]>
Commit | Line | Data |
---|---|---|
d475c634 MW |
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> | |
d77e92e2 | 20 | #include <linux/dax.h> |
d475c634 MW |
21 | #include <linux/fs.h> |
22 | #include <linux/genhd.h> | |
f7ca90b1 MW |
23 | #include <linux/highmem.h> |
24 | #include <linux/memcontrol.h> | |
25 | #include <linux/mm.h> | |
d475c634 | 26 | #include <linux/mutex.h> |
9973c98e | 27 | #include <linux/pagevec.h> |
2765cfbb | 28 | #include <linux/pmem.h> |
289c6aed | 29 | #include <linux/sched.h> |
f361bf4a | 30 | #include <linux/sched/signal.h> |
d475c634 | 31 | #include <linux/uio.h> |
f7ca90b1 | 32 | #include <linux/vmstat.h> |
34c0fd54 | 33 | #include <linux/pfn_t.h> |
0e749e54 | 34 | #include <linux/sizes.h> |
4b4bb46d | 35 | #include <linux/mmu_notifier.h> |
a254e568 CH |
36 | #include <linux/iomap.h> |
37 | #include "internal.h" | |
d475c634 | 38 | |
282a8e03 RZ |
39 | #define CREATE_TRACE_POINTS |
40 | #include <trace/events/fs_dax.h> | |
41 | ||
ac401cc7 JK |
42 | /* We choose 4096 entries - same as per-zone page wait tables */ |
43 | #define DAX_WAIT_TABLE_BITS 12 | |
44 | #define DAX_WAIT_TABLE_ENTRIES (1 << DAX_WAIT_TABLE_BITS) | |
45 | ||
ce95ab0f | 46 | static wait_queue_head_t wait_table[DAX_WAIT_TABLE_ENTRIES]; |
ac401cc7 JK |
47 | |
48 | static int __init init_dax_wait_table(void) | |
49 | { | |
50 | int i; | |
51 | ||
52 | for (i = 0; i < DAX_WAIT_TABLE_ENTRIES; i++) | |
53 | init_waitqueue_head(wait_table + i); | |
54 | return 0; | |
55 | } | |
56 | fs_initcall(init_dax_wait_table); | |
57 | ||
b2e0d162 DW |
58 | static long dax_map_atomic(struct block_device *bdev, struct blk_dax_ctl *dax) |
59 | { | |
60 | struct request_queue *q = bdev->bd_queue; | |
61 | long rc = -EIO; | |
62 | ||
7a9eb206 | 63 | dax->addr = ERR_PTR(-EIO); |
b2e0d162 DW |
64 | if (blk_queue_enter(q, true) != 0) |
65 | return rc; | |
66 | ||
67 | rc = bdev_direct_access(bdev, dax); | |
68 | if (rc < 0) { | |
7a9eb206 | 69 | dax->addr = ERR_PTR(rc); |
b2e0d162 DW |
70 | blk_queue_exit(q); |
71 | return rc; | |
72 | } | |
73 | return rc; | |
74 | } | |
75 | ||
76 | static void dax_unmap_atomic(struct block_device *bdev, | |
77 | const struct blk_dax_ctl *dax) | |
78 | { | |
79 | if (IS_ERR(dax->addr)) | |
80 | return; | |
81 | blk_queue_exit(bdev->bd_queue); | |
82 | } | |
83 | ||
642261ac | 84 | static int dax_is_pmd_entry(void *entry) |
d1a5f2b4 | 85 | { |
642261ac | 86 | return (unsigned long)entry & RADIX_DAX_PMD; |
d1a5f2b4 DW |
87 | } |
88 | ||
642261ac | 89 | static int dax_is_pte_entry(void *entry) |
d475c634 | 90 | { |
642261ac | 91 | return !((unsigned long)entry & RADIX_DAX_PMD); |
d475c634 MW |
92 | } |
93 | ||
642261ac | 94 | static int dax_is_zero_entry(void *entry) |
d475c634 | 95 | { |
642261ac | 96 | return (unsigned long)entry & RADIX_DAX_HZP; |
d475c634 MW |
97 | } |
98 | ||
642261ac | 99 | static int dax_is_empty_entry(void *entry) |
b2e0d162 | 100 | { |
642261ac | 101 | return (unsigned long)entry & RADIX_DAX_EMPTY; |
b2e0d162 DW |
102 | } |
103 | ||
ac401cc7 JK |
104 | /* |
105 | * DAX radix tree locking | |
106 | */ | |
107 | struct exceptional_entry_key { | |
108 | struct address_space *mapping; | |
63e95b5c | 109 | pgoff_t entry_start; |
ac401cc7 JK |
110 | }; |
111 | ||
112 | struct wait_exceptional_entry_queue { | |
113 | wait_queue_t wait; | |
114 | struct exceptional_entry_key key; | |
115 | }; | |
116 | ||
63e95b5c RZ |
117 | static wait_queue_head_t *dax_entry_waitqueue(struct address_space *mapping, |
118 | pgoff_t index, void *entry, struct exceptional_entry_key *key) | |
119 | { | |
120 | unsigned long hash; | |
121 | ||
122 | /* | |
123 | * If 'entry' is a PMD, align the 'index' that we use for the wait | |
124 | * queue to the start of that PMD. This ensures that all offsets in | |
125 | * the range covered by the PMD map to the same bit lock. | |
126 | */ | |
642261ac | 127 | if (dax_is_pmd_entry(entry)) |
63e95b5c RZ |
128 | index &= ~((1UL << (PMD_SHIFT - PAGE_SHIFT)) - 1); |
129 | ||
130 | key->mapping = mapping; | |
131 | key->entry_start = index; | |
132 | ||
133 | hash = hash_long((unsigned long)mapping ^ index, DAX_WAIT_TABLE_BITS); | |
134 | return wait_table + hash; | |
135 | } | |
136 | ||
ac401cc7 JK |
137 | static int wake_exceptional_entry_func(wait_queue_t *wait, unsigned int mode, |
138 | int sync, void *keyp) | |
139 | { | |
140 | struct exceptional_entry_key *key = keyp; | |
141 | struct wait_exceptional_entry_queue *ewait = | |
142 | container_of(wait, struct wait_exceptional_entry_queue, wait); | |
143 | ||
144 | if (key->mapping != ewait->key.mapping || | |
63e95b5c | 145 | key->entry_start != ewait->key.entry_start) |
ac401cc7 JK |
146 | return 0; |
147 | return autoremove_wake_function(wait, mode, sync, NULL); | |
148 | } | |
149 | ||
150 | /* | |
151 | * Check whether the given slot is locked. The function must be called with | |
152 | * mapping->tree_lock held | |
153 | */ | |
154 | static inline int slot_locked(struct address_space *mapping, void **slot) | |
155 | { | |
156 | unsigned long entry = (unsigned long) | |
157 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
158 | return entry & RADIX_DAX_ENTRY_LOCK; | |
159 | } | |
160 | ||
161 | /* | |
162 | * Mark the given slot is locked. The function must be called with | |
163 | * mapping->tree_lock held | |
164 | */ | |
165 | static inline void *lock_slot(struct address_space *mapping, void **slot) | |
166 | { | |
167 | unsigned long entry = (unsigned long) | |
168 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
169 | ||
170 | entry |= RADIX_DAX_ENTRY_LOCK; | |
6d75f366 | 171 | radix_tree_replace_slot(&mapping->page_tree, slot, (void *)entry); |
ac401cc7 JK |
172 | return (void *)entry; |
173 | } | |
174 | ||
175 | /* | |
176 | * Mark the given slot is unlocked. The function must be called with | |
177 | * mapping->tree_lock held | |
178 | */ | |
179 | static inline void *unlock_slot(struct address_space *mapping, void **slot) | |
180 | { | |
181 | unsigned long entry = (unsigned long) | |
182 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
183 | ||
184 | entry &= ~(unsigned long)RADIX_DAX_ENTRY_LOCK; | |
6d75f366 | 185 | radix_tree_replace_slot(&mapping->page_tree, slot, (void *)entry); |
ac401cc7 JK |
186 | return (void *)entry; |
187 | } | |
188 | ||
189 | /* | |
190 | * Lookup entry in radix tree, wait for it to become unlocked if it is | |
191 | * exceptional entry and return it. The caller must call | |
192 | * put_unlocked_mapping_entry() when he decided not to lock the entry or | |
193 | * put_locked_mapping_entry() when he locked the entry and now wants to | |
194 | * unlock it. | |
195 | * | |
196 | * The function must be called with mapping->tree_lock held. | |
197 | */ | |
198 | static void *get_unlocked_mapping_entry(struct address_space *mapping, | |
199 | pgoff_t index, void ***slotp) | |
200 | { | |
e3ad61c6 | 201 | void *entry, **slot; |
ac401cc7 | 202 | struct wait_exceptional_entry_queue ewait; |
63e95b5c | 203 | wait_queue_head_t *wq; |
ac401cc7 JK |
204 | |
205 | init_wait(&ewait.wait); | |
206 | ewait.wait.func = wake_exceptional_entry_func; | |
