Matthew Wilcox | d475c63 | 2015-02-16 15:58:56 -0800 | [diff] [blame] | 1 | /* |
| 2 | * fs/dax.c - Direct Access filesystem code |
| 3 | * Copyright (c) 2013-2014 Intel Corporation |
| 4 | * Author: Matthew Wilcox <matthew.r.wilcox@intel.com> |
| 5 | * Author: Ross Zwisler <ross.zwisler@linux.intel.com> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify it |
| 8 | * under the terms and conditions of the GNU General Public License, |
| 9 | * version 2, as published by the Free Software Foundation. |
| 10 | * |
| 11 | * This program is distributed in the hope it will be useful, but WITHOUT |
| 12 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 13 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
| 14 | * more details. |
| 15 | */ |
| 16 | |
| 17 | #include <linux/atomic.h> |
| 18 | #include <linux/blkdev.h> |
| 19 | #include <linux/buffer_head.h> |
Ross Zwisler | d77e92e | 2015-09-09 10:29:40 -0600 | [diff] [blame] | 20 | #include <linux/dax.h> |
Matthew Wilcox | d475c63 | 2015-02-16 15:58:56 -0800 | [diff] [blame] | 21 | #include <linux/fs.h> |
| 22 | #include <linux/genhd.h> |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 23 | #include <linux/highmem.h> |
| 24 | #include <linux/memcontrol.h> |
| 25 | #include <linux/mm.h> |
Matthew Wilcox | d475c63 | 2015-02-16 15:58:56 -0800 | [diff] [blame] | 26 | #include <linux/mutex.h> |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 27 | #include <linux/pagevec.h> |
Matthew Wilcox | 289c6ae | 2015-02-16 15:58:59 -0800 | [diff] [blame] | 28 | #include <linux/sched.h> |
Ingo Molnar | f361bf4 | 2017-02-03 23:47:37 +0100 | [diff] [blame] | 29 | #include <linux/sched/signal.h> |
Matthew Wilcox | d475c63 | 2015-02-16 15:58:56 -0800 | [diff] [blame] | 30 | #include <linux/uio.h> |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 31 | #include <linux/vmstat.h> |
Dan Williams | 34c0fd5 | 2016-01-15 16:56:14 -0800 | [diff] [blame] | 32 | #include <linux/pfn_t.h> |
Dan Williams | 0e749e5 | 2016-01-15 16:55:53 -0800 | [diff] [blame] | 33 | #include <linux/sizes.h> |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 34 | #include <linux/mmu_notifier.h> |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 35 | #include <linux/iomap.h> |
| 36 | #include "internal.h" |
Matthew Wilcox | d475c63 | 2015-02-16 15:58:56 -0800 | [diff] [blame] | 37 | |
Ross Zwisler | 282a8e0 | 2017-02-22 15:39:50 -0800 | [diff] [blame] | 38 | #define CREATE_TRACE_POINTS |
| 39 | #include <trace/events/fs_dax.h> |
| 40 | |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 41 | /* We choose 4096 entries - same as per-zone page wait tables */ |
| 42 | #define DAX_WAIT_TABLE_BITS 12 |
| 43 | #define DAX_WAIT_TABLE_ENTRIES (1 << DAX_WAIT_TABLE_BITS) |
| 44 | |
Ross Zwisler | 917f345 | 2017-09-06 16:18:58 -0700 | [diff] [blame] | 45 | /* The 'colour' (ie low bits) within a PMD of a page offset. */ |
| 46 | #define PG_PMD_COLOUR ((PMD_SIZE >> PAGE_SHIFT) - 1) |
Matthew Wilcox | 977fbdc | 2018-01-31 16:17:36 -0800 | [diff] [blame] | 47 | #define PG_PMD_NR (PMD_SIZE >> PAGE_SHIFT) |
Ross Zwisler | 917f345 | 2017-09-06 16:18:58 -0700 | [diff] [blame] | 48 | |
Ross Zwisler | ce95ab0 | 2016-11-08 11:31:44 +1100 | [diff] [blame] | 49 | static wait_queue_head_t wait_table[DAX_WAIT_TABLE_ENTRIES]; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 50 | |
| 51 | static int __init init_dax_wait_table(void) |
| 52 | { |
| 53 | int i; |
| 54 | |
| 55 | for (i = 0; i < DAX_WAIT_TABLE_ENTRIES; i++) |
| 56 | init_waitqueue_head(wait_table + i); |
| 57 | return 0; |
| 58 | } |
| 59 | fs_initcall(init_dax_wait_table); |
| 60 | |
Ross Zwisler | 527b19d | 2017-09-06 16:18:51 -0700 | [diff] [blame] | 61 | /* |
| 62 | * We use lowest available bit in exceptional entry for locking, one bit for |
| 63 | * the entry size (PMD) and two more to tell us if the entry is a zero page or |
| 64 | * an empty entry that is just used for locking. In total four special bits. |
| 65 | * |
| 66 | * If the PMD bit isn't set the entry has size PAGE_SIZE, and if the ZERO_PAGE |
| 67 | * and EMPTY bits aren't set the entry is a normal DAX entry with a filesystem |
| 68 | * block allocation. |
| 69 | */ |
| 70 | #define RADIX_DAX_SHIFT (RADIX_TREE_EXCEPTIONAL_SHIFT + 4) |
| 71 | #define RADIX_DAX_ENTRY_LOCK (1 << RADIX_TREE_EXCEPTIONAL_SHIFT) |
| 72 | #define RADIX_DAX_PMD (1 << (RADIX_TREE_EXCEPTIONAL_SHIFT + 1)) |
| 73 | #define RADIX_DAX_ZERO_PAGE (1 << (RADIX_TREE_EXCEPTIONAL_SHIFT + 2)) |
| 74 | #define RADIX_DAX_EMPTY (1 << (RADIX_TREE_EXCEPTIONAL_SHIFT + 3)) |
| 75 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 76 | static unsigned long dax_radix_pfn(void *entry) |
Ross Zwisler | 527b19d | 2017-09-06 16:18:51 -0700 | [diff] [blame] | 77 | { |
| 78 | return (unsigned long)entry >> RADIX_DAX_SHIFT; |
| 79 | } |
| 80 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 81 | static void *dax_radix_locked_entry(unsigned long pfn, unsigned long flags) |
Ross Zwisler | 527b19d | 2017-09-06 16:18:51 -0700 | [diff] [blame] | 82 | { |
| 83 | return (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY | flags | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 84 | (pfn << RADIX_DAX_SHIFT) | RADIX_DAX_ENTRY_LOCK); |
Ross Zwisler | 527b19d | 2017-09-06 16:18:51 -0700 | [diff] [blame] | 85 | } |
| 86 | |
| 87 | static unsigned int dax_radix_order(void *entry) |
| 88 | { |
| 89 | if ((unsigned long)entry & RADIX_DAX_PMD) |
| 90 | return PMD_SHIFT - PAGE_SHIFT; |
| 91 | return 0; |
| 92 | } |
| 93 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 94 | static int dax_is_pmd_entry(void *entry) |
| 95 | { |
| 96 | return (unsigned long)entry & RADIX_DAX_PMD; |
| 97 | } |
| 98 | |
| 99 | static int dax_is_pte_entry(void *entry) |
| 100 | { |
| 101 | return !((unsigned long)entry & RADIX_DAX_PMD); |
| 102 | } |
| 103 | |
| 104 | static int dax_is_zero_entry(void *entry) |
| 105 | { |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 106 | return (unsigned long)entry & RADIX_DAX_ZERO_PAGE; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 107 | } |
| 108 | |
| 109 | static int dax_is_empty_entry(void *entry) |
| 110 | { |
| 111 | return (unsigned long)entry & RADIX_DAX_EMPTY; |
| 112 | } |
| 113 | |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 114 | /* |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 115 | * DAX radix tree locking |
| 116 | */ |
| 117 | struct exceptional_entry_key { |
| 118 | struct address_space *mapping; |
Ross Zwisler | 63e95b5 | 2016-11-08 11:32:20 +1100 | [diff] [blame] | 119 | pgoff_t entry_start; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 120 | }; |
| 121 | |
| 122 | struct wait_exceptional_entry_queue { |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 123 | wait_queue_entry_t wait; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 124 | struct exceptional_entry_key key; |
| 125 | }; |
| 126 | |
Ross Zwisler | 63e95b5 | 2016-11-08 11:32:20 +1100 | [diff] [blame] | 127 | static wait_queue_head_t *dax_entry_waitqueue(struct address_space *mapping, |
| 128 | pgoff_t index, void *entry, struct exceptional_entry_key *key) |
| 129 | { |
| 130 | unsigned long hash; |
| 131 | |
| 132 | /* |
| 133 | * If 'entry' is a PMD, align the 'index' that we use for the wait |
| 134 | * queue to the start of that PMD. This ensures that all offsets in |
| 135 | * the range covered by the PMD map to the same bit lock. |
| 136 | */ |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 137 | if (dax_is_pmd_entry(entry)) |
Ross Zwisler | 917f345 | 2017-09-06 16:18:58 -0700 | [diff] [blame] | 138 | index &= ~PG_PMD_COLOUR; |
Ross Zwisler | 63e95b5 | 2016-11-08 11:32:20 +1100 | [diff] [blame] | 139 | |
| 140 | key->mapping = mapping; |
| 141 | key->entry_start = index; |
| 142 | |
| 143 | hash = hash_long((unsigned long)mapping ^ index, DAX_WAIT_TABLE_BITS); |
| 144 | return wait_table + hash; |
| 145 | } |
| 146 | |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 147 | static int wake_exceptional_entry_func(wait_queue_entry_t *wait, unsigned int mode, |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 148 | int sync, void *keyp) |
| 149 | { |
| 150 | struct exceptional_entry_key *key = keyp; |
| 151 | struct wait_exceptional_entry_queue *ewait = |
| 152 | container_of(wait, struct wait_exceptional_entry_queue, wait); |
| 153 | |
| 154 | if (key->mapping != ewait->key.mapping || |
Ross Zwisler | 63e95b5 | 2016-11-08 11:32:20 +1100 | [diff] [blame] | 155 | key->entry_start != ewait->key.entry_start) |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 156 | return 0; |
| 157 | return autoremove_wake_function(wait, mode, sync, NULL); |
| 158 | } |
| 159 | |
| 160 | /* |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 161 | * @entry may no longer be the entry at the index in the mapping. |
| 162 | * The important information it's conveying is whether the entry at |
| 163 | * this index used to be a PMD entry. |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 164 | */ |
Ross Zwisler | d01ad19 | 2017-09-06 16:18:47 -0700 | [diff] [blame] | 165 | static void dax_wake_mapping_entry_waiter(struct address_space *mapping, |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 166 | pgoff_t index, void *entry, bool wake_all) |
| 167 | { |
| 168 | struct exceptional_entry_key key; |
| 169 | wait_queue_head_t *wq; |
| 170 | |
| 171 | wq = dax_entry_waitqueue(mapping, index, entry, &key); |
| 172 | |
| 173 | /* |
| 174 | * Checking for locked entry and prepare_to_wait_exclusive() happens |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 175 | * under the i_pages lock, ditto for entry handling in our callers. |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 176 | * So at this point all tasks that could have seen our entry locked |
| 177 | * must be in the waitqueue and the following check will see them. |
| 178 | */ |
| 179 | if (waitqueue_active(wq)) |
| 180 | __wake_up(wq, TASK_NORMAL, wake_all ? 0 : 1, &key); |
| 181 | } |
| 182 | |
| 183 | /* |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 184 | * Check whether the given slot is locked. Must be called with the i_pages |
| 185 | * lock held. |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 186 | */ |
| 187 | static inline int slot_locked(struct address_space *mapping, void **slot) |
| 188 | { |
| 189 | unsigned long entry = (unsigned long) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 190 | radix_tree_deref_slot_protected(slot, &mapping->i_pages.xa_lock); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 191 | return entry & RADIX_DAX_ENTRY_LOCK; |
| 192 | } |
| 193 | |
| 194 | /* |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 195 | * Mark the given slot as locked. Must be called with the i_pages lock held. |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 196 | */ |
| 197 | static inline void *lock_slot(struct address_space *mapping, void **slot) |
| 198 | { |
| 199 | unsigned long entry = (unsigned long) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 200 | radix_tree_deref_slot_protected(slot, &mapping->i_pages.xa_lock); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 201 | |
| 202 | entry |= RADIX_DAX_ENTRY_LOCK; |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 203 | radix_tree_replace_slot(&mapping->i_pages, slot, (void *)entry); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 204 | return (void *)entry; |
| 205 | } |
| 206 | |
| 207 | /* |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 208 | * Mark the given slot as unlocked. Must be called with the i_pages lock held. |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 209 | */ |
| 210 | static inline void *unlock_slot(struct address_space *mapping, void **slot) |
| 211 | { |
| 212 | unsigned long entry = (unsigned long) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 213 | radix_tree_deref_slot_protected(slot, &mapping->i_pages.xa_lock); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 214 | |
| 215 | entry &= ~(unsigned long)RADIX_DAX_ENTRY_LOCK; |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 216 | radix_tree_replace_slot(&mapping->i_pages, slot, (void *)entry); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 217 | return (void *)entry; |
| 218 | } |
| 219 | |
Matthew Wilcox | e7a121e | 2018-11-16 15:50:02 -0500 | [diff] [blame] | 220 | static void put_unlocked_mapping_entry(struct address_space *mapping, |
| 221 | pgoff_t index, void *entry); |
| 222 | |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 223 | /* |
| 224 | * Lookup entry in radix tree, wait for it to become unlocked if it is |
| 225 | * exceptional entry and return it. The caller must call |
| 226 | * put_unlocked_mapping_entry() when he decided not to lock the entry or |
| 227 | * put_locked_mapping_entry() when he locked the entry and now wants to |
| 228 | * unlock it. |
| 229 | * |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 230 | * Must be called with the i_pages lock held. |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 231 | */ |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 232 | static void *get_unlocked_mapping_entry(struct address_space *mapping, |
| 233 | pgoff_t index, void ***slotp) |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 234 | { |
