]>
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> |
289c6aed | 28 | #include <linux/sched.h> |
f361bf4a | 29 | #include <linux/sched/signal.h> |
d475c634 | 30 | #include <linux/uio.h> |
f7ca90b1 | 31 | #include <linux/vmstat.h> |
34c0fd54 | 32 | #include <linux/pfn_t.h> |
0e749e54 | 33 | #include <linux/sizes.h> |
4b4bb46d | 34 | #include <linux/mmu_notifier.h> |
a254e568 CH |
35 | #include <linux/iomap.h> |
36 | #include "internal.h" | |
d475c634 | 37 | |
282a8e03 RZ |
38 | #define CREATE_TRACE_POINTS |
39 | #include <trace/events/fs_dax.h> | |
40 | ||
ac401cc7 JK |
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 | ||
ce95ab0f | 45 | static wait_queue_head_t wait_table[DAX_WAIT_TABLE_ENTRIES]; |
ac401cc7 JK |
46 | |
47 | static int __init init_dax_wait_table(void) | |
48 | { | |
49 | int i; | |
50 | ||
51 | for (i = 0; i < DAX_WAIT_TABLE_ENTRIES; i++) | |
52 | init_waitqueue_head(wait_table + i); | |
53 | return 0; | |
54 | } | |
55 | fs_initcall(init_dax_wait_table); | |
56 | ||
642261ac | 57 | static int dax_is_pmd_entry(void *entry) |
d1a5f2b4 | 58 | { |
642261ac | 59 | return (unsigned long)entry & RADIX_DAX_PMD; |
d1a5f2b4 DW |
60 | } |
61 | ||
642261ac | 62 | static int dax_is_pte_entry(void *entry) |
d475c634 | 63 | { |
642261ac | 64 | return !((unsigned long)entry & RADIX_DAX_PMD); |
d475c634 MW |
65 | } |
66 | ||
642261ac | 67 | static int dax_is_zero_entry(void *entry) |
d475c634 | 68 | { |
54585e35 | 69 | return (unsigned long)entry & RADIX_DAX_ZERO_PAGE; |
d475c634 MW |
70 | } |
71 | ||
642261ac | 72 | static int dax_is_empty_entry(void *entry) |
b2e0d162 | 73 | { |
642261ac | 74 | return (unsigned long)entry & RADIX_DAX_EMPTY; |
b2e0d162 DW |
75 | } |
76 | ||
ac401cc7 JK |
77 | /* |
78 | * DAX radix tree locking | |
79 | */ | |
80 | struct exceptional_entry_key { | |
81 | struct address_space *mapping; | |
63e95b5c | 82 | pgoff_t entry_start; |
ac401cc7 JK |
83 | }; |
84 | ||
85 | struct wait_exceptional_entry_queue { | |
ac6424b9 | 86 | wait_queue_entry_t wait; |
ac401cc7 JK |
87 | struct exceptional_entry_key key; |
88 | }; | |
89 | ||
63e95b5c RZ |
90 | static wait_queue_head_t *dax_entry_waitqueue(struct address_space *mapping, |
91 | pgoff_t index, void *entry, struct exceptional_entry_key *key) | |
92 | { | |
93 | unsigned long hash; | |
94 | ||
95 | /* | |
96 | * If 'entry' is a PMD, align the 'index' that we use for the wait | |
97 | * queue to the start of that PMD. This ensures that all offsets in | |
98 | * the range covered by the PMD map to the same bit lock. | |
99 | */ | |
642261ac | 100 | if (dax_is_pmd_entry(entry)) |
63e95b5c RZ |
101 | index &= ~((1UL << (PMD_SHIFT - PAGE_SHIFT)) - 1); |
102 | ||
103 | key->mapping = mapping; | |
104 | key->entry_start = index; | |
105 | ||
106 | hash = hash_long((unsigned long)mapping ^ index, DAX_WAIT_TABLE_BITS); | |
107 | return wait_table + hash; | |
108 | } | |
109 | ||
ac6424b9 | 110 | static int wake_exceptional_entry_func(wait_queue_entry_t *wait, unsigned int mode, |
ac401cc7 JK |
111 | int sync, void *keyp) |
112 | { | |
113 | struct exceptional_entry_key *key = keyp; | |