ac401cc7 JK |
207 | |
208 | for (;;) { | |
e3ad61c6 | 209 | entry = __radix_tree_lookup(&mapping->page_tree, index, NULL, |
ac401cc7 | 210 | &slot); |
e3ad61c6 | 211 | if (!entry || !radix_tree_exceptional_entry(entry) || |
ac401cc7 JK |
212 | !slot_locked(mapping, slot)) { |
213 | if (slotp) | |
214 | *slotp = slot; | |
e3ad61c6 | 215 | return entry; |
ac401cc7 | 216 | } |
63e95b5c RZ |
217 | |
218 | wq = dax_entry_waitqueue(mapping, index, entry, &ewait.key); | |
ac401cc7 JK |
219 | prepare_to_wait_exclusive(wq, &ewait.wait, |
220 | TASK_UNINTERRUPTIBLE); | |
221 | spin_unlock_irq(&mapping->tree_lock); | |
222 | schedule(); | |
223 | finish_wait(wq, &ewait.wait); | |
224 | spin_lock_irq(&mapping->tree_lock); | |
225 | } | |
226 | } | |
227 | ||
b1aa812b JK |
228 | static void dax_unlock_mapping_entry(struct address_space *mapping, |
229 | pgoff_t index) | |
230 | { | |
231 | void *entry, **slot; | |
232 | ||
233 | spin_lock_irq(&mapping->tree_lock); | |
234 | entry = __radix_tree_lookup(&mapping->page_tree, index, NULL, &slot); | |
235 | if (WARN_ON_ONCE(!entry || !radix_tree_exceptional_entry(entry) || | |
236 | !slot_locked(mapping, slot))) { | |
237 | spin_unlock_irq(&mapping->tree_lock); | |
238 | return; | |
239 | } | |
240 | unlock_slot(mapping, slot); | |
241 | spin_unlock_irq(&mapping->tree_lock); | |
242 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); | |
243 | } | |
244 | ||
422476c4 RZ |
245 | static void put_locked_mapping_entry(struct address_space *mapping, |
246 | pgoff_t index, void *entry) | |
247 | { | |
248 | if (!radix_tree_exceptional_entry(entry)) { | |
249 | unlock_page(entry); | |
250 | put_page(entry); | |
251 | } else { | |
252 | dax_unlock_mapping_entry(mapping, index); | |
253 | } | |
254 | } | |
255 | ||
256 | /* | |
257 | * Called when we are done with radix tree entry we looked up via | |
258 | * get_unlocked_mapping_entry() and which we didn't lock in the end. | |
259 | */ | |
260 | static void put_unlocked_mapping_entry(struct address_space *mapping, | |
261 | pgoff_t index, void *entry) | |
262 | { | |
263 | if (!radix_tree_exceptional_entry(entry)) | |
264 | return; | |
265 | ||
266 | /* We have to wake up next waiter for the radix tree entry lock */ | |
267 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); | |
268 | } | |
269 | ||
ac401cc7 JK |
270 | /* |
271 | * Find radix tree entry at given index. If it points to a page, return with | |
272 | * the page locked. If it points to the exceptional entry, return with the | |
273 | * radix tree entry locked. If the radix tree doesn't contain given index, | |
274 | * create empty exceptional entry for the index and return with it locked. | |
275 | * | |
642261ac RZ |
276 | * When requesting an entry with size RADIX_DAX_PMD, grab_mapping_entry() will |
277 | * either return that locked entry or will return an error. This error will | |
278 | * happen if there are any 4k entries (either zero pages or DAX entries) | |
279 | * within the 2MiB range that we are requesting. | |
280 | * | |
281 | * We always favor 4k entries over 2MiB entries. There isn't a flow where we | |
282 | * evict 4k entries in order to 'upgrade' them to a 2MiB entry. A 2MiB | |
283 | * insertion will fail if it finds any 4k entries already in the tree, and a | |
284 | * 4k insertion will cause an existing 2MiB entry to be unmapped and | |
285 | * downgraded to 4k entries. This happens for both 2MiB huge zero pages as | |
286 | * well as 2MiB empty entries. | |
287 | * | |
288 | * The exception to this downgrade path is for 2MiB DAX PMD entries that have | |
289 | * real storage backing them. We will leave these real 2MiB DAX entries in | |
290 | * the tree, and PTE writes will simply dirty the entire 2MiB DAX entry. | |
291 | * | |
ac401cc7 JK |
292 | * Note: Unlike filemap_fault() we don't honor FAULT_FLAG_RETRY flags. For |
293 | * persistent memory the benefit is doubtful. We can add that later if we can | |
294 | * show it helps. | |
295 | */ | |
642261ac RZ |
296 | static void *grab_mapping_entry(struct address_space *mapping, pgoff_t index, |
297 | unsigned long size_flag) | |
ac401cc7 | 298 | { |
642261ac | 299 | bool pmd_downgrade = false; /* splitting 2MiB entry into 4k entries? */ |
e3ad61c6 | 300 | void *entry, **slot; |
ac401cc7 JK |
301 | |
302 | restart: | |
303 | spin_lock_irq(&mapping->tree_lock); | |
e3ad61c6 | 304 | entry = get_unlocked_mapping_entry(mapping, index, &slot); |
642261ac RZ |
305 | |
306 | if (entry) { | |
307 | if (size_flag & RADIX_DAX_PMD) { | |
308 | if (!radix_tree_exceptional_entry(entry) || | |
309 | dax_is_pte_entry(entry)) { | |
310 | put_unlocked_mapping_entry(mapping, index, | |
311 | entry); | |
312 | entry = ERR_PTR(-EEXIST); | |
313 | goto out_unlock; | |
314 | } | |
315 | } else { /* trying to grab a PTE entry */ | |
316 | if (radix_tree_exceptional_entry(entry) && | |
317 | dax_is_pmd_entry(entry) && | |
318 | (dax_is_zero_entry(entry) || | |
319 | dax_is_empty_entry(entry))) { | |
320 | pmd_downgrade = true; | |
321 | } | |
322 | } | |
323 | } | |
324 | ||
ac401cc7 | 325 | /* No entry for given index? Make sure radix tree is big enough. */ |
642261ac | 326 | if (!entry || pmd_downgrade) { |
ac401cc7 JK |
327 | int err; |
328 | ||
642261ac RZ |
329 | if (pmd_downgrade) { |
330 | /* | |
331 | * Make sure 'entry' remains valid while we drop | |
332 | * mapping->tree_lock. | |
333 | */ | |
334 | entry = lock_slot(mapping, slot); | |
335 | } | |
336 | ||
ac401cc7 | 337 | spin_unlock_irq(&mapping->tree_lock); |
642261ac RZ |
338 | /* |
339 | * Besides huge zero pages the only other thing that gets | |
340 | * downgraded are empty entries which don't need to be | |
341 | * unmapped. | |
342 | */ | |
343 | if (pmd_downgrade && dax_is_zero_entry(entry)) | |
344 | unmap_mapping_range(mapping, | |
345 | (index << PAGE_SHIFT) & PMD_MASK, PMD_SIZE, 0); | |
346 | ||
ac401cc7 JK |
347 | err = radix_tree_preload( |
348 | mapping_gfp_mask(mapping) & ~__GFP_HIGHMEM); | |
0cb80b48 JK |
349 | if (err) { |
350 | if (pmd_downgrade) | |
351 | put_locked_mapping_entry(mapping, index, entry); | |
ac401cc7 | 352 | return ERR_PTR(err); |
0cb80b48 | 353 | } |
ac401cc7 | 354 | spin_lock_irq(&mapping->tree_lock); |
642261ac RZ |
355 | |
356 | if (pmd_downgrade) { | |
357 | radix_tree_delete(&mapping->page_tree, index); | |
358 | mapping->nrexceptional--; | |
359 | dax_wake_mapping_entry_waiter(mapping, index, entry, | |
360 | true); | |
361 | } | |
362 | ||
363 | entry = dax_radix_locked_entry(0, size_flag | RADIX_DAX_EMPTY); | |
364 | ||
365 | err = __radix_tree_insert(&mapping->page_tree, index, | |
366 | dax_radix_order(entry), entry); | |
ac401cc7 JK |
367 | radix_tree_preload_end(); |
368 | if (err) { | |
369 | spin_unlock_irq(&mapping->tree_lock); | |
642261ac RZ |
370 | /* |
371 | * Someone already created the entry? This is a | |
372 | * normal failure when inserting PMDs in a range | |
373 | * that already contains PTEs. In that case we want | |
374 | * to return -EEXIST immediately. | |
375 | */ | |
376 | if (err == -EEXIST && !(size_flag & RADIX_DAX_PMD)) | |