Ross Zwisler | e3ad61c | 2016-11-08 11:32:12 +1100 | [diff] [blame] | 235 | void *entry, **slot; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 236 | struct wait_exceptional_entry_queue ewait; |
Ross Zwisler | 63e95b5 | 2016-11-08 11:32:20 +1100 | [diff] [blame] | 237 | wait_queue_head_t *wq; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 238 | |
| 239 | init_wait(&ewait.wait); |
| 240 | ewait.wait.func = wake_exceptional_entry_func; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 241 | |
| 242 | for (;;) { |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 243 | entry = __radix_tree_lookup(&mapping->i_pages, index, NULL, |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 244 | &slot); |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 245 | if (!entry || |
| 246 | WARN_ON_ONCE(!radix_tree_exceptional_entry(entry)) || |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 247 | !slot_locked(mapping, slot)) { |
| 248 | if (slotp) |
| 249 | *slotp = slot; |
Ross Zwisler | e3ad61c | 2016-11-08 11:32:12 +1100 | [diff] [blame] | 250 | return entry; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 251 | } |
Ross Zwisler | 63e95b5 | 2016-11-08 11:32:20 +1100 | [diff] [blame] | 252 | |
| 253 | wq = dax_entry_waitqueue(mapping, index, entry, &ewait.key); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 254 | prepare_to_wait_exclusive(wq, &ewait.wait, |
| 255 | TASK_UNINTERRUPTIBLE); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 256 | xa_unlock_irq(&mapping->i_pages); |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 257 | schedule(); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 258 | finish_wait(wq, &ewait.wait); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 259 | xa_lock_irq(&mapping->i_pages); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 260 | } |
| 261 | } |
| 262 | |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 263 | /* |
| 264 | * The only thing keeping the address space around is the i_pages lock |
| 265 | * (it's cycled in clear_inode() after removing the entries from i_pages) |
| 266 | * After we call xas_unlock_irq(), we cannot touch xas->xa. |
| 267 | */ |
| 268 | static void wait_entry_unlocked(struct address_space *mapping, pgoff_t index, |
| 269 | void ***slotp, void *entry) |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 270 | { |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 271 | struct wait_exceptional_entry_queue ewait; |
| 272 | wait_queue_head_t *wq; |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 273 | |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 274 | init_wait(&ewait.wait); |
| 275 | ewait.wait.func = wake_exceptional_entry_func; |
| 276 | |
| 277 | wq = dax_entry_waitqueue(mapping, index, entry, &ewait.key); |
Dan Williams | 9621ea6 | 2019-01-05 11:45:13 -0800 | [diff] [blame] | 278 | /* |
| 279 | * Unlike get_unlocked_entry() there is no guarantee that this |
| 280 | * path ever successfully retrieves an unlocked entry before an |
| 281 | * inode dies. Perform a non-exclusive wait in case this path |
| 282 | * never successfully performs its own wake up. |
| 283 | */ |
| 284 | prepare_to_wait(wq, &ewait.wait, TASK_UNINTERRUPTIBLE); |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 285 | xa_unlock_irq(&mapping->i_pages); |
| 286 | schedule(); |
| 287 | finish_wait(wq, &ewait.wait); |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 288 | } |
| 289 | |
| 290 | static void unlock_mapping_entry(struct address_space *mapping, pgoff_t index) |
Jan Kara | b1aa812 | 2016-12-14 15:07:24 -0800 | [diff] [blame] | 291 | { |
| 292 | void *entry, **slot; |
| 293 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 294 | xa_lock_irq(&mapping->i_pages); |
| 295 | entry = __radix_tree_lookup(&mapping->i_pages, index, NULL, &slot); |
Jan Kara | b1aa812 | 2016-12-14 15:07:24 -0800 | [diff] [blame] | 296 | if (WARN_ON_ONCE(!entry || !radix_tree_exceptional_entry(entry) || |
| 297 | !slot_locked(mapping, slot))) { |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 298 | xa_unlock_irq(&mapping->i_pages); |
Jan Kara | b1aa812 | 2016-12-14 15:07:24 -0800 | [diff] [blame] | 299 | return; |
| 300 | } |
| 301 | unlock_slot(mapping, slot); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 302 | xa_unlock_irq(&mapping->i_pages); |
Jan Kara | b1aa812 | 2016-12-14 15:07:24 -0800 | [diff] [blame] | 303 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); |
| 304 | } |
| 305 | |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 306 | static void put_locked_mapping_entry(struct address_space *mapping, |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 307 | pgoff_t index) |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 308 | { |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 309 | unlock_mapping_entry(mapping, index); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 310 | } |
| 311 | |
| 312 | /* |
| 313 | * Called when we are done with radix tree entry we looked up via |
| 314 | * get_unlocked_mapping_entry() and which we didn't lock in the end. |
| 315 | */ |
| 316 | static void put_unlocked_mapping_entry(struct address_space *mapping, |
| 317 | pgoff_t index, void *entry) |
| 318 | { |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 319 | if (!entry) |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 320 | return; |
| 321 | |
| 322 | /* We have to wake up next waiter for the radix tree entry lock */ |
Ross Zwisler | 422476c | 2016-11-08 11:33:44 +1100 | [diff] [blame] | 323 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); |
| 324 | } |
| 325 | |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 326 | static unsigned long dax_entry_size(void *entry) |
| 327 | { |
| 328 | if (dax_is_zero_entry(entry)) |
| 329 | return 0; |
| 330 | else if (dax_is_empty_entry(entry)) |
| 331 | return 0; |
| 332 | else if (dax_is_pmd_entry(entry)) |
| 333 | return PMD_SIZE; |
| 334 | else |
| 335 | return PAGE_SIZE; |
| 336 | } |
| 337 | |
| 338 | static unsigned long dax_radix_end_pfn(void *entry) |
| 339 | { |
| 340 | return dax_radix_pfn(entry) + dax_entry_size(entry) / PAGE_SIZE; |
| 341 | } |
| 342 | |
| 343 | /* |
| 344 | * Iterate through all mapped pfns represented by an entry, i.e. skip |
| 345 | * 'empty' and 'zero' entries. |
| 346 | */ |
| 347 | #define for_each_mapped_pfn(entry, pfn) \ |
| 348 | for (pfn = dax_radix_pfn(entry); \ |
| 349 | pfn < dax_radix_end_pfn(entry); pfn++) |
| 350 | |
Dan Williams | 73449da | 2018-07-13 21:49:50 -0700 | [diff] [blame] | 351 | /* |
| 352 | * TODO: for reflink+dax we need a way to associate a single page with |
| 353 | * multiple address_space instances at different linear_page_index() |
| 354 | * offsets. |
| 355 | */ |
| 356 | static void dax_associate_entry(void *entry, struct address_space *mapping, |
| 357 | struct vm_area_struct *vma, unsigned long address) |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 358 | { |
Dan Williams | 73449da | 2018-07-13 21:49:50 -0700 | [diff] [blame] | 359 | unsigned long size = dax_entry_size(entry), pfn, index; |
| 360 | int i = 0; |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 361 | |
| 362 | if (IS_ENABLED(CONFIG_FS_DAX_LIMITED)) |
| 363 | return; |
| 364 | |
Dan Williams | 73449da | 2018-07-13 21:49:50 -0700 | [diff] [blame] | 365 | index = linear_page_index(vma, address & ~(size - 1)); |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 366 | for_each_mapped_pfn(entry, pfn) { |
| 367 | struct page *page = pfn_to_page(pfn); |
| 368 | |
| 369 | WARN_ON_ONCE(page->mapping); |
| 370 | page->mapping = mapping; |
Dan Williams | 73449da | 2018-07-13 21:49:50 -0700 | [diff] [blame] | 371 | page->index = index + i++; |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 372 | } |
| 373 | } |
| 374 | |
| 375 | static void dax_disassociate_entry(void *entry, struct address_space *mapping, |
| 376 | bool trunc) |
| 377 | { |
| 378 | unsigned long pfn; |
| 379 | |
| 380 | if (IS_ENABLED(CONFIG_FS_DAX_LIMITED)) |
| 381 | return; |
| 382 | |
| 383 | for_each_mapped_pfn(entry, pfn) { |
| 384 | struct page *page = pfn_to_page(pfn); |
| 385 | |
| 386 | WARN_ON_ONCE(trunc && page_ref_count(page) > 1); |
| 387 | WARN_ON_ONCE(page->mapping && page->mapping != mapping); |
| 388 | page->mapping = NULL; |
Dan Williams | 73449da | 2018-07-13 21:49:50 -0700 | [diff] [blame] | 389 | page->index = 0; |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 390 | } |
| 391 | } |
| 392 | |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 393 | static struct page *dax_busy_page(void *entry) |
| 394 | { |
| 395 | unsigned long pfn; |
| 396 | |
| 397 | for_each_mapped_pfn(entry, pfn) { |
| 398 | struct page *page = pfn_to_page(pfn); |
| 399 | |
| 400 | if (page_ref_count(page) > 1) |
| 401 | return page; |
| 402 | } |
| 403 | return NULL; |
| 404 | } |
| 405 | |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 406 | bool dax_lock_mapping_entry(struct page *page) |
| 407 | { |
| 408 | pgoff_t index; |
| 409 | struct inode *inode; |
| 410 | bool did_lock = false; |
| 411 | void *entry = NULL, **slot; |
| 412 | struct address_space *mapping; |
| 413 | |
| 414 | rcu_read_lock(); |
| 415 | for (;;) { |
| 416 | mapping = READ_ONCE(page->mapping); |
| 417 | |
Matthew Wilcox | 384f181 | 2018-11-27 13:16:33 -0800 | [diff] [blame] | 418 | if (!mapping || !dax_mapping(mapping)) |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 419 | break; |
| 420 | |
| 421 | /* |
| 422 | * In the device-dax case there's no need to lock, a |
| 423 | * struct dev_pagemap pin is sufficient to keep the |
| 424 | * inode alive, and we assume we have dev_pagemap pin |
| 425 | * otherwise we would not have a valid pfn_to_page() |
| 426 | * translation. |
| 427 | */ |
| 428 | inode = mapping->host; |
| 429 | if (S_ISCHR(inode->i_mode)) { |
| 430 | did_lock = true; |
| 431 | break; |
| 432 | } |
| 433 | |
| 434 | xa_lock_irq(&mapping->i_pages); |
| 435 | if (mapping != page->mapping) { |
| 436 | xa_unlock_irq(&mapping->i_pages); |
| 437 | continue; |
| 438 | } |
| 439 | index = page->index; |
| 440 | |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 441 | entry = __radix_tree_lookup(&mapping->i_pages, index, |
| 442 | NULL, &slot); |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 443 | if (!entry) { |
| 444 | xa_unlock_irq(&mapping->i_pages); |
| 445 | break; |
Matthew Wilcox | c555772 | 2019-01-05 11:45:08 -0800 | [diff] [blame] | 446 | } else if (slot_locked(mapping, slot)) { |
| 447 | rcu_read_unlock(); |
| 448 | wait_entry_unlocked(mapping, index, &slot, entry); |
| 449 | rcu_read_lock(); |
Dan Williams | c2a7d2a | 2018-07-13 21:50:16 -0700 | [diff] [blame] | 450 | continue; |
| 451 | } |
| 452 | lock_slot(mapping, slot); |
| 453 | did_lock = true; |
| 454 | xa_unlock_irq(&mapping->i_pages); |
| 455 | break; |
| 456 | } |
| 457 | rcu_read_unlock(); |
| 458 | |
| 459 | return did_lock; |
| 460 | } |
| 461 | |
| 462 | void dax_unlock_mapping_entry(struct page *page) |
| 463 | { |
| 464 | struct address_space *mapping = page->mapping; |
| 465 | struct inode *inode = mapping->host; |
| 466 | |
| 467 | if (S_ISCHR(inode->i_mode)) |
| 468 | return; |
| 469 | |
| 470 | unlock_mapping_entry(mapping, page->index); |
| 471 | } |
| 472 | |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 473 | /* |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 474 | * Find radix tree entry at given index. If it points to an exceptional entry, |
| 475 | * return it with the radix tree entry locked. If the radix tree doesn't |
| 476 | * contain given index, create an empty exceptional entry for the index and |
| 477 | * return with it locked. |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 478 | * |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 479 | * When requesting an entry with size RADIX_DAX_PMD, grab_mapping_entry() will |
| 480 | * either return that locked entry or will return an error. This error will |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 481 | * happen if there are any 4k entries within the 2MiB range that we are |
| 482 | * requesting. |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 483 | * |
| 484 | * We always favor 4k entries over 2MiB entries. There isn't a flow where we |
| 485 | * evict 4k entries in order to 'upgrade' them to a 2MiB entry. A 2MiB |
| 486 | * insertion will fail if it finds any 4k entries already in the tree, and a |
| 487 | * 4k insertion will cause an existing 2MiB entry to be unmapped and |
| 488 | * downgraded to 4k entries. This happens for both 2MiB huge zero pages as |
| 489 | * well as 2MiB empty entries. |
| 490 | * |
| 491 | * The exception to this downgrade path is for 2MiB DAX PMD entries that have |
| 492 | * real storage backing them. We will leave these real 2MiB DAX entries in |
| 493 | * the tree, and PTE writes will simply dirty the entire 2MiB DAX entry. |
| 494 | * |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 495 | * Note: Unlike filemap_fault() we don't honor FAULT_FLAG_RETRY flags. For |
| 496 | * persistent memory the benefit is doubtful. We can add that later if we can |
| 497 | * show it helps. |