114 | struct wait_exceptional_entry_queue *ewait = | |
115 | container_of(wait, struct wait_exceptional_entry_queue, wait); | |
116 | ||
117 | if (key->mapping != ewait->key.mapping || | |
63e95b5c | 118 | key->entry_start != ewait->key.entry_start) |
ac401cc7 JK |
119 | return 0; |
120 | return autoremove_wake_function(wait, mode, sync, NULL); | |
121 | } | |
122 | ||
7b6956df RZ |
123 | /* |
124 | * We do not necessarily hold the mapping->tree_lock when we call this | |
125 | * function so it is possible that 'entry' is no longer a valid item in the | |
126 | * radix tree. This is okay because all we really need to do is to find the | |
127 | * correct waitqueue where tasks might be waiting for that old 'entry' and | |
128 | * wake them. | |
129 | */ | |
130 | void dax_wake_mapping_entry_waiter(struct address_space *mapping, | |
131 | pgoff_t index, void *entry, bool wake_all) | |
132 | { | |
133 | struct exceptional_entry_key key; | |
134 | wait_queue_head_t *wq; | |
135 | ||
136 | wq = dax_entry_waitqueue(mapping, index, entry, &key); | |
137 | ||
138 | /* | |
139 | * Checking for locked entry and prepare_to_wait_exclusive() happens | |
140 | * under mapping->tree_lock, ditto for entry handling in our callers. | |
141 | * So at this point all tasks that could have seen our entry locked | |
142 | * must be in the waitqueue and the following check will see them. | |
143 | */ | |
144 | if (waitqueue_active(wq)) | |
145 | __wake_up(wq, TASK_NORMAL, wake_all ? 0 : 1, &key); | |
146 | } | |
147 | ||
ac401cc7 JK |
148 | /* |
149 | * Check whether the given slot is locked. The function must be called with | |
150 | * mapping->tree_lock held | |
151 | */ | |
152 | static inline int slot_locked(struct address_space *mapping, void **slot) | |
153 | { | |
154 | unsigned long entry = (unsigned long) | |
155 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
156 | return entry & RADIX_DAX_ENTRY_LOCK; | |
157 | } | |
158 | ||
159 | /* | |
160 | * Mark the given slot is locked. The function must be called with | |
161 | * mapping->tree_lock held | |
162 | */ | |
163 | static inline void *lock_slot(struct address_space *mapping, void **slot) | |
164 | { | |
165 | unsigned long entry = (unsigned long) | |
166 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
167 | ||
168 | entry |= RADIX_DAX_ENTRY_LOCK; | |
6d75f366 | 169 | radix_tree_replace_slot(&mapping->page_tree, slot, (void *)entry); |
ac401cc7 JK |
170 | return (void *)entry; |
171 | } | |
172 | ||
173 | /* | |
174 | * Mark the given slot is unlocked. The function must be called with | |
175 | * mapping->tree_lock held | |
176 | */ | |
177 | static inline void *unlock_slot(struct address_space *mapping, void **slot) | |
178 | { | |
179 | unsigned long entry = (unsigned long) | |
180 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
181 | ||
182 | entry &= ~(unsigned long)RADIX_DAX_ENTRY_LOCK; | |
6d75f366 | 183 | radix_tree_replace_slot(&mapping->page_tree, slot, (void *)entry); |
ac401cc7 JK |
184 | return (void *)entry; |
185 | } | |
186 | ||
187 | /* | |
188 | * Lookup entry in radix tree, wait for it to become unlocked if it is | |