ac401cc7 | 377 | goto restart; |
642261ac RZ |
378 | /* |
379 | * Our insertion of a DAX PMD entry failed, most | |
380 | * likely because it collided with a PTE sized entry | |
381 | * at a different index in the PMD range. We haven't | |
382 | * inserted anything into the radix tree and have no | |
383 | * waiters to wake. | |
384 | */ | |
ac401cc7 JK |
385 | return ERR_PTR(err); |
386 | } | |
387 | /* Good, we have inserted empty locked entry into the tree. */ | |
388 | mapping->nrexceptional++; | |
389 | spin_unlock_irq(&mapping->tree_lock); | |
e3ad61c6 | 390 | return entry; |
ac401cc7 JK |
391 | } |
392 | /* Normal page in radix tree? */ | |
e3ad61c6 RZ |
393 | if (!radix_tree_exceptional_entry(entry)) { |
394 | struct page *page = entry; | |
ac401cc7 JK |
395 | |
396 | get_page(page); | |
397 | spin_unlock_irq(&mapping->tree_lock); | |
398 | lock_page(page); | |
399 | /* Page got truncated? Retry... */ | |
400 | if (unlikely(page->mapping != mapping)) { | |
401 | unlock_page(page); | |
402 | put_page(page); | |
403 | goto restart; | |
404 | } | |
405 | return page; | |
406 | } | |
e3ad61c6 | 407 | entry = lock_slot(mapping, slot); |
642261ac | 408 | out_unlock: |
ac401cc7 | 409 | spin_unlock_irq(&mapping->tree_lock); |
e3ad61c6 | 410 | return entry; |
ac401cc7 JK |
411 | } |
412 | ||
63e95b5c RZ |
413 | /* |
414 | * We do not necessarily hold the mapping->tree_lock when we call this | |
415 | * function so it is possible that 'entry' is no longer a valid item in the | |
642261ac RZ |
416 | * radix tree. This is okay because all we really need to do is to find the |
417 | * correct waitqueue where tasks might be waiting for that old 'entry' and | |
418 | * wake them. | |
63e95b5c | 419 | */ |
ac401cc7 | 420 | void dax_wake_mapping_entry_waiter(struct address_space *mapping, |
63e95b5c | 421 | pgoff_t index, void *entry, bool wake_all) |
ac401cc7 | 422 | { |
63e95b5c RZ |
423 | struct exceptional_entry_key key; |
424 | wait_queue_head_t *wq; | |
425 | ||
426 | wq = dax_entry_waitqueue(mapping, index, entry, &key); | |
ac401cc7 JK |
427 | |
428 | /* | |
429 | * Checking for locked entry and prepare_to_wait_exclusive() happens | |
430 | * under mapping->tree_lock, ditto for entry handling in our callers. | |
431 | * So at this point all tasks that could have seen our entry locked | |
432 | * must be in the waitqueue and the following check will see them. | |
433 | */ | |
63e95b5c | 434 | if (waitqueue_active(wq)) |
ac401cc7 | 435 | __wake_up(wq, TASK_NORMAL, wake_all ? 0 : 1, &key); |
ac401cc7 JK |
436 | } |
437 | ||
c6dcf52c JK |
438 | static int __dax_invalidate_mapping_entry(struct address_space *mapping, |
439 | pgoff_t index, bool trunc) | |
440 | { | |
441 | int ret = 0; | |
442 | void *entry; | |
443 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
444 | ||
445 | spin_lock_irq(&mapping->tree_lock); | |
446 | entry = get_unlocked_mapping_entry(mapping, index, NULL); | |
447 | if (!entry || !radix_tree_exceptional_entry(entry)) | |
448 | goto out; | |
449 | if (!trunc && | |
450 | (radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_DIRTY) || | |
451 | radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE))) | |
452 | goto out; | |
453 | radix_tree_delete(page_tree, index); | |
454 | mapping->nrexceptional--; | |
455 | ret = 1; | |
456 | out: | |
457 | put_unlocked_mapping_entry(mapping, index, entry); | |
458 | spin_unlock_irq(&mapping->tree_lock); | |
459 | return ret; | |
460 | } | |
ac401cc7 JK |
461 | /* |
462 | * Delete exceptional DAX entry at @index from @mapping. Wait for radix tree | |
463 | * entry to get unlocked before deleting it. | |
464 | */ | |
465 | int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index) | |
466 | { | |
c6dcf52c | 467 | int ret = __dax_invalidate_mapping_entry(mapping, index, true); |
ac401cc7 | 468 | |
ac401cc7 JK |
469 | /* |
470 | * This gets called from truncate / punch_hole path. As such, the caller | |
471 | * must hold locks protecting against concurrent modifications of the | |
472 | * radix tree (usually fs-private i_mmap_sem for writing). Since the | |
473 | * caller has seen exceptional entry for this index, we better find it | |
474 | * at that index as well... | |
475 | */ | |
c6dcf52c JK |
476 | WARN_ON_ONCE(!ret); |
477 | return ret; | |
478 | } | |
479 | ||
480 | /* | |
481 | * Invalidate exceptional DAX entry if easily possible. This handles DAX | |
482 | * entries for invalidate_inode_pages() so we evict the entry only if we can | |
483 | * do so without blocking. | |
484 | */ | |
485 | int dax_invalidate_mapping_entry(struct address_space *mapping, pgoff_t index) | |
486 | { | |
487 | int ret = 0; | |
488 | void *entry, **slot; | |
489 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
490 | ||
491 | spin_lock_irq(&mapping->tree_lock); | |
492 | entry = __radix_tree_lookup(page_tree, index, NULL, &slot); | |
493 | if (!entry || !radix_tree_exceptional_entry(entry) || | |
494 | slot_locked(mapping, slot)) | |
495 | goto out; | |
496 | if (radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_DIRTY) || | |
497 | radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE)) | |
498 | goto out; | |
499 | radix_tree_delete(page_tree, index); | |
ac401cc7 | 500 | mapping->nrexceptional--; |
c6dcf52c JK |
501 | ret = 1; |
502 | out: | |
ac401cc7 | 503 | spin_unlock_irq(&mapping->tree_lock); |
c6dcf52c JK |
504 | if (ret) |
505 | dax_wake_mapping_entry_waiter(mapping, index, entry, true); | |
506 | return ret; | |
507 | } | |
ac401cc7 | 508 | |
c6dcf52c JK |
509 | /* |
510 | * Invalidate exceptional DAX entry if it is clean. | |
511 | */ | |
512 | int dax_invalidate_mapping_entry_sync(struct address_space *mapping, | |
513 | pgoff_t index) | |
514 | { | |
515 | return __dax_invalidate_mapping_entry(mapping, index, false); | |
ac401cc7 JK |
516 | } |
517 | ||
f7ca90b1 MW |
518 | /* |
519 | * The user has performed a load from a hole in the file. Allocating | |
520 | * a new page in the file would cause excessive storage usage for | |
521 | * workloads with sparse files. We allocate a page cache page instead. | |
522 | * We'll kick it out of the page cache if it's ever written to, | |
523 | * otherwise it will simply fall out of the page cache under memory | |
524 | * pressure without ever having been dirtied. | |
525 | */ | |
f449b936 | 526 | static int dax_load_hole(struct address_space *mapping, void **entry, |
ac401cc7 | 527 | struct vm_fault *vmf) |
f7ca90b1 | 528 | { |
ac401cc7 | 529 | struct page *page; |
f449b936 | 530 | int ret; |
f7ca90b1 | 531 | |
ac401cc7 | 532 | /* Hole page already exists? Return it... */ |
f449b936 JK |
533 | if (!radix_tree_exceptional_entry(*entry)) { |
534 | page = *entry; | |
535 | goto out; | |
ac401cc7 | 536 | } |
f7ca90b1 | 537 | |
ac401cc7 JK |
538 | /* This will replace locked radix tree entry with a hole page */ |
539 | page = find_or_create_page(mapping, vmf->pgoff, | |
540 | vmf->gfp_mask | __GFP_ZERO); | |
b1aa812b | 541 | if (!page) |
ac401cc7 | 542 | return VM_FAULT_OOM; |
f449b936 | 543 | out: |
f7ca90b1 | 544 | vmf->page = page; |
f449b936 JK |
545 | ret = finish_fault(vmf); |