| 498 | */ |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 499 | static void *grab_mapping_entry(struct address_space *mapping, pgoff_t index, |
| 500 | unsigned long size_flag) |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 501 | { |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 502 | bool pmd_downgrade = false; /* splitting 2MiB entry into 4k entries? */ |
Ross Zwisler | e3ad61c | 2016-11-08 11:32:12 +1100 | [diff] [blame] | 503 | void *entry, **slot; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 504 | |
| 505 | restart: |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 506 | xa_lock_irq(&mapping->i_pages); |
Ross Zwisler | e3ad61c | 2016-11-08 11:32:12 +1100 | [diff] [blame] | 507 | entry = get_unlocked_mapping_entry(mapping, index, &slot); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 508 | |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 509 | if (WARN_ON_ONCE(entry && !radix_tree_exceptional_entry(entry))) { |
| 510 | entry = ERR_PTR(-EIO); |
| 511 | goto out_unlock; |
| 512 | } |
| 513 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 514 | if (entry) { |
| 515 | if (size_flag & RADIX_DAX_PMD) { |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 516 | if (dax_is_pte_entry(entry)) { |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 517 | put_unlocked_mapping_entry(mapping, index, |
| 518 | entry); |
| 519 | entry = ERR_PTR(-EEXIST); |
| 520 | goto out_unlock; |
| 521 | } |
| 522 | } else { /* trying to grab a PTE entry */ |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 523 | if (dax_is_pmd_entry(entry) && |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 524 | (dax_is_zero_entry(entry) || |
| 525 | dax_is_empty_entry(entry))) { |
| 526 | pmd_downgrade = true; |
| 527 | } |
| 528 | } |
| 529 | } |
| 530 | |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 531 | /* No entry for given index? Make sure radix tree is big enough. */ |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 532 | if (!entry || pmd_downgrade) { |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 533 | int err; |
| 534 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 535 | if (pmd_downgrade) { |
| 536 | /* |
| 537 | * Make sure 'entry' remains valid while we drop |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 538 | * the i_pages lock. |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 539 | */ |
| 540 | entry = lock_slot(mapping, slot); |
| 541 | } |
| 542 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 543 | xa_unlock_irq(&mapping->i_pages); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 544 | /* |
| 545 | * Besides huge zero pages the only other thing that gets |
| 546 | * downgraded are empty entries which don't need to be |
| 547 | * unmapped. |
| 548 | */ |
| 549 | if (pmd_downgrade && dax_is_zero_entry(entry)) |
Matthew Wilcox | 977fbdc | 2018-01-31 16:17:36 -0800 | [diff] [blame] | 550 | unmap_mapping_pages(mapping, index & ~PG_PMD_COLOUR, |
| 551 | PG_PMD_NR, false); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 552 | |
Jan Kara | 0cb80b4 | 2016-12-12 21:34:12 -0500 | [diff] [blame] | 553 | err = radix_tree_preload( |
| 554 | mapping_gfp_mask(mapping) & ~__GFP_HIGHMEM); |
| 555 | if (err) { |
| 556 | if (pmd_downgrade) |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 557 | put_locked_mapping_entry(mapping, index); |
Jan Kara | 0cb80b4 | 2016-12-12 21:34:12 -0500 | [diff] [blame] | 558 | return ERR_PTR(err); |
| 559 | } |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 560 | xa_lock_irq(&mapping->i_pages); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 561 | |
Ross Zwisler | e11f8b7 | 2017-04-07 16:04:57 -0700 | [diff] [blame] | 562 | if (!entry) { |
| 563 | /* |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 564 | * We needed to drop the i_pages lock while calling |
Ross Zwisler | e11f8b7 | 2017-04-07 16:04:57 -0700 | [diff] [blame] | 565 | * radix_tree_preload() and we didn't have an entry to |
| 566 | * lock. See if another thread inserted an entry at |
| 567 | * our index during this time. |
| 568 | */ |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 569 | entry = __radix_tree_lookup(&mapping->i_pages, index, |
Ross Zwisler | e11f8b7 | 2017-04-07 16:04:57 -0700 | [diff] [blame] | 570 | NULL, &slot); |
| 571 | if (entry) { |
| 572 | radix_tree_preload_end(); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 573 | xa_unlock_irq(&mapping->i_pages); |
Ross Zwisler | e11f8b7 | 2017-04-07 16:04:57 -0700 | [diff] [blame] | 574 | goto restart; |
| 575 | } |
| 576 | } |
| 577 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 578 | if (pmd_downgrade) { |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 579 | dax_disassociate_entry(entry, mapping, false); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 580 | radix_tree_delete(&mapping->i_pages, index); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 581 | mapping->nrexceptional--; |
| 582 | dax_wake_mapping_entry_waiter(mapping, index, entry, |
| 583 | true); |
| 584 | } |
| 585 | |
| 586 | entry = dax_radix_locked_entry(0, size_flag | RADIX_DAX_EMPTY); |
| 587 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 588 | err = __radix_tree_insert(&mapping->i_pages, index, |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 589 | dax_radix_order(entry), entry); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 590 | radix_tree_preload_end(); |
| 591 | if (err) { |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 592 | xa_unlock_irq(&mapping->i_pages); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 593 | /* |
Ross Zwisler | e11f8b7 | 2017-04-07 16:04:57 -0700 | [diff] [blame] | 594 | * Our insertion of a DAX entry failed, most likely |
| 595 | * because we were inserting a PMD entry and it |
| 596 | * collided with a PTE sized entry at a different |
| 597 | * index in the PMD range. We haven't inserted |
| 598 | * anything into the radix tree and have no waiters to |
| 599 | * wake. |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 600 | */ |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 601 | return ERR_PTR(err); |
| 602 | } |
| 603 | /* Good, we have inserted empty locked entry into the tree. */ |
| 604 | mapping->nrexceptional++; |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 605 | xa_unlock_irq(&mapping->i_pages); |
Ross Zwisler | e3ad61c | 2016-11-08 11:32:12 +1100 | [diff] [blame] | 606 | return entry; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 607 | } |
Ross Zwisler | e3ad61c | 2016-11-08 11:32:12 +1100 | [diff] [blame] | 608 | entry = lock_slot(mapping, slot); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 609 | out_unlock: |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 610 | xa_unlock_irq(&mapping->i_pages); |
Ross Zwisler | e3ad61c | 2016-11-08 11:32:12 +1100 | [diff] [blame] | 611 | return entry; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 612 | } |
| 613 | |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 614 | /** |
| 615 | * dax_layout_busy_page - find first pinned page in @mapping |
| 616 | * @mapping: address space to scan for a page with ref count > 1 |
| 617 | * |
| 618 | * DAX requires ZONE_DEVICE mapped pages. These pages are never |
| 619 | * 'onlined' to the page allocator so they are considered idle when |
| 620 | * page->count == 1. A filesystem uses this interface to determine if |
| 621 | * any page in the mapping is busy, i.e. for DMA, or other |
| 622 | * get_user_pages() usages. |
| 623 | * |
| 624 | * It is expected that the filesystem is holding locks to block the |
| 625 | * establishment of new mappings in this address_space. I.e. it expects |
| 626 | * to be able to run unmap_mapping_range() and subsequently not race |
| 627 | * mapping_mapped() becoming true. |
| 628 | */ |
| 629 | struct page *dax_layout_busy_page(struct address_space *mapping) |
| 630 | { |
| 631 | pgoff_t indices[PAGEVEC_SIZE]; |
| 632 | struct page *page = NULL; |
| 633 | struct pagevec pvec; |
| 634 | pgoff_t index, end; |
| 635 | unsigned i; |
| 636 | |
| 637 | /* |
| 638 | * In the 'limited' case get_user_pages() for dax is disabled. |
| 639 | */ |
| 640 | if (IS_ENABLED(CONFIG_FS_DAX_LIMITED)) |
| 641 | return NULL; |
| 642 | |
| 643 | if (!dax_mapping(mapping) || !mapping_mapped(mapping)) |
| 644 | return NULL; |
| 645 | |
| 646 | pagevec_init(&pvec); |
| 647 | index = 0; |
| 648 | end = -1; |
| 649 | |
| 650 | /* |
| 651 | * If we race get_user_pages_fast() here either we'll see the |
| 652 | * elevated page count in the pagevec_lookup and wait, or |
| 653 | * get_user_pages_fast() will see that the page it took a reference |
| 654 | * against is no longer mapped in the page tables and bail to the |
| 655 | * get_user_pages() slow path. The slow path is protected by |
| 656 | * pte_lock() and pmd_lock(). New references are not taken without |
| 657 | * holding those locks, and unmap_mapping_range() will not zero the |
| 658 | * pte or pmd without holding the respective lock, so we are |
| 659 | * guaranteed to either see new references or prevent new |
| 660 | * references from being established. |
| 661 | */ |
Vivek Goyal | 2afa6c1 | 2019-08-02 15:29:56 -0400 | [diff] [blame] | 662 | unmap_mapping_range(mapping, 0, 0, 0); |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 663 | |
| 664 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
| 665 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
| 666 | indices)) { |
Dan Williams | d778214 | 2018-10-06 10:56:11 -0700 | [diff] [blame] | 667 | pgoff_t nr_pages = 1; |
| 668 | |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 669 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 670 | struct page *pvec_ent = pvec.pages[i]; |
| 671 | void *entry; |
| 672 | |
| 673 | index = indices[i]; |
| 674 | if (index >= end) |
| 675 | break; |
| 676 | |
Ross Zwisler | cdbf889 | 2018-07-29 16:59:16 -0400 | [diff] [blame] | 677 | if (WARN_ON_ONCE( |
| 678 | !radix_tree_exceptional_entry(pvec_ent))) |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 679 | continue; |
| 680 | |
| 681 | xa_lock_irq(&mapping->i_pages); |
| 682 | entry = get_unlocked_mapping_entry(mapping, index, NULL); |
Dan Williams | d778214 | 2018-10-06 10:56:11 -0700 | [diff] [blame] | 683 | if (entry) { |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 684 | page = dax_busy_page(entry); |
Dan Williams | d778214 | 2018-10-06 10:56:11 -0700 | [diff] [blame] | 685 | /* |
| 686 | * Account for multi-order entries at |
| 687 | * the end of the pagevec. |
| 688 | */ |
| 689 | if (i + 1 >= pagevec_count(&pvec)) |
| 690 | nr_pages = 1UL << dax_radix_order(entry); |
| 691 | } |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 692 | put_unlocked_mapping_entry(mapping, index, entry); |
| 693 | xa_unlock_irq(&mapping->i_pages); |
| 694 | if (page) |
| 695 | break; |
| 696 | } |
Ross Zwisler | cdbf889 | 2018-07-29 16:59:16 -0400 | [diff] [blame] | 697 | |
| 698 | /* |
| 699 | * We don't expect normal struct page entries to exist in our |
| 700 | * tree, but we keep these pagevec calls so that this code is |
| 701 | * consistent with the common pattern for handling pagevecs |
| 702 | * throughout the kernel. |
| 703 | */ |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 704 | pagevec_remove_exceptionals(&pvec); |
| 705 | pagevec_release(&pvec); |
Dan Williams | d778214 | 2018-10-06 10:56:11 -0700 | [diff] [blame] | 706 | index += nr_pages; |
Dan Williams | 5fac740 | 2018-03-09 17:44:31 -0800 | [diff] [blame] | 707 | |
| 708 | if (page) |
| 709 | break; |
| 710 | } |
| 711 | return page; |
| 712 | } |
| 713 | EXPORT_SYMBOL_GPL(dax_layout_busy_page); |
| 714 | |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 715 | static int __dax_invalidate_mapping_entry(struct address_space *mapping, |
| 716 | pgoff_t index, bool trunc) |
| 717 | { |
| 718 | int ret = 0; |
| 719 | void *entry; |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 720 | struct radix_tree_root *pages = &mapping->i_pages; |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 721 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 722 | xa_lock_irq(pages); |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 723 | entry = get_unlocked_mapping_entry(mapping, index, NULL); |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 724 | if (!entry || WARN_ON_ONCE(!radix_tree_exceptional_entry(entry))) |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 725 | goto out; |
| 726 | if (!trunc && |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 727 | (radix_tree_tag_get(pages, index, PAGECACHE_TAG_DIRTY) || |
| 728 | radix_tree_tag_get(pages, index, PAGECACHE_TAG_TOWRITE))) |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 729 | goto out; |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 730 | dax_disassociate_entry(entry, mapping, trunc); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 731 | radix_tree_delete(pages, index); |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 732 | mapping->nrexceptional--; |
| 733 | ret = 1; |
| 734 | out: |
| 735 | put_unlocked_mapping_entry(mapping, index, entry); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 736 | xa_unlock_irq(pages); |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 737 | return ret; |