189 | * exceptional entry and return it. The caller must call | |
190 | * put_unlocked_mapping_entry() when he decided not to lock the entry or | |
191 | * put_locked_mapping_entry() when he locked the entry and now wants to | |
192 | * unlock it. | |
193 | * | |
194 | * The function must be called with mapping->tree_lock held. | |
195 | */ | |
196 | static void *get_unlocked_mapping_entry(struct address_space *mapping, | |
197 | pgoff_t index, void ***slotp) | |
198 | { | |
e3ad61c6 | 199 | void *entry, **slot; |
ac401cc7 | 200 | struct wait_exceptional_entry_queue ewait; |
63e95b5c | 201 | wait_queue_head_t *wq; |
ac401cc7 JK |
202 | |
203 | init_wait(&ewait.wait); | |
204 | ewait.wait.func = wake_exceptional_entry_func; | |
ac401cc7 JK |
205 | |
206 | for (;;) { | |
e3ad61c6 | 207 | entry = __radix_tree_lookup(&mapping->page_tree, index, NULL, |
ac401cc7 | 208 | &slot); |
54585e35 RZ |
209 | if (!entry || |
210 | WARN_ON_ONCE(!radix_tree_exceptional_entry(entry)) || | |
ac401cc7 JK |
211 | !slot_locked(mapping, slot)) { |
212 | if (slotp) | |
213 | *slotp = slot; | |
e3ad61c6 | 214 | return entry; |
ac401cc7 | 215 | } |
63e95b5c RZ |
216 | |
217 | wq = dax_entry_waitqueue(mapping, index, entry, &ewait.key); | |
ac401cc7 JK |
218 | prepare_to_wait_exclusive(wq, &ewait.wait, |
219 | TASK_UNINTERRUPTIBLE); | |
220 | spin_unlock_irq(&mapping->tree_lock); | |
221 | schedule(); | |
222 | finish_wait(wq, &ewait.wait); | |
223 | spin_lock_irq(&mapping->tree_lock); | |
224 | } | |
225 | } | |
226 | ||
b1aa812b JK |
227 | static void dax_unlock_mapping_entry(struct address_space *mapping, |
228 | pgoff_t index) | |
229 | { | |
230 | void *entry, **slot; | |
231 | ||
232 | spin_lock_irq(&mapping->tree_lock); | |
233 | entry = __radix_tree_lookup(&mapping->page_tree, index, NULL, &slot); | |
234 | if (WARN_ON_ONCE(!entry || !radix_tree_exceptional_entry(entry) || | |
235 | !slot_locked(mapping, slot))) { | |
236 | spin_unlock_irq(&mapping->tree_lock); | |
237 | return; | |
238 | } | |
239 | unlock_slot(mapping, slot); | |
240 | spin_unlock_irq(&mapping->tree_lock); | |
241 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); | |
242 | } | |
243 | ||
422476c4 | 244 | static void put_locked_mapping_entry(struct address_space *mapping, |
54585e35 | 245 | pgoff_t index) |
422476c4 | 246 | { |
54585e35 | 247 | dax_unlock_mapping_entry(mapping, index); |
422476c4 RZ |
248 | } |
249 | ||
250 | /* | |
251 | * Called when we are done with radix tree entry we looked up via | |
252 | * get_unlocked_mapping_entry() and which we didn't lock in the end. | |
253 | */ | |
254 | static void put_unlocked_mapping_entry(struct address_space *mapping, | |
255 | pgoff_t index, void *entry) | |
256 | { | |
54585e35 | 257 | if (!entry) |
422476c4 RZ |
258 | return; |
259 | ||
260 | /* We have to wake up next waiter for the radix tree entry lock */ | |
261 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); | |
262 | } | |
263 | ||
ac401cc7 | 264 | /* |
54585e35 RZ |
265 | * Find radix tree entry at given index. If it points to an exceptional entry, |
266 | * return it with the radix tree entry locked. If the radix tree doesn't | |
