams } 7557b6be844SDan Williams 7567b6be844SDan Williams MODULE_AUTHOR("Intel Corporation"); 7577b6be844SDan Williams MODULE_LICENSE("GPL v2"); 7589567da0bSDan Williams subsys_initcall(dax_core_init); 7599567da0bSDan Williams module_exit(dax_core_exit); 760