| 738 | } |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 739 | /* |
| 740 | * Delete exceptional DAX entry at @index from @mapping. Wait for radix tree |
| 741 | * entry to get unlocked before deleting it. |
| 742 | */ |
| 743 | int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index) |
| 744 | { |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 745 | int ret = __dax_invalidate_mapping_entry(mapping, index, true); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 746 | |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 747 | /* |
| 748 | * This gets called from truncate / punch_hole path. As such, the caller |
| 749 | * must hold locks protecting against concurrent modifications of the |
| 750 | * radix tree (usually fs-private i_mmap_sem for writing). Since the |
| 751 | * caller has seen exceptional entry for this index, we better find it |
| 752 | * at that index as well... |
| 753 | */ |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 754 | WARN_ON_ONCE(!ret); |
| 755 | return ret; |
| 756 | } |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 757 | |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 758 | /* |
Jan Kara | c6dcf52 | 2016-08-10 17:22:44 +0200 | [diff] [blame] | 759 | * Invalidate exceptional DAX entry if it is clean. |
| 760 | */ |
| 761 | int dax_invalidate_mapping_entry_sync(struct address_space *mapping, |
| 762 | pgoff_t index) |
| 763 | { |
| 764 | return __dax_invalidate_mapping_entry(mapping, index, false); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 765 | } |
| 766 | |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 767 | static int copy_user_dax(struct block_device *bdev, struct dax_device *dax_dev, |
| 768 | sector_t sector, size_t size, struct page *to, |
| 769 | unsigned long vaddr) |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 770 | { |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 771 | void *vto, *kaddr; |
| 772 | pgoff_t pgoff; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 773 | long rc; |
| 774 | int id; |
Ross Zwisler | e2e0539 | 2015-08-18 13:55:41 -0600 | [diff] [blame] | 775 | |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 776 | rc = bdev_dax_pgoff(bdev, sector, size, &pgoff); |
| 777 | if (rc) |
| 778 | return rc; |
| 779 | |
| 780 | id = dax_read_lock(); |
Huaisheng Ye | 86ed913 | 2018-07-30 15:15:48 +0800 | [diff] [blame] | 781 | rc = dax_direct_access(dax_dev, pgoff, PHYS_PFN(size), &kaddr, NULL); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 782 | if (rc < 0) { |
| 783 | dax_read_unlock(id); |
| 784 | return rc; |
| 785 | } |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 786 | vto = kmap_atomic(to); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 787 | copy_user_page(vto, (void __force *)kaddr, vaddr, to); |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 788 | kunmap_atomic(vto); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 789 | dax_read_unlock(id); |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 790 | return 0; |
| 791 | } |
| 792 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 793 | /* |
| 794 | * By this point grab_mapping_entry() has ensured that we have a locked entry |
| 795 | * of the appropriate size so we don't have to worry about downgrading PMDs to |
| 796 | * PTEs. If we happen to be trying to insert a PTE and there is a PMD |
| 797 | * already in the tree, we will skip the insertion and just dirty the PMD as |
| 798 | * appropriate. |
| 799 | */ |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 800 | static void *dax_insert_mapping_entry(struct address_space *mapping, |
| 801 | struct vm_fault *vmf, |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 802 | void *entry, pfn_t pfn_t, |
Jan Kara | f5b7b74 | 2017-11-01 16:36:40 +0100 | [diff] [blame] | 803 | unsigned long flags, bool dirty) |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 804 | { |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 805 | struct radix_tree_root *pages = &mapping->i_pages; |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 806 | unsigned long pfn = pfn_t_to_pfn(pfn_t); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 807 | pgoff_t index = vmf->pgoff; |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 808 | void *new_entry; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 809 | |
Jan Kara | f5b7b74 | 2017-11-01 16:36:40 +0100 | [diff] [blame] | 810 | if (dirty) |
Dmitry Monakhov | d2b2a28 | 2016-02-05 15:36:55 -0800 | [diff] [blame] | 811 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 812 | |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 813 | if (dax_is_zero_entry(entry) && !(flags & RADIX_DAX_ZERO_PAGE)) { |
| 814 | /* we are replacing a zero page with block mapping */ |
| 815 | if (dax_is_pmd_entry(entry)) |
Matthew Wilcox | 977fbdc | 2018-01-31 16:17:36 -0800 | [diff] [blame] | 816 | unmap_mapping_pages(mapping, index & ~PG_PMD_COLOUR, |
| 817 | PG_PMD_NR, false); |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 818 | else /* pte entry */ |
Matthew Wilcox | 977fbdc | 2018-01-31 16:17:36 -0800 | [diff] [blame] | 819 | unmap_mapping_pages(mapping, vmf->pgoff, 1, false); |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 820 | } |
| 821 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 822 | xa_lock_irq(pages); |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 823 | new_entry = dax_radix_locked_entry(pfn, flags); |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 824 | if (dax_entry_size(entry) != dax_entry_size(new_entry)) { |
| 825 | dax_disassociate_entry(entry, mapping, false); |
Dan Williams | 73449da | 2018-07-13 21:49:50 -0700 | [diff] [blame] | 826 | dax_associate_entry(new_entry, mapping, vmf->vma, vmf->address); |
Dan Williams | d2c997c | 2017-12-22 22:02:48 -0800 | [diff] [blame] | 827 | } |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 828 | |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 829 | if (dax_is_zero_entry(entry) || dax_is_empty_entry(entry)) { |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 830 | /* |
| 831 | * Only swap our new entry into the radix tree if the current |
| 832 | * entry is a zero page or an empty entry. If a normal PTE or |
| 833 | * PMD entry is already in the tree, we leave it alone. This |
| 834 | * means that if we are trying to insert a PTE and the |
| 835 | * existing entry is a PMD, we will just leave the PMD in the |
| 836 | * tree and dirty it if necessary. |
| 837 | */ |
Johannes Weiner | f794243 | 2016-12-12 16:43:41 -0800 | [diff] [blame] | 838 | struct radix_tree_node *node; |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 839 | void **slot; |
| 840 | void *ret; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 841 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 842 | ret = __radix_tree_lookup(pages, index, &node, &slot); |
Jan Kara | ac401cc | 2016-05-12 18:29:18 +0200 | [diff] [blame] | 843 | WARN_ON_ONCE(ret != entry); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 844 | __radix_tree_replace(pages, node, slot, |
Mel Gorman | c7df8ad | 2017-11-15 17:37:41 -0800 | [diff] [blame] | 845 | new_entry, NULL); |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 846 | entry = new_entry; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 847 | } |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 848 | |
Jan Kara | f5b7b74 | 2017-11-01 16:36:40 +0100 | [diff] [blame] | 849 | if (dirty) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 850 | radix_tree_tag_set(pages, index, PAGECACHE_TAG_DIRTY); |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 851 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 852 | xa_unlock_irq(pages); |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 853 | return entry; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 854 | } |
| 855 | |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 856 | static inline unsigned long |
| 857 | pgoff_address(pgoff_t pgoff, struct vm_area_struct *vma) |
| 858 | { |
| 859 | unsigned long address; |
| 860 | |
| 861 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
| 862 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); |
| 863 | return address; |
| 864 | } |
| 865 | |
| 866 | /* Walk all mappings of a given index of a file and writeprotect them */ |
| 867 | static void dax_mapping_entry_mkclean(struct address_space *mapping, |
| 868 | pgoff_t index, unsigned long pfn) |
| 869 | { |
| 870 | struct vm_area_struct *vma; |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 871 | pte_t pte, *ptep = NULL; |
| 872 | pmd_t *pmdp = NULL; |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 873 | spinlock_t *ptl; |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 874 | |
| 875 | i_mmap_lock_read(mapping); |
| 876 | vma_interval_tree_foreach(vma, &mapping->i_mmap, index, index) { |
Jérôme Glisse | a4d1a88 | 2017-08-31 17:17:26 -0400 | [diff] [blame] | 877 | unsigned long address, start, end; |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 878 | |
| 879 | cond_resched(); |
| 880 | |
| 881 | if (!(vma->vm_flags & VM_SHARED)) |
| 882 | continue; |
| 883 | |
| 884 | address = pgoff_address(index, vma); |
Jérôme Glisse | a4d1a88 | 2017-08-31 17:17:26 -0400 | [diff] [blame] | 885 | |
| 886 | /* |
| 887 | * Note because we provide start/end to follow_pte_pmd it will |
| 888 | * call mmu_notifier_invalidate_range_start() on our behalf |
| 889 | * before taking any lock. |
| 890 | */ |
| 891 | if (follow_pte_pmd(vma->vm_mm, address, &start, &end, &ptep, &pmdp, &ptl)) |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 892 | continue; |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 893 | |
Jérôme Glisse | 0f10851 | 2017-11-15 17:34:07 -0800 | [diff] [blame] | 894 | /* |
| 895 | * No need to call mmu_notifier_invalidate_range() as we are |
| 896 | * downgrading page table protection not changing it to point |
| 897 | * to a new page. |
| 898 | * |
Mike Rapoport | ad56b73 | 2018-03-21 21:22:47 +0200 | [diff] [blame] | 899 | * See Documentation/vm/mmu_notifier.rst |
Jérôme Glisse | 0f10851 | 2017-11-15 17:34:07 -0800 | [diff] [blame] | 900 | */ |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 901 | if (pmdp) { |
| 902 | #ifdef CONFIG_FS_DAX_PMD |
| 903 | pmd_t pmd; |
| 904 | |
| 905 | if (pfn != pmd_pfn(*pmdp)) |
| 906 | goto unlock_pmd; |
Linus Torvalds | f6f3732 | 2017-12-15 18:53:22 -0800 | [diff] [blame] | 907 | if (!pmd_dirty(*pmdp) && !pmd_write(*pmdp)) |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 908 | goto unlock_pmd; |
| 909 | |
| 910 | flush_cache_page(vma, address, pfn); |
Aneesh Kumar K.V | 38c5fce | 2019-05-13 17:19:11 -0700 | [diff] [blame] | 911 | pmd = pmdp_invalidate(vma, address, pmdp); |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 912 | pmd = pmd_wrprotect(pmd); |
| 913 | pmd = pmd_mkclean(pmd); |
| 914 | set_pmd_at(vma->vm_mm, address, pmdp, pmd); |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 915 | unlock_pmd: |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 916 | #endif |
Jan H. Schönherr | ee190ca | 2018-01-31 16:14:04 -0800 | [diff] [blame] | 917 | spin_unlock(ptl); |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 918 | } else { |
| 919 | if (pfn != pte_pfn(*ptep)) |
| 920 | goto unlock_pte; |
| 921 | if (!pte_dirty(*ptep) && !pte_write(*ptep)) |
| 922 | goto unlock_pte; |
| 923 | |
| 924 | flush_cache_page(vma, address, pfn); |
| 925 | pte = ptep_clear_flush(vma, address, ptep); |
| 926 | pte = pte_wrprotect(pte); |
| 927 | pte = pte_mkclean(pte); |
| 928 | set_pte_at(vma->vm_mm, address, ptep, pte); |
Ross Zwisler | f729c8c | 2017-01-10 16:57:24 -0800 | [diff] [blame] | 929 | unlock_pte: |
| 930 | pte_unmap_unlock(ptep, ptl); |
| 931 | } |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 932 | |
Jérôme Glisse | a4d1a88 | 2017-08-31 17:17:26 -0400 | [diff] [blame] | 933 | mmu_notifier_invalidate_range_end(vma->vm_mm, start, end); |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 934 | } |
| 935 | i_mmap_unlock_read(mapping); |
| 936 | } |
| 937 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 938 | static int dax_writeback_one(struct dax_device *dax_dev, |
| 939 | struct address_space *mapping, pgoff_t index, void *entry) |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 940 | { |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 941 | struct radix_tree_root *pages = &mapping->i_pages; |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 942 | void *entry2, **slot; |
| 943 | unsigned long pfn; |
| 944 | long ret = 0; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 945 | size_t size; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 946 | |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 947 | /* |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 948 | * A page got tagged dirty in DAX mapping? Something is seriously |
| 949 | * wrong. |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 950 | */ |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 951 | if (WARN_ON(!radix_tree_exceptional_entry(entry))) |