267 | * contain given index, create an empty exceptional entry for the index and | |
268 | * return with it locked. | |
ac401cc7 | 269 | * |
642261ac RZ |
270 | * When requesting an entry with size RADIX_DAX_PMD, grab_mapping_entry() will |
271 | * either return that locked entry or will return an error. This error will | |
54585e35 RZ |
272 | * happen if there are any 4k entries within the 2MiB range that we are |
273 | * requesting. | |
642261ac RZ |
274 | * |
275 | * We always favor 4k entries over 2MiB entries. There isn't a flow where we | |
276 | * evict 4k entries in order to 'upgrade' them to a 2MiB entry. A 2MiB | |
277 | * insertion will fail if it finds any 4k entries already in the tree, and a | |
278 | * 4k insertion will cause an existing 2MiB entry to be unmapped and | |
279 | * downgraded to 4k entries. This happens for both 2MiB huge zero pages as | |
280 | * well as 2MiB empty entries. | |
281 | * | |
282 | * The exception to this downgrade path is for 2MiB DAX PMD entries that have | |
283 | * real storage backing them. We will leave these real 2MiB DAX entries in | |
284 | * the tree, and PTE writes will simply dirty the entire 2MiB DAX entry. | |
285 | * | |
ac401cc7 JK |
286 | * Note: Unlike filemap_fault() we don't honor FAULT_FLAG_RETRY flags. For |
287 | * persistent memory the benefit is doubtful. We can add that later if we can | |
288 | * show it helps. | |
289 | */ | |
642261ac RZ |
290 | static void *grab_mapping_entry(struct address_space *mapping, pgoff_t index, |
291 | unsigned long size_flag) | |
ac401cc7 | 292 | { |
642261ac | 293 | bool pmd_downgrade = false; /* splitting 2MiB entry into 4k entries? */ |
e3ad61c6 | 294 | void *entry, **slot; |
ac401cc7 JK |
295 | |
296 | restart: | |
297 | spin_lock_irq(&mapping->tree_lock); | |
e3ad61c6 | 298 | entry = get_unlocked_mapping_entry(mapping, index, &slot); |
642261ac | 299 | |
54585e35 RZ |
300 | if (WARN_ON_ONCE(entry && !radix_tree_exceptional_entry(entry))) { |
301 | entry = ERR_PTR(-EIO); | |
302 | goto out_unlock; | |
303 | } | |
304 | ||
642261ac RZ |
305 | if (entry) { |
306 | if (size_flag & RADIX_DAX_PMD) { | |
54585e35 | 307 | if (dax_is_pte_entry(entry)) { |
642261ac RZ |
308 | put_unlocked_mapping_entry(mapping, index, |
309 | entry); | |
310 | entry = ERR_PTR(-EEXIST); | |
311 | goto out_unlock; | |
312 | } | |
313 | } else { /* trying to grab a PTE entry */ | |
54585e35 | 314 | if (dax_is_pmd_entry(entry) && |
642261ac RZ |
315 | (dax_is_zero_entry(entry) || |
316 | dax_is_empty_entry(entry))) { | |
317 | pmd_downgrade = true; | |
318 | } | |
319 | } | |
320 | } | |
321 | ||
ac401cc7 | 322 | /* No entry for given index? Make sure radix tree is big enough. */ |
642261ac | 323 | if (!entry || pmd_downgrade) { |
ac401cc7 JK |
324 | int err; |
325 | ||
642261ac RZ |
326 | if (pmd_downgrade) { |
327 | /* | |
328 | * Make sure 'entry' remains valid while we drop | |
329 | * mapping->tree_lock. | |
330 | */ | |
331 | entry = lock_slot(mapping, slot); | |
332 | } | |
333 | ||
ac401cc7 | 334 | spin_unlock_irq(&mapping->tree_lock); |
642261ac RZ |
335 | /* |