546 | vmf->page = NULL; | |
547 | *entry = page; | |
548 | if (!ret) { | |
549 | /* Grab reference for PTE that is now referencing the page */ | |
550 | get_page(page); | |
551 | return VM_FAULT_NOPAGE; | |
552 | } | |
553 | return ret; | |
f7ca90b1 MW |
554 | } |
555 | ||
b0d5e82f CH |
556 | static int copy_user_dax(struct block_device *bdev, sector_t sector, size_t size, |
557 | struct page *to, unsigned long vaddr) | |
f7ca90b1 | 558 | { |
b2e0d162 | 559 | struct blk_dax_ctl dax = { |
b0d5e82f CH |
560 | .sector = sector, |
561 | .size = size, | |
b2e0d162 | 562 | }; |
e2e05394 RZ |
563 | void *vto; |
564 | ||
b2e0d162 DW |
565 | if (dax_map_atomic(bdev, &dax) < 0) |
566 | return PTR_ERR(dax.addr); | |
f7ca90b1 | 567 | vto = kmap_atomic(to); |
b2e0d162 | 568 | copy_user_page(vto, (void __force *)dax.addr, vaddr, to); |
f7ca90b1 | 569 | kunmap_atomic(vto); |
b2e0d162 | 570 | dax_unmap_atomic(bdev, &dax); |
f7ca90b1 MW |
571 | return 0; |
572 | } | |
573 | ||
642261ac RZ |
574 | /* |
575 | * By this point grab_mapping_entry() has ensured that we have a locked entry | |
576 | * of the appropriate size so we don't have to worry about downgrading PMDs to | |
577 | * PTEs. If we happen to be trying to insert a PTE and there is a PMD | |
578 | * already in the tree, we will skip the insertion and just dirty the PMD as | |
579 | * appropriate. | |
580 | */ | |
ac401cc7 JK |
581 | static void *dax_insert_mapping_entry(struct address_space *mapping, |
582 | struct vm_fault *vmf, | |
642261ac RZ |
583 | void *entry, sector_t sector, |
584 | unsigned long flags) | |
9973c98e RZ |
585 | { |
586 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
ac401cc7 JK |
587 | int error = 0; |
588 | bool hole_fill = false; | |
589 | void *new_entry; | |
590 | pgoff_t index = vmf->pgoff; | |
9973c98e | 591 | |
ac401cc7 | 592 | if (vmf->flags & FAULT_FLAG_WRITE) |
d2b2a28e | 593 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
9973c98e | 594 | |
ac401cc7 JK |
595 | /* Replacing hole page with block mapping? */ |
596 | if (!radix_tree_exceptional_entry(entry)) { | |
597 | hole_fill = true; | |
598 | /* | |
599 | * Unmap the page now before we remove it from page cache below. | |
600 | * The page is locked so it cannot be faulted in again. | |
601 | */ | |
602 | unmap_mapping_range(mapping, vmf->pgoff << PAGE_SHIFT, | |
603 | PAGE_SIZE, 0); | |
604 | error = radix_tree_preload(vmf->gfp_mask & ~__GFP_HIGHMEM); | |
605 | if (error) | |
606 | return ERR_PTR(error); | |
642261ac RZ |
607 | } else if (dax_is_zero_entry(entry) && !(flags & RADIX_DAX_HZP)) { |
608 | /* replacing huge zero page with PMD block mapping */ | |
609 | unmap_mapping_range(mapping, | |
610 | (vmf->pgoff << PAGE_SHIFT) & PMD_MASK, PMD_SIZE, 0); | |
9973c98e RZ |
611 | } |
612 | ||
ac401cc7 | 613 | spin_lock_irq(&mapping->tree_lock); |
642261ac RZ |
614 | new_entry = dax_radix_locked_entry(sector, flags); |
615 | ||
ac401cc7 JK |
616 | if (hole_fill) { |
617 | __delete_from_page_cache(entry, NULL); | |
618 | /* Drop pagecache reference */ | |
619 | put_page(entry); | |
642261ac RZ |
620 | error = __radix_tree_insert(page_tree, index, |
621 | dax_radix_order(new_entry), new_entry); | |
ac401cc7 JK |
622 | if (error) { |
623 | new_entry = ERR_PTR(error); | |
9973c98e RZ |
624 | goto unlock; |
625 | } | |
ac401cc7 | 626 | mapping->nrexceptional++; |
642261ac RZ |
627 | } else if (dax_is_zero_entry(entry) || dax_is_empty_entry(entry)) { |
628 | /* | |
629 | * Only swap our new entry into the radix tree if the current | |
630 | * entry is a zero page or an empty entry. If a normal PTE or | |
631 | * PMD entry is already in the tree, we leave it alone. This | |
632 | * means that if we are trying to insert a PTE and the | |
633 | * existing entry is a PMD, we will just leave the PMD in the | |
634 | * tree and dirty it if necessary. | |
635 | */ | |
f7942430 | 636 | struct radix_tree_node *node; |
ac401cc7 JK |
637 | void **slot; |
638 | void *ret; | |
9973c98e | 639 | |
f7942430 | 640 | ret = __radix_tree_lookup(page_tree, index, &node, &slot); |
ac401cc7 | 641 | WARN_ON_ONCE(ret != entry); |
4d693d08 JW |
642 | __radix_tree_replace(page_tree, node, slot, |
643 | new_entry, NULL, NULL); | |
9973c98e | 644 | } |
ac401cc7 | 645 | if (vmf->flags & FAULT_FLAG_WRITE) |
9973c98e RZ |
646 | radix_tree_tag_set(page_tree, index, PAGECACHE_TAG_DIRTY); |
647 | unlock: | |
648 | spin_unlock_irq(&mapping->tree_lock); | |
ac401cc7 JK |
649 | if (hole_fill) { |
650 | radix_tree_preload_end(); | |
651 | /* | |
652 | * We don't need hole page anymore, it has been replaced with | |
653 | * locked radix tree entry now. | |
654 | */ | |
655 | if (mapping->a_ops->freepage) | |
656 | mapping->a_ops->freepage(entry); | |
657 | unlock_page(entry); | |
658 | put_page(entry); | |
659 | } | |
660 | return new_entry; | |
9973c98e RZ |
661 | } |
662 | ||
4b4bb46d JK |
663 | static inline unsigned long |
664 | pgoff_address(pgoff_t pgoff, struct vm_area_struct *vma) | |
665 | { | |
666 | unsigned long address; | |
667 | ||
668 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
669 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); | |
670 | return address; | |
671 | } | |
672 | ||
673 | /* Walk all mappings of a given index of a file and writeprotect them */ | |
674 | static void dax_mapping_entry_mkclean(struct address_space *mapping, | |
675 | pgoff_t index, unsigned long pfn) | |
676 | { | |
677 | struct vm_area_struct *vma; | |
f729c8c9 RZ |
678 | pte_t pte, *ptep = NULL; |
679 | pmd_t *pmdp = NULL; | |
4b4bb46d JK |
680 | spinlock_t *ptl; |
681 | bool changed; | |
682 | ||
683 | i_mmap_lock_read(mapping); | |
684 | vma_interval_tree_foreach(vma, &mapping->i_mmap, index, index) { | |
685 | unsigned long address; | |
686 | ||
687 | cond_resched(); | |
688 | ||
689 | if (!(vma->vm_flags & VM_SHARED)) | |
690 | continue; | |
691 | ||
692 | address = pgoff_address(index, vma); | |
693 | changed = false; | |
f729c8c9 | 694 | if (follow_pte_pmd(vma->vm_mm, address, &ptep, &pmdp, &ptl)) |
4b4bb46d | 695 | continue; |
4b4bb46d | 696 | |
f729c8c9 RZ |
697 | if (pmdp) { |
698 | #ifdef CONFIG_FS_DAX_PMD | |
699 | pmd_t pmd; | |
700 | ||
701 | if (pfn != pmd_pfn(*pmdp)) | |
702 | goto unlock_pmd; | |
703 | if (!pmd_dirty(*pmdp) && !pmd_write(*pmdp)) | |
704 | goto unlock_pmd; | |
705 | ||
706 | flush_cache_page(vma, address, pfn); | |
707 | pmd = pmdp_huge_clear_flush(vma, address, pmdp); | |
708 | pmd = pmd_wrprotect(pmd); | |
709 | pmd = pmd_mkclean(pmd); | |
710 | set_pmd_at(vma->vm_mm, address, pmdp, pmd); | |
711 | changed = true; | |
712 | unlock_pmd: | |
713 | spin_unlock(ptl); | |
714 | #endif | |
715 | } else { | |
716 | if (pfn != pte_pfn(*ptep)) | |
717 | goto unlock_pte; | |
718 | if (!pte_dirty(*ptep) && !pte_write(*ptep)) | |
719 | goto unlock_pte; | |
720 | ||
721 | flush_cache_page(vma, address, pfn); | |
722 | pte = ptep_clear_flush(vma, address, ptep); | |
723 | pte = pte_wrprotect(pte); | |
724 | pte = pte_mkclean(pte); | |
725 | set_pte_at(vma->vm_mm, address, ptep, pte); | |
726 | changed = true; | |
727 | unlock_pte: | |
728 | pte_unmap_unlock(ptep, ptl); | |