| 952 | return -EIO; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 953 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 954 | xa_lock_irq(pages); |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 955 | entry2 = get_unlocked_mapping_entry(mapping, index, &slot); |
| 956 | /* Entry got punched out / reallocated? */ |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 957 | if (!entry2 || WARN_ON_ONCE(!radix_tree_exceptional_entry(entry2))) |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 958 | goto put_unlocked; |
| 959 | /* |
| 960 | * Entry got reallocated elsewhere? No need to writeback. We have to |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 961 | * compare pfns as we must not bail out due to difference in lockbit |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 962 | * or entry type. |
| 963 | */ |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 964 | if (dax_radix_pfn(entry2) != dax_radix_pfn(entry)) |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 965 | goto put_unlocked; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 966 | if (WARN_ON_ONCE(dax_is_empty_entry(entry) || |
| 967 | dax_is_zero_entry(entry))) { |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 968 | ret = -EIO; |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 969 | goto put_unlocked; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 970 | } |
| 971 | |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 972 | /* Another fsync thread may have already written back this entry */ |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 973 | if (!radix_tree_tag_get(pages, index, PAGECACHE_TAG_TOWRITE)) |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 974 | goto put_unlocked; |
| 975 | /* Lock the entry to serialize with page faults */ |
| 976 | entry = lock_slot(mapping, slot); |
| 977 | /* |
| 978 | * We can clear the tag now but we have to be careful so that concurrent |
| 979 | * dax_writeback_one() calls for the same index cannot finish before we |
| 980 | * actually flush the caches. This is achieved as the calls will look |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 981 | * at the entry only under the i_pages lock and once they do that |
| 982 | * they will see the entry locked and wait for it to unlock. |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 983 | */ |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 984 | radix_tree_tag_clear(pages, index, PAGECACHE_TAG_TOWRITE); |
| 985 | xa_unlock_irq(pages); |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 986 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 987 | /* |
| 988 | * Even if dax_writeback_mapping_range() was given a wbc->range_start |
| 989 | * in the middle of a PMD, the 'index' we are given will be aligned to |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 990 | * the start index of the PMD, as will the pfn we pull from 'entry'. |
| 991 | * This allows us to flush for PMD_SIZE and not have to worry about |
| 992 | * partial PMD writebacks. |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 993 | */ |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 994 | pfn = dax_radix_pfn(entry); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 995 | size = PAGE_SIZE << dax_radix_order(entry); |
| 996 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 997 | dax_mapping_entry_mkclean(mapping, index, pfn); |
| 998 | dax_flush(dax_dev, page_address(pfn_to_page(pfn)), size); |
Jan Kara | 4b4bb46 | 2016-12-14 15:07:53 -0800 | [diff] [blame] | 999 | /* |
| 1000 | * After we have flushed the cache, we can clear the dirty tag. There |
| 1001 | * cannot be new dirty data in the pfn after the flush has completed as |
| 1002 | * the pfn mappings are writeprotected and fault waits for mapping |
| 1003 | * entry lock. |
| 1004 | */ |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 1005 | xa_lock_irq(pages); |
| 1006 | radix_tree_tag_clear(pages, index, PAGECACHE_TAG_DIRTY); |
| 1007 | xa_unlock_irq(pages); |
Ross Zwisler | f9bc3a0 | 2017-05-08 16:00:13 -0700 | [diff] [blame] | 1008 | trace_dax_writeback_one(mapping->host, index, size >> PAGE_SHIFT); |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1009 | put_locked_mapping_entry(mapping, index); |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1010 | return ret; |
| 1011 | |
Jan Kara | a6abc2c | 2016-12-14 15:07:47 -0800 | [diff] [blame] | 1012 | put_unlocked: |
| 1013 | put_unlocked_mapping_entry(mapping, index, entry2); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 1014 | xa_unlock_irq(pages); |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1015 | return ret; |
| 1016 | } |
| 1017 | |
| 1018 | /* |
| 1019 | * Flush the mapping to the persistent domain within the byte range of [start, |
| 1020 | * end]. This is required by data integrity operations to ensure file data is |
| 1021 | * on persistent storage prior to completion of the operation. |
| 1022 | */ |
Ross Zwisler | 7f6d5b5 | 2016-02-26 15:19:55 -0800 | [diff] [blame] | 1023 | int dax_writeback_mapping_range(struct address_space *mapping, |
| 1024 | struct block_device *bdev, struct writeback_control *wbc) |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1025 | { |
| 1026 | struct inode *inode = mapping->host; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1027 | pgoff_t start_index, end_index; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1028 | pgoff_t indices[PAGEVEC_SIZE]; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1029 | struct dax_device *dax_dev; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1030 | struct pagevec pvec; |
| 1031 | bool done = false; |
| 1032 | int i, ret = 0; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1033 | |
| 1034 | if (WARN_ON_ONCE(inode->i_blkbits != PAGE_SHIFT)) |
| 1035 | return -EIO; |
| 1036 | |
Ross Zwisler | 7f6d5b5 | 2016-02-26 15:19:55 -0800 | [diff] [blame] | 1037 | if (!mapping->nrexceptional || wbc->sync_mode != WB_SYNC_ALL) |
| 1038 | return 0; |
| 1039 | |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1040 | dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
| 1041 | if (!dax_dev) |
| 1042 | return -EIO; |
| 1043 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1044 | start_index = wbc->range_start >> PAGE_SHIFT; |
| 1045 | end_index = wbc->range_end >> PAGE_SHIFT; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1046 | |
Ross Zwisler | d14a3f4 | 2017-05-08 16:00:10 -0700 | [diff] [blame] | 1047 | trace_dax_writeback_range(inode, start_index, end_index); |
| 1048 | |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1049 | tag_pages_for_writeback(mapping, start_index, end_index); |
| 1050 | |
Mel Gorman | 8667982 | 2017-11-15 17:37:52 -0800 | [diff] [blame] | 1051 | pagevec_init(&pvec); |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1052 | while (!done) { |
| 1053 | pvec.nr = find_get_entries_tag(mapping, start_index, |
| 1054 | PAGECACHE_TAG_TOWRITE, PAGEVEC_SIZE, |
| 1055 | pvec.pages, indices); |
| 1056 | |
| 1057 | if (pvec.nr == 0) |
| 1058 | break; |
| 1059 | |
| 1060 | for (i = 0; i < pvec.nr; i++) { |
| 1061 | if (indices[i] > end_index) { |
| 1062 | done = true; |
| 1063 | break; |
| 1064 | } |
| 1065 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 1066 | ret = dax_writeback_one(dax_dev, mapping, indices[i], |
| 1067 | pvec.pages[i]); |
Jeff Layton | 819ec6b | 2017-07-06 07:02:27 -0400 | [diff] [blame] | 1068 | if (ret < 0) { |
| 1069 | mapping_set_error(mapping, ret); |
Ross Zwisler | d14a3f4 | 2017-05-08 16:00:10 -0700 | [diff] [blame] | 1070 | goto out; |
Jeff Layton | 819ec6b | 2017-07-06 07:02:27 -0400 | [diff] [blame] | 1071 | } |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1072 | } |
Jan Kara | 1eb643d | 2017-06-23 15:08:46 -0700 | [diff] [blame] | 1073 | start_index = indices[pvec.nr - 1] + 1; |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1074 | } |
Ross Zwisler | d14a3f4 | 2017-05-08 16:00:10 -0700 | [diff] [blame] | 1075 | out: |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1076 | put_dax(dax_dev); |
Ross Zwisler | d14a3f4 | 2017-05-08 16:00:10 -0700 | [diff] [blame] | 1077 | trace_dax_writeback_range_done(inode, start_index, end_index); |
| 1078 | return (ret < 0 ? ret : 0); |
Ross Zwisler | 9973c98 | 2016-01-22 15:10:47 -0800 | [diff] [blame] | 1079 | } |
| 1080 | EXPORT_SYMBOL_GPL(dax_writeback_mapping_range); |
| 1081 | |
Jan Kara | 31a6f1a | 2017-11-01 16:36:32 +0100 | [diff] [blame] | 1082 | static sector_t dax_iomap_sector(struct iomap *iomap, loff_t pos) |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 1083 | { |
Linus Torvalds | a3841f9 | 2017-11-17 09:51:57 -0800 | [diff] [blame] | 1084 | return (iomap->addr + (pos & PAGE_MASK) - iomap->offset) >> 9; |
Jan Kara | 31a6f1a | 2017-11-01 16:36:32 +0100 | [diff] [blame] | 1085 | } |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 1086 | |
Jan Kara | 5e161e4 | 2017-11-01 16:36:33 +0100 | [diff] [blame] | 1087 | static int dax_iomap_pfn(struct iomap *iomap, loff_t pos, size_t size, |
| 1088 | pfn_t *pfnp) |
| 1089 | { |
| 1090 | const sector_t sector = dax_iomap_sector(iomap, pos); |
| 1091 | pgoff_t pgoff; |
Jan Kara | 5e161e4 | 2017-11-01 16:36:33 +0100 | [diff] [blame] | 1092 | int id, rc; |
| 1093 | long length; |
| 1094 | |
| 1095 | rc = bdev_dax_pgoff(iomap->bdev, sector, size, &pgoff); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1096 | if (rc) |
| 1097 | return rc; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1098 | id = dax_read_lock(); |
Jan Kara | 5e161e4 | 2017-11-01 16:36:33 +0100 | [diff] [blame] | 1099 | length = dax_direct_access(iomap->dax_dev, pgoff, PHYS_PFN(size), |
Huaisheng Ye | 86ed913 | 2018-07-30 15:15:48 +0800 | [diff] [blame] | 1100 | NULL, pfnp); |
Jan Kara | 5e161e4 | 2017-11-01 16:36:33 +0100 | [diff] [blame] | 1101 | if (length < 0) { |
| 1102 | rc = length; |
| 1103 | goto out; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1104 | } |
Jan Kara | 5e161e4 | 2017-11-01 16:36:33 +0100 | [diff] [blame] | 1105 | rc = -EINVAL; |
| 1106 | if (PFN_PHYS(length) < size) |
| 1107 | goto out; |
| 1108 | if (pfn_t_to_pfn(*pfnp) & (PHYS_PFN(size)-1)) |
| 1109 | goto out; |
| 1110 | /* For larger pages we need devmap */ |
| 1111 | if (length > 1 && !pfn_t_devmap(*pfnp)) |
| 1112 | goto out; |
| 1113 | rc = 0; |
| 1114 | out: |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1115 | dax_read_unlock(id); |
Jan Kara | 5e161e4 | 2017-11-01 16:36:33 +0100 | [diff] [blame] | 1116 | return rc; |
Matthew Wilcox | f7ca90b | 2015-02-16 15:59:02 -0800 | [diff] [blame] | 1117 | } |
| 1118 | |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 1119 | /* |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1120 | * The user has performed a load from a hole in the file. Allocating a new |
| 1121 | * page in the file would cause excessive storage usage for workloads with |
| 1122 | * sparse files. Instead we insert a read-only mapping of the 4k zero page. |
| 1123 | * If this page is ever written to we will re-fault and change the mapping to |
| 1124 | * point to real DAX storage instead. |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 1125 | */ |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1126 | static vm_fault_t dax_load_hole(struct address_space *mapping, void *entry, |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 1127 | struct vm_fault *vmf) |
| 1128 | { |
| 1129 | struct inode *inode = mapping->host; |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1130 | unsigned long vaddr = vmf->address; |
Matthew Wilcox | b90ca5c | 2018-09-11 21:27:44 -0700 | [diff] [blame] | 1131 | pfn_t pfn = pfn_to_pfn_t(my_zero_pfn(vaddr)); |
| 1132 | vm_fault_t ret; |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 1133 | |
Matthew Wilcox | cc4a90a | 2018-06-02 19:39:37 -0700 | [diff] [blame] | 1134 | dax_insert_mapping_entry(mapping, vmf, entry, pfn, RADIX_DAX_ZERO_PAGE, |
| 1135 | false); |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1136 | ret = vmf_insert_mixed(vmf->vma, vaddr, pfn); |
Ross Zwisler | e30331f | 2017-09-06 16:18:39 -0700 | [diff] [blame] | 1137 | trace_dax_load_hole(inode, vmf, ret); |
| 1138 | return ret; |
| 1139 | } |
| 1140 | |
Vishal Verma | 4b0228f | 2016-04-21 15:13:46 -0400 | [diff] [blame] | 1141 | static bool dax_range_is_aligned(struct block_device *bdev, |
| 1142 | unsigned int offset, unsigned int length) |
| 1143 | { |
| 1144 | unsigned short sector_size = bdev_logical_block_size(bdev); |
| 1145 | |
| 1146 | if (!IS_ALIGNED(offset, sector_size)) |
| 1147 | return false; |
| 1148 | if (!IS_ALIGNED(length, sector_size)) |
| 1149 | return false; |
| 1150 | |
| 1151 | return true; |
| 1152 | } |
| 1153 | |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1154 | int __dax_zero_page_range(struct block_device *bdev, |
| 1155 | struct dax_device *dax_dev, sector_t sector, |
| 1156 | unsigned int offset, unsigned int size) |
Christoph Hellwig | 679c8bd | 2016-05-09 10:47:04 +0200 | [diff] [blame] | 1157 | { |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1158 | if (dax_range_is_aligned(bdev, offset, size)) { |
| 1159 | sector_t start_sector = sector + (offset >> 9); |
Vishal Verma | 4b0228f | 2016-04-21 15:13:46 -0400 | [diff] [blame] | 1160 | |
| 1161 | return blkdev_issue_zeroout(bdev, start_sector, |