336 | * Besides huge zero pages the only other thing that gets | |
337 | * downgraded are empty entries which don't need to be | |
338 | * unmapped. | |
339 | */ | |
340 | if (pmd_downgrade && dax_is_zero_entry(entry)) | |
341 | unmap_mapping_range(mapping, | |
342 | (index << PAGE_SHIFT) & PMD_MASK, PMD_SIZE, 0); | |
343 | ||
ac401cc7 JK |
344 | err = radix_tree_preload( |
345 | mapping_gfp_mask(mapping) & ~__GFP_HIGHMEM); | |
0cb80b48 JK |
346 | if (err) { |
347 | if (pmd_downgrade) | |
54585e35 | 348 | put_locked_mapping_entry(mapping, index); |
ac401cc7 | 349 | return ERR_PTR(err); |
0cb80b48 | 350 | } |
ac401cc7 | 351 | spin_lock_irq(&mapping->tree_lock); |
642261ac | 352 | |
e11f8b7b RZ |
353 | if (!entry) { |
354 | /* | |
355 | * We needed to drop the page_tree lock while calling | |
356 | * radix_tree_preload() and we didn't have an entry to | |
357 | * lock. See if another thread inserted an entry at | |
358 | * our index during this time. | |
359 | */ | |
360 | entry = __radix_tree_lookup(&mapping->page_tree, index, | |
361 | NULL, &slot); | |
362 | if (entry) { | |
363 | radix_tree_preload_end(); | |
364 | spin_unlock_irq(&mapping->tree_lock); | |
365 | goto restart; | |
366 | } | |
367 | } | |
368 | ||
642261ac RZ |
369 | if (pmd_downgrade) { |
370 | radix_tree_delete(&mapping->page_tree, index); | |
371 | mapping->nrexceptional--; | |
372 | dax_wake_mapping_entry_waiter(mapping, index, entry, | |
373 | true); | |
374 | } | |
375 | ||
376 | entry = dax_radix_locked_entry(0, size_flag | RADIX_DAX_EMPTY); | |
377 | ||
378 | err = __radix_tree_insert(&mapping->page_tree, index, | |
379 | dax_radix_order(entry), entry); | |
ac401cc7 JK |
380 | radix_tree_preload_end(); |
381 | if (err) { | |
382 | spin_unlock_irq(&mapping->tree_lock); | |
642261ac | 383 | /* |
e11f8b7b RZ |
384 | * Our insertion of a DAX entry failed, most likely |
385 | * because we were inserting a PMD entry and it | |
386 | * collided with a PTE sized entry at a different | |
387 | * index in the PMD range. We haven't inserted | |
388 | * anything into the radix tree and have no waiters to | |
389 | * wake. | |
642261ac | 390 | */ |
ac401cc7 JK |
391 | return ERR_PTR(err); |
392 | } | |
393 | /* Good, we have inserted empty locked entry into the tree. */ | |
394 | mapping->nrexceptional++; | |
395 | spin_unlock_irq(&mapping->tree_lock); | |
e3ad61c6 | 396 | return entry; |
ac401cc7 | 397 | } |
e3ad61c6 | 398 | entry = lock_slot(mapping, slot); |
642261ac | 399 | out_unlock: |
ac401cc7 | 400 | spin_unlock_irq(&mapping->tree_lock); |
e3ad61c6 | 401 | return entry; |
ac401cc7 JK |
402 | } |
403 | ||
c6dcf52c JK |
404 | static int __dax_invalidate_mapping_entry(struct address_space *mapping, |
405 | pgoff_t index, bool trunc) | |
406 | { | |
407 | int ret = 0; | |
408 | void *entry; | |
409 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
410 | ||
411 | spin_lock_irq(&mapping->tree_lock); | |
412 | entry = get_unlocked_mapping_entry(mapping, index, NULL); | |
54585e35 | 413 | if (!entry || WARN_ON_ONCE(!radix_tree_exceptional_entry(entry))) |
c6dcf52c JK |
414 | goto out; |
415 | if (!trunc && | |