729 | } | |
4b4bb46d JK |
730 | |
731 | if (changed) | |
732 | mmu_notifier_invalidate_page(vma->vm_mm, address); | |
733 | } | |
734 | i_mmap_unlock_read(mapping); | |
735 | } | |
736 | ||
9973c98e RZ |
737 | static int dax_writeback_one(struct block_device *bdev, |
738 | struct address_space *mapping, pgoff_t index, void *entry) | |
739 | { | |
740 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
9973c98e | 741 | struct blk_dax_ctl dax; |
a6abc2c0 | 742 | void *entry2, **slot; |
9973c98e RZ |
743 | int ret = 0; |
744 | ||
9973c98e | 745 | /* |
a6abc2c0 JK |
746 | * A page got tagged dirty in DAX mapping? Something is seriously |
747 | * wrong. | |
9973c98e | 748 | */ |
a6abc2c0 JK |
749 | if (WARN_ON(!radix_tree_exceptional_entry(entry))) |
750 | return -EIO; | |
9973c98e | 751 | |
a6abc2c0 JK |
752 | spin_lock_irq(&mapping->tree_lock); |
753 | entry2 = get_unlocked_mapping_entry(mapping, index, &slot); | |
754 | /* Entry got punched out / reallocated? */ | |
755 | if (!entry2 || !radix_tree_exceptional_entry(entry2)) | |
756 | goto put_unlocked; | |
757 | /* | |
758 | * Entry got reallocated elsewhere? No need to writeback. We have to | |
759 | * compare sectors as we must not bail out due to difference in lockbit | |
760 | * or entry type. | |
761 | */ | |
762 | if (dax_radix_sector(entry2) != dax_radix_sector(entry)) | |
763 | goto put_unlocked; | |
642261ac RZ |
764 | if (WARN_ON_ONCE(dax_is_empty_entry(entry) || |
765 | dax_is_zero_entry(entry))) { | |
9973c98e | 766 | ret = -EIO; |
a6abc2c0 | 767 | goto put_unlocked; |
9973c98e RZ |
768 | } |
769 | ||
a6abc2c0 JK |
770 | /* Another fsync thread may have already written back this entry */ |
771 | if (!radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE)) | |
772 | goto put_unlocked; | |
773 | /* Lock the entry to serialize with page faults */ | |
774 | entry = lock_slot(mapping, slot); | |
775 | /* | |
776 | * We can clear the tag now but we have to be careful so that concurrent | |
777 | * dax_writeback_one() calls for the same index cannot finish before we | |
778 | * actually flush the caches. This is achieved as the calls will look | |
779 | * at the entry only under tree_lock and once they do that they will | |
780 | * see the entry locked and wait for it to unlock. | |
781 | */ | |
782 | radix_tree_tag_clear(page_tree, index, PAGECACHE_TAG_TOWRITE); | |
783 | spin_unlock_irq(&mapping->tree_lock); | |
784 | ||
642261ac RZ |
785 | /* |
786 | * Even if dax_writeback_mapping_range() was given a wbc->range_start | |
787 | * in the middle of a PMD, the 'index' we are given will be aligned to | |
788 | * the start index of the PMD, as will the sector we pull from | |
789 | * 'entry'. This allows us to flush for PMD_SIZE and not have to | |
790 | * worry about partial PMD writebacks. | |
791 | */ | |
792 | dax.sector = dax_radix_sector(entry); | |
793 | dax.size = PAGE_SIZE << dax_radix_order(entry); | |
9973c98e RZ |
794 | |
795 | /* | |
796 | * We cannot hold tree_lock while calling dax_map_atomic() because it | |
797 | * eventually calls cond_resched(). | |
798 | */ | |
799 | ret = dax_map_atomic(bdev, &dax); | |
a6abc2c0 JK |
800 | if (ret < 0) { |
801 | put_locked_mapping_entry(mapping, index, entry); | |
9973c98e | 802 | return ret; |
a6abc2c0 | 803 | } |
9973c98e RZ |
804 | |
805 | if (WARN_ON_ONCE(ret < dax.size)) { | |
806 | ret = -EIO; | |
807 | goto unmap; | |
808 | } | |
809 | ||
4b4bb46d | 810 | dax_mapping_entry_mkclean(mapping, index, pfn_t_to_pfn(dax.pfn)); |
9973c98e | 811 | wb_cache_pmem(dax.addr, dax.size); |
4b4bb46d JK |
812 | /* |
813 | * After we have flushed the cache, we can clear the dirty tag. There | |
814 | * cannot be new dirty data in the pfn after the flush has completed as | |
815 | * the pfn mappings are writeprotected and fault waits for mapping | |
816 | * entry lock. | |
817 | */ | |
818 | spin_lock_irq(&mapping->tree_lock); | |
819 | radix_tree_tag_clear(page_tree, index, PAGECACHE_TAG_DIRTY); | |
820 | spin_unlock_irq(&mapping->tree_lock); | |
9973c98e RZ |
821 | unmap: |
822 | dax_unmap_atomic(bdev, &dax); | |
a6abc2c0 | 823 | put_locked_mapping_entry(mapping, index, entry); |
9973c98e RZ |
824 | return ret; |
825 | ||
a6abc2c0 JK |
826 | put_unlocked: |
827 | put_unlocked_mapping_entry(mapping, index, entry2); | |
9973c98e RZ |
828 | spin_unlock_irq(&mapping->tree_lock); |
829 | return ret; | |
830 | } | |
831 | ||
832 | /* | |
833 | * Flush the mapping to the persistent domain within the byte range of [start, | |
834 | * end]. This is required by data integrity operations to ensure file data is | |
835 | * on persistent storage prior to completion of the operation. | |
836 | */ | |
7f6d5b52 RZ |
837 | int dax_writeback_mapping_range(struct address_space *mapping, |
838 | struct block_device *bdev, struct writeback_control *wbc) | |
9973c98e RZ |
839 | { |
840 | struct inode *inode = mapping->host; | |
642261ac | 841 | pgoff_t start_index, end_index; |
9973c98e RZ |
842 | pgoff_t indices[PAGEVEC_SIZE]; |
843 | struct pagevec pvec; | |
844 | bool done = false; | |
845 | int i, ret = 0; | |
9973c98e RZ |
846 | |
847 | if (WARN_ON_ONCE(inode->i_blkbits != PAGE_SHIFT)) | |
848 | return -EIO; | |
849 | ||
7f6d5b52 RZ |
850 | if (!mapping->nrexceptional || wbc->sync_mode != WB_SYNC_ALL) |
851 | return 0; | |
852 | ||
09cbfeaf KS |
853 | start_index = wbc->range_start >> PAGE_SHIFT; |
854 | end_index = wbc->range_end >> PAGE_SHIFT; | |
9973c98e RZ |
855 | |
856 | tag_pages_for_writeback(mapping, start_index, end_index); | |
857 | ||
858 | pagevec_init(&pvec, 0); | |
859 | while (!done) { | |
860 | pvec.nr = find_get_entries_tag(mapping, start_index, | |
861 | PAGECACHE_TAG_TOWRITE, PAGEVEC_SIZE, | |
862 | pvec.pages, indices); | |
863 | ||
864 | if (pvec.nr == 0) | |
865 | break; | |
866 | ||
867 | for (i = 0; i < pvec.nr; i++) { | |
868 | if (indices[i] > end_index) { | |
869 | done = true; | |
870 | break; | |
871 | } | |
872 | ||
873 | ret = dax_writeback_one(bdev, mapping, indices[i], | |
874 | pvec.pages[i]); | |
875 | if (ret < 0) | |
876 | return ret; | |
877 | } | |
878 | } | |
9973c98e RZ |
879 | return 0; |
880 | } | |
881 | EXPORT_SYMBOL_GPL(dax_writeback_mapping_range); | |
882 | ||
ac401cc7 | 883 | static int dax_insert_mapping(struct address_space *mapping, |
1aaba095 CH |
884 | struct block_device *bdev, sector_t sector, size_t size, |
885 | void **entryp, struct vm_area_struct *vma, struct vm_fault *vmf) | |
f7ca90b1 | 886 | { |
1a29d85e | 887 | unsigned long vaddr = vmf->address; |
b2e0d162 | 888 | struct blk_dax_ctl dax = { |
1aaba095 CH |
889 | .sector = sector, |
890 | .size = size, | |
b2e0d162 | 891 | }; |
ac401cc7 JK |
892 | void *ret; |
893 | void *entry = *entryp; | |
f7ca90b1 | 894 | |
4d9a2c87 JK |
895 | if (dax_map_atomic(bdev, &dax) < 0) |
896 | return PTR_ERR(dax.addr); | |
b2e0d162 | 897 | dax_unmap_atomic(bdev, &dax); |
f7ca90b1 | 898 | |
642261ac | 899 | ret = dax_insert_mapping_entry(mapping, vmf, entry, dax.sector, 0); |
4d9a2c87 JK |
900 | if (IS_ERR(ret)) |
901 | return PTR_ERR(ret); | |
ac401cc7 | 902 | *entryp = ret; |