Linus Torvalds | 53ef7d0 | 2017-05-05 18:49:20 -0700 | [diff] [blame] | 1162 | size >> 9, GFP_NOFS, 0); |
Vishal Verma | 4b0228f | 2016-04-21 15:13:46 -0400 | [diff] [blame] | 1163 | } else { |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1164 | pgoff_t pgoff; |
| 1165 | long rc, id; |
| 1166 | void *kaddr; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1167 | |
Dan Williams | e84b83b | 2017-05-10 19:38:13 -0700 | [diff] [blame] | 1168 | rc = bdev_dax_pgoff(bdev, sector, PAGE_SIZE, &pgoff); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1169 | if (rc) |
| 1170 | return rc; |
| 1171 | |
| 1172 | id = dax_read_lock(); |
Huaisheng Ye | 86ed913 | 2018-07-30 15:15:48 +0800 | [diff] [blame] | 1173 | rc = dax_direct_access(dax_dev, pgoff, 1, &kaddr, NULL); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1174 | if (rc < 0) { |
| 1175 | dax_read_unlock(id); |
| 1176 | return rc; |
| 1177 | } |
Dan Williams | 81f5587 | 2017-05-29 13:12:20 -0700 | [diff] [blame] | 1178 | memset(kaddr + offset, 0, size); |
Mikulas Patocka | c3ca015 | 2017-08-31 21:47:43 -0400 | [diff] [blame] | 1179 | dax_flush(dax_dev, kaddr + offset, size); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1180 | dax_read_unlock(id); |
Vishal Verma | 4b0228f | 2016-04-21 15:13:46 -0400 | [diff] [blame] | 1181 | } |
Christoph Hellwig | 679c8bd | 2016-05-09 10:47:04 +0200 | [diff] [blame] | 1182 | return 0; |
| 1183 | } |
| 1184 | EXPORT_SYMBOL_GPL(__dax_zero_page_range); |
| 1185 | |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1186 | static loff_t |
Ross Zwisler | 11c59c9 | 2016-11-08 11:32:46 +1100 | [diff] [blame] | 1187 | dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data, |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1188 | struct iomap *iomap) |
| 1189 | { |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1190 | struct block_device *bdev = iomap->bdev; |
| 1191 | struct dax_device *dax_dev = iomap->dax_dev; |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1192 | struct iov_iter *iter = data; |
| 1193 | loff_t end = pos + length, done = 0; |
| 1194 | ssize_t ret = 0; |
Dan Williams | a77d478 | 2018-03-16 17:36:44 -0700 | [diff] [blame] | 1195 | size_t xfer; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1196 | int id; |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1197 | |
| 1198 | if (iov_iter_rw(iter) == READ) { |
| 1199 | end = min(end, i_size_read(inode)); |
| 1200 | if (pos >= end) |
| 1201 | return 0; |
| 1202 | |
| 1203 | if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN) |
| 1204 | return iov_iter_zero(min(length, end - pos), iter); |
| 1205 | } |
| 1206 | |
| 1207 | if (WARN_ON_ONCE(iomap->type != IOMAP_MAPPED)) |
| 1208 | return -EIO; |
| 1209 | |
Jan Kara | e3fce68 | 2016-08-10 17:10:28 +0200 | [diff] [blame] | 1210 | /* |
| 1211 | * Write can allocate block for an area which has a hole page mapped |
| 1212 | * into page tables. We have to tear down these mappings so that data |
| 1213 | * written by write(2) is visible in mmap. |
| 1214 | */ |
Jan Kara | cd65637 | 2017-05-12 15:46:50 -0700 | [diff] [blame] | 1215 | if (iomap->flags & IOMAP_F_NEW) { |
Jan Kara | e3fce68 | 2016-08-10 17:10:28 +0200 | [diff] [blame] | 1216 | invalidate_inode_pages2_range(inode->i_mapping, |
| 1217 | pos >> PAGE_SHIFT, |
| 1218 | (end - 1) >> PAGE_SHIFT); |
| 1219 | } |
| 1220 | |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1221 | id = dax_read_lock(); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1222 | while (pos < end) { |
| 1223 | unsigned offset = pos & (PAGE_SIZE - 1); |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1224 | const size_t size = ALIGN(length + offset, PAGE_SIZE); |
| 1225 | const sector_t sector = dax_iomap_sector(iomap, pos); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1226 | ssize_t map_len; |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1227 | pgoff_t pgoff; |
| 1228 | void *kaddr; |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1229 | |
Michal Hocko | d1908f5 | 2017-02-03 13:13:26 -0800 | [diff] [blame] | 1230 | if (fatal_signal_pending(current)) { |
| 1231 | ret = -EINTR; |
| 1232 | break; |
| 1233 | } |
| 1234 | |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1235 | ret = bdev_dax_pgoff(bdev, sector, size, &pgoff); |
| 1236 | if (ret) |
| 1237 | break; |
| 1238 | |
| 1239 | map_len = dax_direct_access(dax_dev, pgoff, PHYS_PFN(size), |
Huaisheng Ye | 86ed913 | 2018-07-30 15:15:48 +0800 | [diff] [blame] | 1240 | &kaddr, NULL); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1241 | if (map_len < 0) { |
| 1242 | ret = map_len; |
| 1243 | break; |
| 1244 | } |
| 1245 | |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1246 | map_len = PFN_PHYS(map_len); |
| 1247 | kaddr += offset; |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1248 | map_len -= offset; |
| 1249 | if (map_len > end - pos) |
| 1250 | map_len = end - pos; |
| 1251 | |
Ross Zwisler | a2e050f | 2017-09-06 16:18:54 -0700 | [diff] [blame] | 1252 | /* |
| 1253 | * The userspace address for the memory copy has already been |
| 1254 | * validated via access_ok() in either vfs_read() or |
| 1255 | * vfs_write(), depending on which operation we are doing. |
| 1256 | */ |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1257 | if (iov_iter_rw(iter) == WRITE) |
Dan Williams | a77d478 | 2018-03-16 17:36:44 -0700 | [diff] [blame] | 1258 | xfer = dax_copy_from_iter(dax_dev, pgoff, kaddr, |
Dan Williams | fec5377 | 2017-05-29 21:56:49 -0700 | [diff] [blame] | 1259 | map_len, iter); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1260 | else |
Dan Williams | a77d478 | 2018-03-16 17:36:44 -0700 | [diff] [blame] | 1261 | xfer = dax_copy_to_iter(dax_dev, pgoff, kaddr, |
Dan Williams | b3a9a0c | 2018-05-02 06:46:33 -0700 | [diff] [blame] | 1262 | map_len, iter); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1263 | |
Dan Williams | a77d478 | 2018-03-16 17:36:44 -0700 | [diff] [blame] | 1264 | pos += xfer; |
| 1265 | length -= xfer; |
| 1266 | done += xfer; |
| 1267 | |
| 1268 | if (xfer == 0) |
| 1269 | ret = -EFAULT; |
| 1270 | if (xfer < map_len) |
| 1271 | break; |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1272 | } |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1273 | dax_read_unlock(id); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1274 | |
| 1275 | return done ? done : ret; |
| 1276 | } |
| 1277 | |
| 1278 | /** |
Ross Zwisler | 11c59c9 | 2016-11-08 11:32:46 +1100 | [diff] [blame] | 1279 | * dax_iomap_rw - Perform I/O to a DAX file |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1280 | * @iocb: The control block for this I/O |
| 1281 | * @iter: The addresses to do I/O from or to |
| 1282 | * @ops: iomap ops passed from the file system |
| 1283 | * |
| 1284 | * This function performs read and write operations to directly mapped |
| 1285 | * persistent memory. The callers needs to take care of read/write exclusion |
| 1286 | * and evicting any page cache pages in the region under I/O. |
| 1287 | */ |
| 1288 | ssize_t |
Ross Zwisler | 11c59c9 | 2016-11-08 11:32:46 +1100 | [diff] [blame] | 1289 | dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter, |
Christoph Hellwig | 8ff6daa | 2017-01-27 23:20:26 -0800 | [diff] [blame] | 1290 | const struct iomap_ops *ops) |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1291 | { |
| 1292 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
| 1293 | struct inode *inode = mapping->host; |
| 1294 | loff_t pos = iocb->ki_pos, ret = 0, done = 0; |
| 1295 | unsigned flags = 0; |
| 1296 | |
Christoph Hellwig | 168316d | 2017-02-08 14:43:13 -0500 | [diff] [blame] | 1297 | if (iov_iter_rw(iter) == WRITE) { |
| 1298 | lockdep_assert_held_exclusive(&inode->i_rwsem); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1299 | flags |= IOMAP_WRITE; |
Christoph Hellwig | 168316d | 2017-02-08 14:43:13 -0500 | [diff] [blame] | 1300 | } else { |
| 1301 | lockdep_assert_held(&inode->i_rwsem); |
| 1302 | } |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1303 | |
Jeff Moyer | 5195d8c | 2020-02-05 14:15:58 -0500 | [diff] [blame] | 1304 | if (iocb->ki_flags & IOCB_NOWAIT) |
| 1305 | flags |= IOMAP_NOWAIT; |
| 1306 | |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1307 | while (iov_iter_count(iter)) { |
| 1308 | ret = iomap_apply(inode, pos, iov_iter_count(iter), flags, ops, |
Ross Zwisler | 11c59c9 | 2016-11-08 11:32:46 +1100 | [diff] [blame] | 1309 | iter, dax_iomap_actor); |
Christoph Hellwig | a254e56 | 2016-09-19 11:24:49 +1000 | [diff] [blame] | 1310 | if (ret <= 0) |
| 1311 | break; |
| 1312 | pos += ret; |
| 1313 | done += ret; |
| 1314 | } |
| 1315 | |
| 1316 | iocb->ki_pos += done; |
| 1317 | return done ? done : ret; |
| 1318 | } |
Ross Zwisler | 11c59c9 | 2016-11-08 11:32:46 +1100 | [diff] [blame] | 1319 | EXPORT_SYMBOL_GPL(dax_iomap_rw); |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1320 | |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1321 | static vm_fault_t dax_fault_return(int error) |
Jan Kara | 9f141d6 | 2016-10-19 14:34:31 +0200 | [diff] [blame] | 1322 | { |
| 1323 | if (error == 0) |
| 1324 | return VM_FAULT_NOPAGE; |
| 1325 | if (error == -ENOMEM) |
| 1326 | return VM_FAULT_OOM; |
| 1327 | return VM_FAULT_SIGBUS; |
| 1328 | } |
| 1329 | |
Dan Williams | aaa422c | 2017-11-13 16:38:44 -0800 | [diff] [blame] | 1330 | /* |
| 1331 | * MAP_SYNC on a dax mapping guarantees dirty metadata is |
| 1332 | * flushed on write-faults (non-cow), but not read-faults. |
| 1333 | */ |
| 1334 | static bool dax_fault_is_synchronous(unsigned long flags, |
| 1335 | struct vm_area_struct *vma, struct iomap *iomap) |
| 1336 | { |
| 1337 | return (flags & IOMAP_WRITE) && (vma->vm_flags & VM_SYNC) |
| 1338 | && (iomap->flags & IOMAP_F_DIRTY); |
| 1339 | } |
| 1340 | |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1341 | static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp, |
Jan Kara | c0b2462 | 2018-01-07 16:38:43 -0500 | [diff] [blame] | 1342 | int *iomap_errp, const struct iomap_ops *ops) |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1343 | { |
Jan Kara | a0987ad | 2017-11-01 16:36:34 +0100 | [diff] [blame] | 1344 | struct vm_area_struct *vma = vmf->vma; |
| 1345 | struct address_space *mapping = vma->vm_file->f_mapping; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1346 | struct inode *inode = mapping->host; |
Jan Kara | 1a29d85 | 2016-12-14 15:07:01 -0800 | [diff] [blame] | 1347 | unsigned long vaddr = vmf->address; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1348 | loff_t pos = (loff_t)vmf->pgoff << PAGE_SHIFT; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1349 | struct iomap iomap = { 0 }; |
Jan Kara | 9484ab1 | 2016-11-10 10:26:50 +1100 | [diff] [blame] | 1350 | unsigned flags = IOMAP_FAULT; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1351 | int error, major = 0; |
Jan Kara | d2c43ef | 2017-11-01 16:36:35 +0100 | [diff] [blame] | 1352 | bool write = vmf->flags & FAULT_FLAG_WRITE; |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1353 | bool sync; |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1354 | vm_fault_t ret = 0; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1355 | void *entry; |
Jan Kara | 1b5a1cb | 2017-11-01 16:36:36 +0100 | [diff] [blame] | 1356 | pfn_t pfn; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1357 | |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1358 | trace_dax_pte_fault(inode, vmf, ret); |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1359 | /* |
| 1360 | * Check whether offset isn't beyond end of file now. Caller is supposed |
| 1361 | * to hold locks serializing us with truncate / punch hole so this is |
| 1362 | * a reliable test. |
| 1363 | */ |
Ross Zwisler | a9c42b3 | 2017-05-08 16:00:00 -0700 | [diff] [blame] | 1364 | if (pos >= i_size_read(inode)) { |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1365 | ret = VM_FAULT_SIGBUS; |
Ross Zwisler | a9c42b3 | 2017-05-08 16:00:00 -0700 | [diff] [blame] | 1366 | goto out; |
| 1367 | } |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1368 | |
Jan Kara | d2c43ef | 2017-11-01 16:36:35 +0100 | [diff] [blame] | 1369 | if (write && !vmf->cow_page) |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1370 | flags |= IOMAP_WRITE; |
| 1371 | |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1372 | entry = grab_mapping_entry(mapping, vmf->pgoff, 0); |
| 1373 | if (IS_ERR(entry)) { |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1374 | ret = dax_fault_return(PTR_ERR(entry)); |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1375 | goto out; |
| 1376 | } |
| 1377 | |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1378 | /* |
Ross Zwisler | e209392 | 2017-06-02 14:46:37 -0700 | [diff] [blame] | 1379 | * It is possible, particularly with mixed reads & writes to private |
| 1380 | * mappings, that we have raced with a PMD fault that overlaps with |
| 1381 | * the PTE we need to set up. If so just return and the fault will be |
| 1382 | * retried. |
| 1383 | */ |
| 1384 | if (pmd_trans_huge(*vmf->pmd) || pmd_devmap(*vmf->pmd)) { |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1385 | ret = VM_FAULT_NOPAGE; |