416 | (radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_DIRTY) || | |
417 | radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE))) | |
418 | goto out; | |
419 | radix_tree_delete(page_tree, index); | |
420 | mapping->nrexceptional--; | |
421 | ret = 1; | |
422 | out: | |
423 | put_unlocked_mapping_entry(mapping, index, entry); | |
424 | spin_unlock_irq(&mapping->tree_lock); | |
425 | return ret; | |
426 | } | |
ac401cc7 JK |
427 | /* |
428 | * Delete exceptional DAX entry at @index from @mapping. Wait for radix tree | |
429 | * entry to get unlocked before deleting it. | |
430 | */ | |
431 | int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index) | |
432 | { | |
c6dcf52c | 433 | int ret = __dax_invalidate_mapping_entry(mapping, index, true); |
ac401cc7 | 434 | |
ac401cc7 JK |
435 | /* |
436 | * This gets called from truncate / punch_hole path. As such, the caller | |
437 | * must hold locks protecting against concurrent modifications of the | |
438 | * radix tree (usually fs-private i_mmap_sem for writing). Since the | |
439 | * caller has seen exceptional entry for this index, we better find it | |
440 | * at that index as well... | |
441 | */ | |
c6dcf52c JK |
442 | WARN_ON_ONCE(!ret); |
443 | return ret; | |
444 | } | |
445 | ||
c6dcf52c JK |
446 | /* |
447 | * Invalidate exceptional DAX entry if it is clean. | |
448 | */ | |
449 | int dax_invalidate_mapping_entry_sync(struct address_space *mapping, | |
450 | pgoff_t index) | |
451 | { | |
452 | return __dax_invalidate_mapping_entry(mapping, index, false); | |
ac401cc7 JK |
453 | } |
454 | ||
cccbce67 DW |
455 | static int copy_user_dax(struct block_device *bdev, struct dax_device *dax_dev, |
456 | sector_t sector, size_t size, struct page *to, | |
457 | unsigned long vaddr) | |
f7ca90b1 | 458 | { |
cccbce67 DW |
459 | void *vto, *kaddr; |
460 | pgoff_t pgoff; | |
461 | pfn_t pfn; | |
462 | long rc; | |
463 | int id; | |
464 | ||
465 | rc = bdev_dax_pgoff(bdev, sector, size, &pgoff); | |
466 | if (rc) | |
467 | return rc; | |
468 | ||
469 | id = dax_read_lock(); | |
470 | rc = dax_direct_access(dax_dev, pgoff, PHYS_PFN(size), &kaddr, &pfn); | |
471 | if (rc < 0) { | |
472 | dax_read_unlock(id); | |
473 | return rc; | |
474 | } | |
f7ca90b1 | 475 | vto = kmap_atomic(to); |
cccbce67 | 476 | copy_user_page(vto, (void __force *)kaddr, vaddr, to); |
f7ca90b1 | 477 | kunmap_atomic(vto); |
cccbce67 | 478 | dax_read_unlock(id); |
f7ca90b1 MW |
479 | return 0; |
480 | } | |
481 | ||
642261ac RZ |
482 | /* |
483 | * By this point grab_mapping_entry() has ensured that we have a locked entry | |
484 | * of the appropriate size so we don't have to worry about downgrading PMDs to | |
485 | * PTEs. If we happen to be trying to insert a PTE and there is a PMD | |
486 | * already in the tree, we will skip the insertion and just dirty the PMD as | |
487 | * appropriate. | |
488 | */ | |
ac401cc7 JK |
489 | static void *dax_insert_mapping_entry(struct address_space *mapping, |
490 | struct vm_fault *vmf, | |
642261ac RZ |
491 | void *entry, sector_t sector, |
492 | unsigned long flags) | |
9973c98e RZ |
493 | { |
494 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