9973c98e | 903 | |
4d9a2c87 | 904 | return vm_insert_mixed(vma, vaddr, dax.pfn); |
f7ca90b1 MW |
905 | } |
906 | ||
0e3b210c BH |
907 | /** |
908 | * dax_pfn_mkwrite - handle first write to DAX page | |
0e3b210c | 909 | * @vmf: The description of the fault |
0e3b210c | 910 | */ |
11bac800 | 911 | int dax_pfn_mkwrite(struct vm_fault *vmf) |
0e3b210c | 912 | { |
11bac800 | 913 | struct file *file = vmf->vma->vm_file; |
ac401cc7 | 914 | struct address_space *mapping = file->f_mapping; |
2f89dc12 | 915 | void *entry, **slot; |
ac401cc7 | 916 | pgoff_t index = vmf->pgoff; |
30f471fd | 917 | |
ac401cc7 | 918 | spin_lock_irq(&mapping->tree_lock); |
2f89dc12 JK |
919 | entry = get_unlocked_mapping_entry(mapping, index, &slot); |
920 | if (!entry || !radix_tree_exceptional_entry(entry)) { | |
921 | if (entry) | |
922 | put_unlocked_mapping_entry(mapping, index, entry); | |
923 | spin_unlock_irq(&mapping->tree_lock); | |
924 | return VM_FAULT_NOPAGE; | |
925 | } | |
ac401cc7 | 926 | radix_tree_tag_set(&mapping->page_tree, index, PAGECACHE_TAG_DIRTY); |
2f89dc12 | 927 | entry = lock_slot(mapping, slot); |
ac401cc7 | 928 | spin_unlock_irq(&mapping->tree_lock); |
2f89dc12 JK |
929 | /* |
930 | * If we race with somebody updating the PTE and finish_mkwrite_fault() | |
931 | * fails, we don't care. We need to return VM_FAULT_NOPAGE and retry | |
932 | * the fault in either case. | |
933 | */ | |
934 | finish_mkwrite_fault(vmf); | |
935 | put_locked_mapping_entry(mapping, index, entry); | |
0e3b210c BH |
936 | return VM_FAULT_NOPAGE; |
937 | } | |
938 | EXPORT_SYMBOL_GPL(dax_pfn_mkwrite); | |
939 | ||
4b0228fa VV |
940 | static bool dax_range_is_aligned(struct block_device *bdev, |
941 | unsigned int offset, unsigned int length) | |
942 | { | |
943 | unsigned short sector_size = bdev_logical_block_size(bdev); | |
944 | ||
945 | if (!IS_ALIGNED(offset, sector_size)) | |
946 | return false; | |
947 | if (!IS_ALIGNED(length, sector_size)) | |
948 | return false; | |
949 | ||
950 | return true; | |
951 | } | |
952 | ||
679c8bd3 CH |
953 | int __dax_zero_page_range(struct block_device *bdev, sector_t sector, |
954 | unsigned int offset, unsigned int length) | |
955 | { | |
956 | struct blk_dax_ctl dax = { | |
957 | .sector = sector, | |
958 | .size = PAGE_SIZE, | |
959 | }; | |
960 | ||
4b0228fa VV |
961 | if (dax_range_is_aligned(bdev, offset, length)) { |
962 | sector_t start_sector = dax.sector + (offset >> 9); | |
963 | ||
964 | return blkdev_issue_zeroout(bdev, start_sector, | |
965 | length >> 9, GFP_NOFS, true); | |
966 | } else { | |
967 | if (dax_map_atomic(bdev, &dax) < 0) | |
968 | return PTR_ERR(dax.addr); | |
969 | clear_pmem(dax.addr + offset, length); | |
4b0228fa VV |
970 | dax_unmap_atomic(bdev, &dax); |
971 | } | |
679c8bd3 CH |
972 | return 0; |
973 | } | |
974 | EXPORT_SYMBOL_GPL(__dax_zero_page_range); | |
975 | ||
333ccc97 | 976 | static sector_t dax_iomap_sector(struct iomap *iomap, loff_t pos) |
25726bc1 | 977 | { |
333ccc97 | 978 | return iomap->blkno + (((pos & PAGE_MASK) - iomap->offset) >> 9); |
25726bc1 | 979 | } |
a254e568 | 980 | |
a254e568 | 981 | static loff_t |
11c59c92 | 982 | dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data, |
a254e568 CH |
983 | struct iomap *iomap) |
984 | { | |
985 | struct iov_iter *iter = data; | |
986 | loff_t end = pos + length, done = 0; | |
987 | ssize_t ret = 0; | |
988 | ||
989 | if (iov_iter_rw(iter) == READ) { | |
990 | end = min(end, i_size_read(inode)); | |
991 | if (pos >= end) | |
992 | return 0; | |
993 | ||
994 | if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN) | |
995 | return iov_iter_zero(min(length, end - pos), iter); | |
996 | } | |
997 | ||
998 | if (WARN_ON_ONCE(iomap->type != IOMAP_MAPPED)) | |
999 | return -EIO; | |
1000 | ||
e3fce68c JK |
1001 | /* |
1002 | * Write can allocate block for an area which has a hole page mapped | |
1003 | * into page tables. We have to tear down these mappings so that data | |
1004 | * written by write(2) is visible in mmap. | |
1005 | */ | |
1006 | if ((iomap->flags & IOMAP_F_NEW) && inode->i_mapping->nrpages) { | |
1007 | invalidate_inode_pages2_range(inode->i_mapping, | |
1008 | pos >> PAGE_SHIFT, | |
1009 | (end - 1) >> PAGE_SHIFT); | |
1010 | } | |
1011 | ||
a254e568 CH |
1012 | while (pos < end) { |
1013 | unsigned offset = pos & (PAGE_SIZE - 1); | |
1014 | struct blk_dax_ctl dax = { 0 }; | |
1015 | ssize_t map_len; | |
1016 | ||
d1908f52 MH |
1017 | if (fatal_signal_pending(current)) { |
1018 | ret = -EINTR; | |
1019 | break; | |
1020 | } | |
1021 | ||
333ccc97 | 1022 | dax.sector = dax_iomap_sector(iomap, pos); |
a254e568 CH |
1023 | dax.size = (length + offset + PAGE_SIZE - 1) & PAGE_MASK; |
1024 | map_len = dax_map_atomic(iomap->bdev, &dax); | |
1025 | if (map_len < 0) { | |
1026 | ret = map_len; | |
1027 | break; | |
1028 | } | |
1029 | ||
1030 | dax.addr += offset; | |
1031 | map_len -= offset; | |
1032 | if (map_len > end - pos) | |
1033 | map_len = end - pos; | |
1034 | ||
1035 | if (iov_iter_rw(iter) == WRITE) | |
1036 | map_len = copy_from_iter_pmem(dax.addr, map_len, iter); | |
1037 | else | |
1038 | map_len = copy_to_iter(dax.addr, map_len, iter); | |
1039 | dax_unmap_atomic(iomap->bdev, &dax); | |
1040 | if (map_len <= 0) { | |
1041 | ret = map_len ? map_len : -EFAULT; | |
1042 | break; | |
1043 | } | |
1044 | ||
1045 | pos += map_len; | |
1046 | length -= map_len; | |
1047 | done += map_len; | |
1048 | } | |
1049 | ||
1050 | return done ? done : ret; | |
1051 | } | |
1052 | ||
1053 | /** | |
11c59c92 | 1054 | * dax_iomap_rw - Perform I/O to a DAX file |
a254e568 CH |
1055 | * @iocb: The control block for this I/O |
1056 | * @iter: The addresses to do I/O from or to | |
1057 | * @ops: iomap ops passed from the file system | |
1058 | * | |
1059 | * This function performs read and write operations to directly mapped | |
1060 | * persistent memory. The callers needs to take care of read/write exclusion | |
1061 | * and evicting any page cache pages in the region under I/O. | |
1062 | */ | |
1063 | ssize_t | |
11c59c92 | 1064 | dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter, |
8ff6daa1 | 1065 | const struct iomap_ops *ops) |
a254e568 CH |
1066 | { |
1067 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
1068 | struct inode *inode = mapping->host; | |
1069 | loff_t pos = iocb->ki_pos, ret = 0, done = 0; | |
1070 | unsigned flags = 0; | |
1071 | ||
168316db CH |
1072 | if (iov_iter_rw(iter) == WRITE) { |
1073 | lockdep_assert_held_exclusive(&inode->i_rwsem); | |
a254e568 | 1074 | flags |= IOMAP_WRITE; |
168316db CH |
1075 | } else { |
1076 | lockdep_assert_held(&inode->i_rwsem); | |
1077 | } | |
a254e568 | 1078 | |
a254e568 CH |
1079 | while (iov_iter_count(iter)) { |
1080 | ret = iomap_apply(inode, pos, iov_iter_count(iter), flags, ops, | |
11c59c92 | 1081 | iter, dax_iomap_actor); |
a254e568 CH |
1082 | if (ret <= 0) |
1083 | break; | |
1084 | pos += ret; | |
1085 | done += ret; | |
1086 | } | |
1087 | ||
1088 | iocb->ki_pos += done; | |
1089 | return done ? done : ret; | |
1090 | } | |
11c59c92 | 1091 | EXPORT_SYMBOL_GPL(dax_iomap_rw); |
a7d73fe6 | 1092 | |