Ross Zwisler | e209392 | 2017-06-02 14:46:37 -0700 | [diff] [blame] | 1386 | goto unlock_entry; |
| 1387 | } |
| 1388 | |
| 1389 | /* |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1390 | * Note that we don't bother to use iomap_apply here: DAX required |
| 1391 | * the file system block size to be equal the page size, which means |
| 1392 | * that we never have to deal with more than a single extent here. |
| 1393 | */ |
| 1394 | error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap); |
Jan Kara | c0b2462 | 2018-01-07 16:38:43 -0500 | [diff] [blame] | 1395 | if (iomap_errp) |
| 1396 | *iomap_errp = error; |
Ross Zwisler | a9c42b3 | 2017-05-08 16:00:00 -0700 | [diff] [blame] | 1397 | if (error) { |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1398 | ret = dax_fault_return(error); |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1399 | goto unlock_entry; |
Ross Zwisler | a9c42b3 | 2017-05-08 16:00:00 -0700 | [diff] [blame] | 1400 | } |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1401 | if (WARN_ON_ONCE(iomap.offset + iomap.length < pos + PAGE_SIZE)) { |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1402 | error = -EIO; /* fs corruption? */ |
| 1403 | goto error_finish_iomap; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1404 | } |
| 1405 | |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1406 | if (vmf->cow_page) { |
Jan Kara | 31a6f1a | 2017-11-01 16:36:32 +0100 | [diff] [blame] | 1407 | sector_t sector = dax_iomap_sector(&iomap, pos); |
| 1408 | |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1409 | switch (iomap.type) { |
| 1410 | case IOMAP_HOLE: |
| 1411 | case IOMAP_UNWRITTEN: |
| 1412 | clear_user_highpage(vmf->cow_page, vaddr); |
| 1413 | break; |
| 1414 | case IOMAP_MAPPED: |
Dan Williams | cccbce6 | 2017-01-27 13:31:42 -0800 | [diff] [blame] | 1415 | error = copy_user_dax(iomap.bdev, iomap.dax_dev, |
| 1416 | sector, PAGE_SIZE, vmf->cow_page, vaddr); |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1417 | break; |
| 1418 | default: |
| 1419 | WARN_ON_ONCE(1); |
| 1420 | error = -EIO; |
| 1421 | break; |
| 1422 | } |
| 1423 | |
| 1424 | if (error) |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1425 | goto error_finish_iomap; |
Jan Kara | b1aa812 | 2016-12-14 15:07:24 -0800 | [diff] [blame] | 1426 | |
| 1427 | __SetPageUptodate(vmf->cow_page); |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1428 | ret = finish_fault(vmf); |
| 1429 | if (!ret) |
| 1430 | ret = VM_FAULT_DONE_COW; |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1431 | goto finish_iomap; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1432 | } |
| 1433 | |
Dan Williams | aaa422c | 2017-11-13 16:38:44 -0800 | [diff] [blame] | 1434 | sync = dax_fault_is_synchronous(flags, vma, &iomap); |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1435 | |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1436 | switch (iomap.type) { |
| 1437 | case IOMAP_MAPPED: |
| 1438 | if (iomap.flags & IOMAP_F_NEW) { |
| 1439 | count_vm_event(PGMAJFAULT); |
Jan Kara | a0987ad | 2017-11-01 16:36:34 +0100 | [diff] [blame] | 1440 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1441 | major = VM_FAULT_MAJOR; |
| 1442 | } |
Jan Kara | 1b5a1cb | 2017-11-01 16:36:36 +0100 | [diff] [blame] | 1443 | error = dax_iomap_pfn(&iomap, pos, PAGE_SIZE, &pfn); |
| 1444 | if (error < 0) |
| 1445 | goto error_finish_iomap; |
| 1446 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 1447 | entry = dax_insert_mapping_entry(mapping, vmf, entry, pfn, |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1448 | 0, write && !sync); |
Jan Kara | 1b5a1cb | 2017-11-01 16:36:36 +0100 | [diff] [blame] | 1449 | |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1450 | /* |
| 1451 | * If we are doing synchronous page fault and inode needs fsync, |
| 1452 | * we can insert PTE into page tables only after that happens. |
| 1453 | * Skip insertion for now and return the pfn so that caller can |
| 1454 | * insert it after fsync is done. |
| 1455 | */ |
| 1456 | if (sync) { |
| 1457 | if (WARN_ON_ONCE(!pfnp)) { |
| 1458 | error = -EIO; |
| 1459 | goto error_finish_iomap; |
| 1460 | } |
| 1461 | *pfnp = pfn; |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1462 | ret = VM_FAULT_NEEDDSYNC | major; |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1463 | goto finish_iomap; |
| 1464 | } |
Jan Kara | 1b5a1cb | 2017-11-01 16:36:36 +0100 | [diff] [blame] | 1465 | trace_dax_insert_mapping(inode, vmf, entry); |
| 1466 | if (write) |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1467 | ret = vmf_insert_mixed_mkwrite(vma, vaddr, pfn); |
Jan Kara | 1b5a1cb | 2017-11-01 16:36:36 +0100 | [diff] [blame] | 1468 | else |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1469 | ret = vmf_insert_mixed(vma, vaddr, pfn); |
Jan Kara | 1b5a1cb | 2017-11-01 16:36:36 +0100 | [diff] [blame] | 1470 | |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1471 | goto finish_iomap; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1472 | case IOMAP_UNWRITTEN: |
| 1473 | case IOMAP_HOLE: |
Jan Kara | d2c43ef | 2017-11-01 16:36:35 +0100 | [diff] [blame] | 1474 | if (!write) { |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1475 | ret = dax_load_hole(mapping, entry, vmf); |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1476 | goto finish_iomap; |
Ross Zwisler | 1550290 | 2016-11-08 11:33:26 +1100 | [diff] [blame] | 1477 | } |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1478 | /*FALLTHRU*/ |
| 1479 | default: |
| 1480 | WARN_ON_ONCE(1); |
| 1481 | error = -EIO; |
| 1482 | break; |
| 1483 | } |
| 1484 | |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1485 | error_finish_iomap: |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1486 | ret = dax_fault_return(error); |
Jan Kara | 9f141d6 | 2016-10-19 14:34:31 +0200 | [diff] [blame] | 1487 | finish_iomap: |
| 1488 | if (ops->iomap_end) { |
| 1489 | int copied = PAGE_SIZE; |
| 1490 | |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1491 | if (ret & VM_FAULT_ERROR) |
Jan Kara | 9f141d6 | 2016-10-19 14:34:31 +0200 | [diff] [blame] | 1492 | copied = 0; |
| 1493 | /* |
| 1494 | * The fault is done by now and there's no way back (other |
| 1495 | * thread may be already happily using PTE we have installed). |
| 1496 | * Just ignore error from ->iomap_end since we cannot do much |
| 1497 | * with it. |
| 1498 | */ |
| 1499 | ops->iomap_end(inode, pos, PAGE_SIZE, copied, flags, &iomap); |
Ross Zwisler | 1550290 | 2016-11-08 11:33:26 +1100 | [diff] [blame] | 1500 | } |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1501 | unlock_entry: |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1502 | put_locked_mapping_entry(mapping, vmf->pgoff); |
Jan Kara | 13e451f | 2017-05-12 15:46:57 -0700 | [diff] [blame] | 1503 | out: |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1504 | trace_dax_pte_fault_done(inode, vmf, ret); |
| 1505 | return ret | major; |
Christoph Hellwig | a7d73fe | 2016-09-19 11:24:50 +1000 | [diff] [blame] | 1506 | } |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1507 | |
| 1508 | #ifdef CONFIG_FS_DAX_PMD |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1509 | static vm_fault_t dax_pmd_load_hole(struct vm_fault *vmf, struct iomap *iomap, |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1510 | void *entry) |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1511 | { |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1512 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
| 1513 | unsigned long pmd_addr = vmf->address & PMD_MASK; |
Ross Zwisler | 653b2ea | 2017-02-22 15:39:57 -0800 | [diff] [blame] | 1514 | struct inode *inode = mapping->host; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1515 | struct page *zero_page; |
Ross Zwisler | 653b2ea | 2017-02-22 15:39:57 -0800 | [diff] [blame] | 1516 | void *ret = NULL; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1517 | spinlock_t *ptl; |
| 1518 | pmd_t pmd_entry; |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 1519 | pfn_t pfn; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1520 | |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1521 | zero_page = mm_get_huge_zero_page(vmf->vma->vm_mm); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1522 | |
| 1523 | if (unlikely(!zero_page)) |
Ross Zwisler | 653b2ea | 2017-02-22 15:39:57 -0800 | [diff] [blame] | 1524 | goto fallback; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1525 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 1526 | pfn = page_to_pfn_t(zero_page); |
| 1527 | ret = dax_insert_mapping_entry(mapping, vmf, entry, pfn, |
Jan Kara | f5b7b74 | 2017-11-01 16:36:40 +0100 | [diff] [blame] | 1528 | RADIX_DAX_PMD | RADIX_DAX_ZERO_PAGE, false); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1529 | |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1530 | ptl = pmd_lock(vmf->vma->vm_mm, vmf->pmd); |
| 1531 | if (!pmd_none(*(vmf->pmd))) { |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1532 | spin_unlock(ptl); |
Ross Zwisler | 653b2ea | 2017-02-22 15:39:57 -0800 | [diff] [blame] | 1533 | goto fallback; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1534 | } |
| 1535 | |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1536 | pmd_entry = mk_pmd(zero_page, vmf->vma->vm_page_prot); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1537 | pmd_entry = pmd_mkhuge(pmd_entry); |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1538 | set_pmd_at(vmf->vma->vm_mm, pmd_addr, vmf->pmd, pmd_entry); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1539 | spin_unlock(ptl); |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1540 | trace_dax_pmd_load_hole(inode, vmf, zero_page, ret); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1541 | return VM_FAULT_NOPAGE; |
Ross Zwisler | 653b2ea | 2017-02-22 15:39:57 -0800 | [diff] [blame] | 1542 | |
| 1543 | fallback: |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1544 | trace_dax_pmd_load_hole_fallback(inode, vmf, zero_page, ret); |
Ross Zwisler | 653b2ea | 2017-02-22 15:39:57 -0800 | [diff] [blame] | 1545 | return VM_FAULT_FALLBACK; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1546 | } |
| 1547 | |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1548 | static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp, |
Dave Jiang | a2d5816 | 2017-02-24 14:56:59 -0800 | [diff] [blame] | 1549 | const struct iomap_ops *ops) |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1550 | { |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1551 | struct vm_area_struct *vma = vmf->vma; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1552 | struct address_space *mapping = vma->vm_file->f_mapping; |
Dave Jiang | d8a849e | 2017-02-22 15:40:03 -0800 | [diff] [blame] | 1553 | unsigned long pmd_addr = vmf->address & PMD_MASK; |
| 1554 | bool write = vmf->flags & FAULT_FLAG_WRITE; |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1555 | bool sync; |
Jan Kara | 9484ab1 | 2016-11-10 10:26:50 +1100 | [diff] [blame] | 1556 | unsigned int iomap_flags = (write ? IOMAP_WRITE : 0) | IOMAP_FAULT; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1557 | struct inode *inode = mapping->host; |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1558 | vm_fault_t result = VM_FAULT_FALLBACK; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1559 | struct iomap iomap = { 0 }; |
| 1560 | pgoff_t max_pgoff, pgoff; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1561 | void *entry; |
| 1562 | loff_t pos; |
| 1563 | int error; |
Jan Kara | 302a5e3 | 2017-11-01 16:36:37 +0100 | [diff] [blame] | 1564 | pfn_t pfn; |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1565 | |
Ross Zwisler | 282a8e0 | 2017-02-22 15:39:50 -0800 | [diff] [blame] | 1566 | /* |
| 1567 | * Check whether offset isn't beyond end of file now. Caller is |
| 1568 | * supposed to hold locks serializing us with truncate / punch hole so |
| 1569 | * this is a reliable test. |
| 1570 | */ |
| 1571 | pgoff = linear_page_index(vma, pmd_addr); |
Jeff Moyer | 957ac8c | 2017-11-14 20:37:27 -0500 | [diff] [blame] | 1572 | max_pgoff = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
Ross Zwisler | 282a8e0 | 2017-02-22 15:39:50 -0800 | [diff] [blame] | 1573 | |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1574 | trace_dax_pmd_fault(inode, vmf, max_pgoff, 0); |
Ross Zwisler | 282a8e0 | 2017-02-22 15:39:50 -0800 | [diff] [blame] | 1575 | |
Ross Zwisler | fffa281 | 2017-08-25 15:55:36 -0700 | [diff] [blame] | 1576 | /* |
| 1577 | * Make sure that the faulting address's PMD offset (color) matches |
| 1578 | * the PMD offset from the start of the file. This is necessary so |
| 1579 | * that a PMD range in the page table overlaps exactly with a PMD |
| 1580 | * range in the radix tree. |
| 1581 | */ |
| 1582 | if ((vmf->pgoff & PG_PMD_COLOUR) != |
| 1583 | ((vmf->address >> PAGE_SHIFT) & PG_PMD_COLOUR)) |
| 1584 | goto fallback; |
| 1585 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1586 | /* Fall back to PTEs if we're going to COW */ |
| 1587 | if (write && !(vma->vm_flags & VM_SHARED)) |
| 1588 | goto fallback; |
| 1589 | |
| 1590 | /* If the PMD would extend outside the VMA */ |
| 1591 | if (pmd_addr < vma->vm_start) |