ac401cc7 JK |
495 | void *new_entry; |
496 | pgoff_t index = vmf->pgoff; | |
9973c98e | 497 | |
ac401cc7 | 498 | if (vmf->flags & FAULT_FLAG_WRITE) |
d2b2a28e | 499 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
9973c98e | 500 | |
54585e35 RZ |
501 | if (dax_is_zero_entry(entry) && !(flags & RADIX_DAX_ZERO_PAGE)) { |
502 | /* we are replacing a zero page with block mapping */ | |
503 | if (dax_is_pmd_entry(entry)) | |
504 | unmap_mapping_range(mapping, | |
505 | (vmf->pgoff << PAGE_SHIFT) & PMD_MASK, | |
506 | PMD_SIZE, 0); | |
507 | else /* pte entry */ | |
508 | unmap_mapping_range(mapping, vmf->pgoff << PAGE_SHIFT, | |
509 | PAGE_SIZE, 0); | |
9973c98e RZ |
510 | } |
511 | ||
ac401cc7 | 512 | spin_lock_irq(&mapping->tree_lock); |
642261ac RZ |
513 | new_entry = dax_radix_locked_entry(sector, flags); |
514 | ||
54585e35 | 515 | if (dax_is_zero_entry(entry) || dax_is_empty_entry(entry)) { |
642261ac RZ |
516 | /* |
517 | * Only swap our new entry into the radix tree if the current | |
518 | * entry is a zero page or an empty entry. If a normal PTE or | |
519 | * PMD entry is already in the tree, we leave it alone. This | |
520 | * means that if we are trying to insert a PTE and the | |
521 | * existing entry is a PMD, we will just leave the PMD in the | |
522 | * tree and dirty it if necessary. | |
523 | */ | |
f7942430 | 524 | struct radix_tree_node *node; |
ac401cc7 JK |
525 | void **slot; |
526 | void *ret; | |
9973c98e | 527 | |
f7942430 | 528 | ret = __radix_tree_lookup(page_tree, index, &node, &slot); |
ac401cc7 | 529 | WARN_ON_ONCE(ret != entry); |
4d693d08 JW |
530 | __radix_tree_replace(page_tree, node, slot, |
531 | new_entry, NULL, NULL); | |
54585e35 | 532 | entry = new_entry; |
9973c98e | 533 | } |
54585e35 | 534 | |
ac401cc7 | 535 | if (vmf->flags & FAULT_FLAG_WRITE) |
9973c98e | 536 | radix_tree_tag_set(page_tree, index, PAGECACHE_TAG_DIRTY); |
54585e35 | 537 | |
9973c98e | 538 | spin_unlock_irq(&mapping->tree_lock); |
54585e35 | 539 | return entry; |
9973c98e RZ |
540 | } |
541 | ||
4b4bb46d JK |
542 | static inline unsigned long |
543 | pgoff_address(pgoff_t pgoff, struct vm_area_struct *vma) | |
544 | { | |
545 | unsigned long address; | |
546 | ||
547 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
548 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); | |
549 | return address; | |
550 | } | |
551 | ||
552 | /* Walk all mappings of a given index of a file and writeprotect them */ | |
553 | static void dax_mapping_entry_mkclean(struct address_space *mapping, | |
554 | pgoff_t index, unsigned long pfn) | |
555 | { | |
556 | struct vm_area_struct *vma; | |
f729c8c9 RZ |
557 | pte_t pte, *ptep = NULL; |
558 | pmd_t *pmdp = NULL; | |
4b4bb46d | 559 | spinlock_t *ptl; |
4b4bb46d JK |
560 | |
561 | i_mmap_lock_read(mapping); | |
562 | vma_interval_tree_foreach(vma, &mapping->i_mmap, index, index) { | |
a4d1a885 | 563 | unsigned long address, start, end; |
4b4bb46d JK |
564 | |
565 | cond_resched(); | |
566 | ||
567 | if (!(vma->vm_flags & VM_SHARED)) | |
568 | continue; | |
569 | ||
570 | address = pgoff_address(index, vma); | |
a4d1a885 JG |