9f141d6e JK |
1093 | static int dax_fault_return(int error) |
1094 | { | |
1095 | if (error == 0) | |
1096 | return VM_FAULT_NOPAGE; | |
1097 | if (error == -ENOMEM) | |
1098 | return VM_FAULT_OOM; | |
1099 | return VM_FAULT_SIGBUS; | |
1100 | } | |
1101 | ||
a2d58167 DJ |
1102 | static int dax_iomap_pte_fault(struct vm_fault *vmf, |
1103 | const struct iomap_ops *ops) | |
a7d73fe6 | 1104 | { |
11bac800 | 1105 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
a7d73fe6 | 1106 | struct inode *inode = mapping->host; |
1a29d85e | 1107 | unsigned long vaddr = vmf->address; |
a7d73fe6 CH |
1108 | loff_t pos = (loff_t)vmf->pgoff << PAGE_SHIFT; |
1109 | sector_t sector; | |
1110 | struct iomap iomap = { 0 }; | |
9484ab1b | 1111 | unsigned flags = IOMAP_FAULT; |
a7d73fe6 | 1112 | int error, major = 0; |
b1aa812b | 1113 | int vmf_ret = 0; |
a7d73fe6 CH |
1114 | void *entry; |
1115 | ||
1116 | /* | |
1117 | * Check whether offset isn't beyond end of file now. Caller is supposed | |
1118 | * to hold locks serializing us with truncate / punch hole so this is | |
1119 | * a reliable test. | |
1120 | */ | |
1121 | if (pos >= i_size_read(inode)) | |
1122 | return VM_FAULT_SIGBUS; | |
1123 | ||
a7d73fe6 CH |
1124 | if ((vmf->flags & FAULT_FLAG_WRITE) && !vmf->cow_page) |
1125 | flags |= IOMAP_WRITE; | |
1126 | ||
1127 | /* | |
1128 | * Note that we don't bother to use iomap_apply here: DAX required | |
1129 | * the file system block size to be equal the page size, which means | |
1130 | * that we never have to deal with more than a single extent here. | |
1131 | */ | |
1132 | error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap); | |
1133 | if (error) | |
9f141d6e | 1134 | return dax_fault_return(error); |
a7d73fe6 | 1135 | if (WARN_ON_ONCE(iomap.offset + iomap.length < pos + PAGE_SIZE)) { |
9f141d6e JK |
1136 | vmf_ret = dax_fault_return(-EIO); /* fs corruption? */ |
1137 | goto finish_iomap; | |
1138 | } | |
1139 | ||
1140 | entry = grab_mapping_entry(mapping, vmf->pgoff, 0); | |
1141 | if (IS_ERR(entry)) { | |
1142 | vmf_ret = dax_fault_return(PTR_ERR(entry)); | |
1550290b | 1143 | goto finish_iomap; |
a7d73fe6 CH |
1144 | } |
1145 | ||
333ccc97 | 1146 | sector = dax_iomap_sector(&iomap, pos); |
a7d73fe6 CH |
1147 | |
1148 | if (vmf->cow_page) { | |
1149 | switch (iomap.type) { | |
1150 | case IOMAP_HOLE: | |
1151 | case IOMAP_UNWRITTEN: | |
1152 | clear_user_highpage(vmf->cow_page, vaddr); | |
1153 | break; | |
1154 | case IOMAP_MAPPED: | |
1155 | error = copy_user_dax(iomap.bdev, sector, PAGE_SIZE, | |
1156 | vmf->cow_page, vaddr); | |
1157 | break; | |
1158 | default: | |
1159 | WARN_ON_ONCE(1); | |
1160 | error = -EIO; | |
1161 | break; | |
1162 | } | |
1163 | ||
1164 | if (error) | |
9f141d6e | 1165 | goto error_unlock_entry; |
b1aa812b JK |
1166 | |
1167 | __SetPageUptodate(vmf->cow_page); | |
1168 | vmf_ret = finish_fault(vmf); | |
1169 | if (!vmf_ret) | |
1170 | vmf_ret = VM_FAULT_DONE_COW; | |
9f141d6e | 1171 | goto unlock_entry; |
a7d73fe6 CH |
1172 | } |
1173 | ||
1174 | switch (iomap.type) { | |
1175 | case IOMAP_MAPPED: | |
1176 | if (iomap.flags & IOMAP_F_NEW) { | |
1177 | count_vm_event(PGMAJFAULT); | |
11bac800 | 1178 | mem_cgroup_count_vm_event(vmf->vma->vm_mm, PGMAJFAULT); |
a7d73fe6 CH |
1179 | major = VM_FAULT_MAJOR; |
1180 | } | |
1181 | error = dax_insert_mapping(mapping, iomap.bdev, sector, | |
11bac800 | 1182 | PAGE_SIZE, &entry, vmf->vma, vmf); |
9f141d6e JK |
1183 | /* -EBUSY is fine, somebody else faulted on the same PTE */ |
1184 | if (error == -EBUSY) | |
1185 | error = 0; | |
a7d73fe6 CH |
1186 | break; |
1187 | case IOMAP_UNWRITTEN: | |
1188 | case IOMAP_HOLE: | |
1550290b | 1189 | if (!(vmf->flags & FAULT_FLAG_WRITE)) { |
f449b936 | 1190 | vmf_ret = dax_load_hole(mapping, &entry, vmf); |
9f141d6e | 1191 | goto unlock_entry; |
1550290b | 1192 | } |
a7d73fe6 CH |
1193 | /*FALLTHRU*/ |
1194 | default: | |
1195 | WARN_ON_ONCE(1); | |
1196 | error = -EIO; | |
1197 | break; | |
1198 | } | |
1199 | ||
9f141d6e JK |
1200 | error_unlock_entry: |
1201 | vmf_ret = dax_fault_return(error) | major; | |
a7d73fe6 | 1202 | unlock_entry: |
f449b936 | 1203 | put_locked_mapping_entry(mapping, vmf->pgoff, entry); |
9f141d6e JK |
1204 | finish_iomap: |
1205 | if (ops->iomap_end) { | |
1206 | int copied = PAGE_SIZE; | |
1207 | ||
1208 | if (vmf_ret & VM_FAULT_ERROR) | |
1209 | copied = 0; | |
1210 | /* | |
1211 | * The fault is done by now and there's no way back (other | |
1212 | * thread may be already happily using PTE we have installed). | |
1213 | * Just ignore error from ->iomap_end since we cannot do much | |
1214 | * with it. | |
1215 | */ | |
1216 | ops->iomap_end(inode, pos, PAGE_SIZE, copied, flags, &iomap); | |
1550290b | 1217 | } |
9f141d6e | 1218 | return vmf_ret; |
a7d73fe6 | 1219 | } |
642261ac RZ |
1220 | |
1221 | #ifdef CONFIG_FS_DAX_PMD | |
1222 | /* | |
1223 | * The 'colour' (ie low bits) within a PMD of a page offset. This comes up | |
1224 | * more often than one might expect in the below functions. | |
1225 | */ | |
1226 | #define PG_PMD_COLOUR ((PMD_SIZE >> PAGE_SHIFT) - 1) | |
1227 | ||
f4200391 DJ |
1228 | static int dax_pmd_insert_mapping(struct vm_fault *vmf, struct iomap *iomap, |
1229 | loff_t pos, void **entryp) | |
642261ac | 1230 | { |
f4200391 | 1231 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
642261ac | 1232 | struct block_device *bdev = iomap->bdev; |
27a7ffac | 1233 | struct inode *inode = mapping->host; |
642261ac RZ |
1234 | struct blk_dax_ctl dax = { |
1235 | .sector = dax_iomap_sector(iomap, pos), | |
1236 | .size = PMD_SIZE, | |
1237 | }; | |
1238 | long length = dax_map_atomic(bdev, &dax); | |
27a7ffac | 1239 | void *ret = NULL; |
642261ac RZ |
1240 | |
1241 | if (length < 0) /* dax_map_atomic() failed */ | |
27a7ffac | 1242 | goto fallback; |
642261ac RZ |
1243 | if (length < PMD_SIZE) |
1244 | goto unmap_fallback; | |
1245 | if (pfn_t_to_pfn(dax.pfn) & PG_PMD_COLOUR) | |
1246 | goto unmap_fallback; | |
1247 | if (!pfn_t_devmap(dax.pfn)) | |
1248 | goto unmap_fallback; | |
1249 | ||
1250 | dax_unmap_atomic(bdev, &dax); | |
1251 | ||
1252 | ret = dax_insert_mapping_entry(mapping, vmf, *entryp, dax.sector, | |
1253 | RADIX_DAX_PMD); | |
1254 | if (IS_ERR(ret)) | |
27a7ffac | 1255 | goto fallback; |
642261ac RZ |
1256 | *entryp = ret; |
1257 | ||
f4200391 DJ |
1258 | trace_dax_pmd_insert_mapping(inode, vmf, length, dax.pfn, ret); |
1259 | return vmf_insert_pfn_pmd(vmf->vma, vmf->address, vmf->pmd, | |
1260 | dax.pfn, vmf->flags & FAULT_FLAG_WRITE); | |
642261ac RZ |
1261 | |
1262 | unmap_fallback: | |
1263 | dax_unmap_atomic(bdev, &dax); | |
27a7ffac | 1264 | fallback: |
f4200391 DJ |
1265 | trace_dax_pmd_insert_mapping_fallback(inode, vmf, length, |
1266 | dax.pfn, ret); | |
642261ac RZ |
1267 | return VM_FAULT_FALLBACK; |
1268 | } | |
1269 | ||
f4200391 DJ |
1270 | static int dax_pmd_load_hole(struct vm_fault *vmf, struct iomap *iomap, |
1271 | void **entryp) | |
642261ac | 1272 | { |
f4200391 DJ |
1273 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
1274 | unsigned long pmd_addr = vmf->address & PMD_MASK; | |