| 1592 | goto fallback; |
| 1593 | if ((pmd_addr + PMD_SIZE) > vma->vm_end) |
| 1594 | goto fallback; |
| 1595 | |
Jeff Moyer | 957ac8c | 2017-11-14 20:37:27 -0500 | [diff] [blame] | 1596 | if (pgoff >= max_pgoff) { |
Ross Zwisler | 282a8e0 | 2017-02-22 15:39:50 -0800 | [diff] [blame] | 1597 | result = VM_FAULT_SIGBUS; |
| 1598 | goto out; |
| 1599 | } |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1600 | |
| 1601 | /* If the PMD would extend beyond the file size */ |
Jeff Moyer | 957ac8c | 2017-11-14 20:37:27 -0500 | [diff] [blame] | 1602 | if ((pgoff | PG_PMD_COLOUR) >= max_pgoff) |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1603 | goto fallback; |
| 1604 | |
| 1605 | /* |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1606 | * grab_mapping_entry() will make sure we get a 2MiB empty entry, a |
| 1607 | * 2MiB zero page entry or a DAX PMD. If it can't (because a 4k page |
| 1608 | * is already in the tree, for instance), it will return -EEXIST and |
| 1609 | * we just fall back to 4k entries. |
Jan Kara | 9f141d6 | 2016-10-19 14:34:31 +0200 | [diff] [blame] | 1610 | */ |
| 1611 | entry = grab_mapping_entry(mapping, pgoff, RADIX_DAX_PMD); |
| 1612 | if (IS_ERR(entry)) |
Ross Zwisler | 876f294 | 2017-05-12 15:47:00 -0700 | [diff] [blame] | 1613 | goto fallback; |
| 1614 | |
| 1615 | /* |
Ross Zwisler | e209392 | 2017-06-02 14:46:37 -0700 | [diff] [blame] | 1616 | * It is possible, particularly with mixed reads & writes to private |
| 1617 | * mappings, that we have raced with a PTE fault that overlaps with |
| 1618 | * the PMD we need to set up. If so just return and the fault will be |
| 1619 | * retried. |
| 1620 | */ |
| 1621 | if (!pmd_none(*vmf->pmd) && !pmd_trans_huge(*vmf->pmd) && |
| 1622 | !pmd_devmap(*vmf->pmd)) { |
| 1623 | result = 0; |
| 1624 | goto unlock_entry; |
| 1625 | } |
| 1626 | |
| 1627 | /* |
Ross Zwisler | 876f294 | 2017-05-12 15:47:00 -0700 | [diff] [blame] | 1628 | * Note that we don't use iomap_apply here. We aren't doing I/O, only |
| 1629 | * setting up a mapping, so really we're using iomap_begin() as a way |
| 1630 | * to look up our filesystem block. |
| 1631 | */ |
| 1632 | pos = (loff_t)pgoff << PAGE_SHIFT; |
| 1633 | error = ops->iomap_begin(inode, pos, PMD_SIZE, iomap_flags, &iomap); |
| 1634 | if (error) |
| 1635 | goto unlock_entry; |
| 1636 | |
| 1637 | if (iomap.offset + iomap.length < pos + PMD_SIZE) |
Jan Kara | 9f141d6 | 2016-10-19 14:34:31 +0200 | [diff] [blame] | 1638 | goto finish_iomap; |
| 1639 | |
Dan Williams | aaa422c | 2017-11-13 16:38:44 -0800 | [diff] [blame] | 1640 | sync = dax_fault_is_synchronous(iomap_flags, vma, &iomap); |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1641 | |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1642 | switch (iomap.type) { |
| 1643 | case IOMAP_MAPPED: |
Jan Kara | 302a5e3 | 2017-11-01 16:36:37 +0100 | [diff] [blame] | 1644 | error = dax_iomap_pfn(&iomap, pos, PMD_SIZE, &pfn); |
| 1645 | if (error < 0) |
| 1646 | goto finish_iomap; |
| 1647 | |
Dan Williams | 3fe0791 | 2017-10-14 17:13:45 -0700 | [diff] [blame] | 1648 | entry = dax_insert_mapping_entry(mapping, vmf, entry, pfn, |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1649 | RADIX_DAX_PMD, write && !sync); |
Jan Kara | 302a5e3 | 2017-11-01 16:36:37 +0100 | [diff] [blame] | 1650 | |
Jan Kara | caa51d2 | 2017-11-01 16:36:42 +0100 | [diff] [blame] | 1651 | /* |
| 1652 | * If we are doing synchronous page fault and inode needs fsync, |
| 1653 | * we can insert PMD into page tables only after that happens. |
| 1654 | * Skip insertion for now and return the pfn so that caller can |
| 1655 | * insert it after fsync is done. |
| 1656 | */ |
| 1657 | if (sync) { |
| 1658 | if (WARN_ON_ONCE(!pfnp)) |
| 1659 | goto finish_iomap; |
| 1660 | *pfnp = pfn; |
| 1661 | result = VM_FAULT_NEEDDSYNC; |
| 1662 | goto finish_iomap; |
| 1663 | } |
| 1664 | |
Jan Kara | 302a5e3 | 2017-11-01 16:36:37 +0100 | [diff] [blame] | 1665 | trace_dax_pmd_insert_mapping(inode, vmf, PMD_SIZE, pfn, entry); |
Dan Williams | 58db381 | 2019-05-13 17:15:33 -0700 | [diff] [blame] | 1666 | result = vmf_insert_pfn_pmd(vmf, pfn, write); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1667 | break; |
| 1668 | case IOMAP_UNWRITTEN: |
| 1669 | case IOMAP_HOLE: |
| 1670 | if (WARN_ON_ONCE(write)) |
Ross Zwisler | 876f294 | 2017-05-12 15:47:00 -0700 | [diff] [blame] | 1671 | break; |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1672 | result = dax_pmd_load_hole(vmf, &iomap, entry); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1673 | break; |
| 1674 | default: |
| 1675 | WARN_ON_ONCE(1); |
| 1676 | break; |
| 1677 | } |
| 1678 | |
Jan Kara | 9f141d6 | 2016-10-19 14:34:31 +0200 | [diff] [blame] | 1679 | finish_iomap: |
| 1680 | if (ops->iomap_end) { |
| 1681 | int copied = PMD_SIZE; |
| 1682 | |
| 1683 | if (result == VM_FAULT_FALLBACK) |
| 1684 | copied = 0; |
| 1685 | /* |
| 1686 | * The fault is done by now and there's no way back (other |
| 1687 | * thread may be already happily using PMD we have installed). |
| 1688 | * Just ignore error from ->iomap_end since we cannot do much |
| 1689 | * with it. |
| 1690 | */ |
| 1691 | ops->iomap_end(inode, pos, PMD_SIZE, copied, iomap_flags, |
| 1692 | &iomap); |
| 1693 | } |
Ross Zwisler | 876f294 | 2017-05-12 15:47:00 -0700 | [diff] [blame] | 1694 | unlock_entry: |
Ross Zwisler | 91d25ba | 2017-09-06 16:18:43 -0700 | [diff] [blame] | 1695 | put_locked_mapping_entry(mapping, pgoff); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1696 | fallback: |
| 1697 | if (result == VM_FAULT_FALLBACK) { |
Dave Jiang | d8a849e | 2017-02-22 15:40:03 -0800 | [diff] [blame] | 1698 | split_huge_pmd(vma, vmf->pmd, vmf->address); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1699 | count_vm_event(THP_FAULT_FALLBACK); |
| 1700 | } |
Ross Zwisler | 282a8e0 | 2017-02-22 15:39:50 -0800 | [diff] [blame] | 1701 | out: |
Dave Jiang | f420039 | 2017-02-22 15:40:06 -0800 | [diff] [blame] | 1702 | trace_dax_pmd_fault_done(inode, vmf, max_pgoff, result); |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1703 | return result; |
| 1704 | } |
Dave Jiang | a2d5816 | 2017-02-24 14:56:59 -0800 | [diff] [blame] | 1705 | #else |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1706 | static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp, |
Arnd Bergmann | 01cddfe | 2017-02-27 14:26:44 -0800 | [diff] [blame] | 1707 | const struct iomap_ops *ops) |
Dave Jiang | a2d5816 | 2017-02-24 14:56:59 -0800 | [diff] [blame] | 1708 | { |
| 1709 | return VM_FAULT_FALLBACK; |
| 1710 | } |
Ross Zwisler | 642261a | 2016-11-08 11:34:45 +1100 | [diff] [blame] | 1711 | #endif /* CONFIG_FS_DAX_PMD */ |
Dave Jiang | a2d5816 | 2017-02-24 14:56:59 -0800 | [diff] [blame] | 1712 | |
| 1713 | /** |
| 1714 | * dax_iomap_fault - handle a page fault on a DAX file |
| 1715 | * @vmf: The description of the fault |
Jan Kara | cec04e8 | 2017-11-01 16:36:38 +0100 | [diff] [blame] | 1716 | * @pe_size: Size of the page to fault in |
Jan Kara | 9a0dd42 | 2017-11-01 16:36:39 +0100 | [diff] [blame] | 1717 | * @pfnp: PFN to insert for synchronous faults if fsync is required |
Jan Kara | c0b2462 | 2018-01-07 16:38:43 -0500 | [diff] [blame] | 1718 | * @iomap_errp: Storage for detailed error code in case of error |
Jan Kara | cec04e8 | 2017-11-01 16:36:38 +0100 | [diff] [blame] | 1719 | * @ops: Iomap ops passed from the file system |
Dave Jiang | a2d5816 | 2017-02-24 14:56:59 -0800 | [diff] [blame] | 1720 | * |
| 1721 | * When a page fault occurs, filesystems may call this helper in |
| 1722 | * their fault handler for DAX files. dax_iomap_fault() assumes the caller |
| 1723 | * has done all the necessary locking for page fault to proceed |
| 1724 | * successfully. |
| 1725 | */ |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1726 | vm_fault_t dax_iomap_fault(struct vm_fault *vmf, enum page_entry_size pe_size, |
Jan Kara | c0b2462 | 2018-01-07 16:38:43 -0500 | [diff] [blame] | 1727 | pfn_t *pfnp, int *iomap_errp, const struct iomap_ops *ops) |
Dave Jiang | a2d5816 | 2017-02-24 14:56:59 -0800 | [diff] [blame] | 1728 | { |
Dave Jiang | c791ace | 2017-02-24 14:57:08 -0800 | [diff] [blame] | 1729 | switch (pe_size) { |
| 1730 | case PE_SIZE_PTE: |
Jan Kara | c0b2462 | 2018-01-07 16:38:43 -0500 | [diff] [blame] | 1731 | return dax_iomap_pte_fault(vmf, pfnp, iomap_errp, ops); |
Dave Jiang | c791ace | 2017-02-24 14:57:08 -0800 | [diff] [blame] | 1732 | case PE_SIZE_PMD: |
Jan Kara | 9a0dd42 | 2017-11-01 16:36:39 +0100 | [diff] [blame] | 1733 | return dax_iomap_pmd_fault(vmf, pfnp, ops); |
Dave Jiang | a2d5816 | 2017-02-24 14:56:59 -0800 | [diff] [blame] | 1734 | default: |
| 1735 | return VM_FAULT_FALLBACK; |
| 1736 | } |
| 1737 | } |
| 1738 | EXPORT_SYMBOL_GPL(dax_iomap_fault); |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1739 | |
| 1740 | /** |
| 1741 | * dax_insert_pfn_mkwrite - insert PTE or PMD entry into page tables |
| 1742 | * @vmf: The description of the fault |
| 1743 | * @pe_size: Size of entry to be inserted |
| 1744 | * @pfn: PFN to insert |
| 1745 | * |
| 1746 | * This function inserts writeable PTE or PMD entry into page tables for mmaped |
| 1747 | * DAX file. It takes care of marking corresponding radix tree entry as dirty |
| 1748 | * as well. |
| 1749 | */ |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1750 | static vm_fault_t dax_insert_pfn_mkwrite(struct vm_fault *vmf, |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1751 | enum page_entry_size pe_size, |
| 1752 | pfn_t pfn) |
| 1753 | { |
| 1754 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
| 1755 | void *entry, **slot; |
| 1756 | pgoff_t index = vmf->pgoff; |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1757 | vm_fault_t ret; |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1758 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 1759 | xa_lock_irq(&mapping->i_pages); |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1760 | entry = get_unlocked_mapping_entry(mapping, index, &slot); |
| 1761 | /* Did we race with someone splitting entry or so? */ |
| 1762 | if (!entry || |
| 1763 | (pe_size == PE_SIZE_PTE && !dax_is_pte_entry(entry)) || |
| 1764 | (pe_size == PE_SIZE_PMD && !dax_is_pmd_entry(entry))) { |
| 1765 | put_unlocked_mapping_entry(mapping, index, entry); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 1766 | xa_unlock_irq(&mapping->i_pages); |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1767 | trace_dax_insert_pfn_mkwrite_no_entry(mapping->host, vmf, |
| 1768 | VM_FAULT_NOPAGE); |
| 1769 | return VM_FAULT_NOPAGE; |
| 1770 | } |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 1771 | radix_tree_tag_set(&mapping->i_pages, index, PAGECACHE_TAG_DIRTY); |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1772 | entry = lock_slot(mapping, slot); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 1773 | xa_unlock_irq(&mapping->i_pages); |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1774 | switch (pe_size) { |
| 1775 | case PE_SIZE_PTE: |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1776 | ret = vmf_insert_mixed_mkwrite(vmf->vma, vmf->address, pfn); |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1777 | break; |
| 1778 | #ifdef CONFIG_FS_DAX_PMD |
| 1779 | case PE_SIZE_PMD: |
Dan Williams | 58db381 | 2019-05-13 17:15:33 -0700 | [diff] [blame] | 1780 | ret = vmf_insert_pfn_pmd(vmf, pfn, FAULT_FLAG_WRITE); |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1781 | break; |
| 1782 | #endif |
| 1783 | default: |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1784 | ret = VM_FAULT_FALLBACK; |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1785 | } |
| 1786 | put_locked_mapping_entry(mapping, index); |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1787 | trace_dax_insert_pfn_mkwrite(mapping->host, vmf, ret); |
| 1788 | return ret; |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1789 | } |
| 1790 | |
| 1791 | /** |
| 1792 | * dax_finish_sync_fault - finish synchronous page fault |
| 1793 | * @vmf: The description of the fault |
| 1794 | * @pe_size: Size of entry to be inserted |
| 1795 | * @pfn: PFN to insert |
| 1796 | * |
| 1797 | * This function ensures that the file range touched by the page fault is |
| 1798 | * stored persistently on the media and handles inserting of appropriate page |
| 1799 | * table entry. |
| 1800 | */ |
Souptick Joarder | ab77dab | 2018-06-07 17:04:29 -0700 | [diff] [blame] | 1801 | vm_fault_t dax_finish_sync_fault(struct vm_fault *vmf, |
| 1802 | enum page_entry_size pe_size, pfn_t pfn) |
Jan Kara | 71eab6d | 2017-11-01 16:36:43 +0100 | [diff] [blame] | 1803 | { |
| 1804 | int err; |
| 1805 | loff_t start = ((loff_t)vmf->pgoff) << PAGE_SHIFT; |
| 1806 | size_t len = 0; |
| 1807 | |
| 1808 | if (pe_size == PE_SIZE_PTE) |
| 1809 | len = PAGE_SIZE; |
| 1810 | else if (pe_size == PE_SIZE_PMD) |
| 1811 | len = PMD_SIZE; |
| 1812 | else |
| 1813 | WARN_ON_ONCE(1); |
| 1814 | err = vfs_fsync_range(vmf->vma->vm_file, start, start + len - 1, 1); |
| 1815 | if (err) |
| 1816 | return VM_FAULT_SIGBUS; |
| 1817 | return dax_insert_pfn_mkwrite(vmf, pe_size, pfn); |
| 1818 | } |
| 1819 | EXPORT_SYMBOL_GPL(dax_finish_sync_fault); |