653b2ea3 | 1275 | struct inode *inode = mapping->host; |
642261ac | 1276 | struct page *zero_page; |
653b2ea3 | 1277 | void *ret = NULL; |
642261ac RZ |
1278 | spinlock_t *ptl; |
1279 | pmd_t pmd_entry; | |
642261ac | 1280 | |
f4200391 | 1281 | zero_page = mm_get_huge_zero_page(vmf->vma->vm_mm); |
642261ac RZ |
1282 | |
1283 | if (unlikely(!zero_page)) | |
653b2ea3 | 1284 | goto fallback; |
642261ac RZ |
1285 | |
1286 | ret = dax_insert_mapping_entry(mapping, vmf, *entryp, 0, | |
1287 | RADIX_DAX_PMD | RADIX_DAX_HZP); | |
1288 | if (IS_ERR(ret)) | |
653b2ea3 | 1289 | goto fallback; |
642261ac RZ |
1290 | *entryp = ret; |
1291 | ||
f4200391 DJ |
1292 | ptl = pmd_lock(vmf->vma->vm_mm, vmf->pmd); |
1293 | if (!pmd_none(*(vmf->pmd))) { | |
642261ac | 1294 | spin_unlock(ptl); |
653b2ea3 | 1295 | goto fallback; |
642261ac RZ |
1296 | } |
1297 | ||
f4200391 | 1298 | pmd_entry = mk_pmd(zero_page, vmf->vma->vm_page_prot); |
642261ac | 1299 | pmd_entry = pmd_mkhuge(pmd_entry); |
f4200391 | 1300 | set_pmd_at(vmf->vma->vm_mm, pmd_addr, vmf->pmd, pmd_entry); |
642261ac | 1301 | spin_unlock(ptl); |
f4200391 | 1302 | trace_dax_pmd_load_hole(inode, vmf, zero_page, ret); |
642261ac | 1303 | return VM_FAULT_NOPAGE; |
653b2ea3 RZ |
1304 | |
1305 | fallback: | |
f4200391 | 1306 | trace_dax_pmd_load_hole_fallback(inode, vmf, zero_page, ret); |
653b2ea3 | 1307 | return VM_FAULT_FALLBACK; |
642261ac RZ |
1308 | } |
1309 | ||
a2d58167 DJ |
1310 | static int dax_iomap_pmd_fault(struct vm_fault *vmf, |
1311 | const struct iomap_ops *ops) | |
642261ac | 1312 | { |
f4200391 | 1313 | struct vm_area_struct *vma = vmf->vma; |
642261ac | 1314 | struct address_space *mapping = vma->vm_file->f_mapping; |
d8a849e1 DJ |
1315 | unsigned long pmd_addr = vmf->address & PMD_MASK; |
1316 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
9484ab1b | 1317 | unsigned int iomap_flags = (write ? IOMAP_WRITE : 0) | IOMAP_FAULT; |
642261ac RZ |
1318 | struct inode *inode = mapping->host; |
1319 | int result = VM_FAULT_FALLBACK; | |
1320 | struct iomap iomap = { 0 }; | |
1321 | pgoff_t max_pgoff, pgoff; | |
642261ac RZ |
1322 | void *entry; |
1323 | loff_t pos; | |
1324 | int error; | |
1325 | ||
282a8e03 RZ |
1326 | /* |
1327 | * Check whether offset isn't beyond end of file now. Caller is | |
1328 | * supposed to hold locks serializing us with truncate / punch hole so | |
1329 | * this is a reliable test. | |
1330 | */ | |
1331 | pgoff = linear_page_index(vma, pmd_addr); | |
1332 | max_pgoff = (i_size_read(inode) - 1) >> PAGE_SHIFT; | |
1333 | ||
f4200391 | 1334 | trace_dax_pmd_fault(inode, vmf, max_pgoff, 0); |
282a8e03 | 1335 | |
642261ac RZ |
1336 | /* Fall back to PTEs if we're going to COW */ |
1337 | if (write && !(vma->vm_flags & VM_SHARED)) | |
1338 | goto fallback; | |
1339 | ||
1340 | /* If the PMD would extend outside the VMA */ | |
1341 | if (pmd_addr < vma->vm_start) | |
1342 | goto fallback; | |
1343 | if ((pmd_addr + PMD_SIZE) > vma->vm_end) | |
1344 | goto fallback; | |
1345 | ||
282a8e03 RZ |
1346 | if (pgoff > max_pgoff) { |
1347 | result = VM_FAULT_SIGBUS; | |
1348 | goto out; | |
1349 | } | |
642261ac RZ |
1350 | |
1351 | /* If the PMD would extend beyond the file size */ | |
1352 | if ((pgoff | PG_PMD_COLOUR) > max_pgoff) | |
1353 | goto fallback; | |
1354 | ||
642261ac RZ |
1355 | /* |
1356 | * Note that we don't use iomap_apply here. We aren't doing I/O, only | |
1357 | * setting up a mapping, so really we're using iomap_begin() as a way | |
1358 | * to look up our filesystem block. | |
1359 | */ | |
1360 | pos = (loff_t)pgoff << PAGE_SHIFT; | |
1361 | error = ops->iomap_begin(inode, pos, PMD_SIZE, iomap_flags, &iomap); | |
1362 | if (error) | |
9f141d6e JK |
1363 | goto fallback; |
1364 | ||
642261ac RZ |
1365 | if (iomap.offset + iomap.length < pos + PMD_SIZE) |
1366 | goto finish_iomap; | |
1367 | ||
9f141d6e JK |
1368 | /* |
1369 | * grab_mapping_entry() will make sure we get a 2M empty entry, a DAX | |
1370 | * PMD or a HZP entry. If it can't (because a 4k page is already in | |
1371 | * the tree, for instance), it will return -EEXIST and we just fall | |
1372 | * back to 4k entries. | |
1373 | */ | |
1374 | entry = grab_mapping_entry(mapping, pgoff, RADIX_DAX_PMD); | |
1375 | if (IS_ERR(entry)) | |
1376 | goto finish_iomap; | |
1377 | ||
642261ac RZ |
1378 | switch (iomap.type) { |
1379 | case IOMAP_MAPPED: | |
f4200391 | 1380 | result = dax_pmd_insert_mapping(vmf, &iomap, pos, &entry); |
642261ac RZ |
1381 | break; |
1382 | case IOMAP_UNWRITTEN: | |
1383 | case IOMAP_HOLE: | |
1384 | if (WARN_ON_ONCE(write)) | |
9f141d6e | 1385 | goto unlock_entry; |
f4200391 | 1386 | result = dax_pmd_load_hole(vmf, &iomap, &entry); |
642261ac RZ |
1387 | break; |
1388 | default: | |
1389 | WARN_ON_ONCE(1); | |
1390 | break; | |
1391 | } | |
1392 | ||
9f141d6e JK |
1393 | unlock_entry: |
1394 | put_locked_mapping_entry(mapping, pgoff, entry); | |
642261ac RZ |
1395 | finish_iomap: |
1396 | if (ops->iomap_end) { | |
9f141d6e JK |
1397 | int copied = PMD_SIZE; |
1398 | ||
1399 | if (result == VM_FAULT_FALLBACK) | |
1400 | copied = 0; | |
1401 | /* | |
1402 | * The fault is done by now and there's no way back (other | |
1403 | * thread may be already happily using PMD we have installed). | |
1404 | * Just ignore error from ->iomap_end since we cannot do much | |
1405 | * with it. | |
1406 | */ | |
1407 | ops->iomap_end(inode, pos, PMD_SIZE, copied, iomap_flags, | |
1408 | &iomap); | |
642261ac | 1409 | } |
642261ac RZ |
1410 | fallback: |
1411 | if (result == VM_FAULT_FALLBACK) { | |
d8a849e1 | 1412 | split_huge_pmd(vma, vmf->pmd, vmf->address); |
642261ac RZ |
1413 | count_vm_event(THP_FAULT_FALLBACK); |
1414 | } | |
282a8e03 | 1415 | out: |
f4200391 | 1416 | trace_dax_pmd_fault_done(inode, vmf, max_pgoff, result); |
642261ac RZ |
1417 | return result; |
1418 | } | |
a2d58167 | 1419 | #else |
01cddfe9 AB |
1420 | static int dax_iomap_pmd_fault(struct vm_fault *vmf, |
1421 | const struct iomap_ops *ops) | |
a2d58167 DJ |
1422 | { |
1423 | return VM_FAULT_FALLBACK; | |
1424 | } | |
642261ac | 1425 | #endif /* CONFIG_FS_DAX_PMD */ |
a2d58167 DJ |
1426 | |
1427 | /** | |
1428 | * dax_iomap_fault - handle a page fault on a DAX file | |
1429 | * @vmf: The description of the fault | |
1430 | * @ops: iomap ops passed from the file system | |
1431 | * | |
1432 | * When a page fault occurs, filesystems may call this helper in | |
1433 | * their fault handler for DAX files. dax_iomap_fault() assumes the caller | |
1434 | * has done all the necessary locking for page fault to proceed | |
1435 | * successfully. | |
1436 | */ | |
c791ace1 DJ |
1437 | int dax_iomap_fault(struct vm_fault *vmf, enum page_entry_size pe_size, |
1438 | const struct iomap_ops *ops) | |
a2d58167 | 1439 | { |
c791ace1 DJ |
1440 | switch (pe_size) { |
1441 | case PE_SIZE_PTE: | |
a2d58167 | 1442 | return dax_iomap_pte_fault(vmf, ops); |
c791ace1 | 1443 | case PE_SIZE_PMD: |
a2d58167 DJ |
1444 | return dax_iomap_pmd_fault(vmf, ops); |
1445 | default: | |
1446 | return VM_FAULT_FALLBACK; | |
1447 | } | |
1448 | } | |
1449 | EXPORT_SYMBOL_GPL(dax_iomap_fault); |