571 | |
572 | /* | |
573 | * Note because we provide start/end to follow_pte_pmd it will | |
574 | * call mmu_notifier_invalidate_range_start() on our behalf | |
575 | * before taking any lock. | |
576 | */ | |
577 | if (follow_pte_pmd(vma->vm_mm, address, &start, &end, &ptep, &pmdp, &ptl)) | |
4b4bb46d | 578 | continue; |
4b4bb46d | 579 | |
f729c8c9 RZ |
580 | if (pmdp) { |
581 | #ifdef CONFIG_FS_DAX_PMD | |
582 | pmd_t pmd; | |
583 | ||
584 | if (pfn != pmd_pfn(*pmdp)) | |
585 | goto unlock_pmd; | |
586 | if (!pmd_dirty(*pmdp) && !pmd_write(*pmdp)) | |
587 | goto unlock_pmd; | |
588 | ||
589 | flush_cache_page(vma, address, pfn); | |
590 | pmd = pmdp_huge_clear_flush(vma, address, pmdp); | |
591 | pmd = pmd_wrprotect(pmd); | |
592 | pmd = pmd_mkclean(pmd); | |
593 | set_pmd_at(vma->vm_mm, address, pmdp, pmd); | |
a4d1a885 | 594 | mmu_notifier_invalidate_range(vma->vm_mm, start, end); |
f729c8c9 RZ |
595 | unlock_pmd: |
596 | spin_unlock(ptl); | |
597 | #endif | |
598 | } else { | |
599 | if (pfn != pte_pfn(*ptep)) | |
600 | goto unlock_pte; | |
601 | if (!pte_dirty(*ptep) && !pte_write(*ptep)) | |
602 | goto unlock_pte; | |
603 | ||
604 | flush_cache_page(vma, address, pfn); | |
605 | pte = ptep_clear_flush(vma, address, ptep); | |
606 | pte = pte_wrprotect(pte); | |
607 | pte = pte_mkclean(pte); | |
608 | set_pte_at(vma->vm_mm, address, ptep, pte); | |
a4d1a885 | 609 | mmu_notifier_invalidate_range(vma->vm_mm, start, end); |
f729c8c9 RZ |
610 | unlock_pte: |
611 | pte_unmap_unlock(ptep, ptl); | |
612 | } | |
4b4bb46d | 613 | |
a4d1a885 | 614 | mmu_notifier_invalidate_range_end(vma->vm_mm, start, end); |
4b4bb46d JK |
615 | } |
616 | i_mmap_unlock_read(mapping); | |
617 | } | |
618 | ||
9973c98e | 619 | static int dax_writeback_one(struct block_device *bdev, |
cccbce67 DW |
620 | struct dax_device *dax_dev, struct address_space *mapping, |
621 | pgoff_t index, void *entry) | |
9973c98e RZ |
622 | { |
623 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
cccbce67 DW |
624 | void *entry2, **slot, *kaddr; |
625 | long ret = 0, id; | |
626 | sector_t sector; | |
627 | pgoff_t pgoff; | |
628 | size_t size; | |
629 | pfn_t pfn; | |
9973c98e | 630 | |
9973c98e | 631 | /* |
a6abc2c0 JK |
632 | * A page got tagged dirty in DAX mapping? Something is seriously |
633 | * wrong. | |
9973c98e | 634 | */ |
a6abc2c0 JK |
635 | if (WARN_ON(!radix_tree_exceptional_entry(entry))) |
636 | return -EIO; | |
9973c98e | 637 | |
a6abc2c0 JK |
638 | spin_lock_irq(&mapping->tree_lock); |
639 | entry2 = get_unlocked_mapping_entry(mapping, index, &slot); | |
640 | /* Entry got punched out / reallocated? */ | |
54585e35 | 641 | if (!entry2 || WARN_ON_ONCE(!radix_tree_exceptional_entry(entry2))) |
a6abc2c0 JK |
642 | goto put_unlocked; |
643 | /* | |
644 | * Entry got reallocated elsewhere? No need to writeback. We have to | |
645 | * compare sectors as we must not bail out due to difference in lockbit | |
646 | * or entry type. | |
647 | */ | |
648 | if (dax_radix_sector(entry2) != dax_radix_sector(entry)) | |
649 | goto put_unlocked; | |
642261ac RZ |
650 | if (WARN_ON_ONCE(dax_is_empty_entry(entry) || |
651 | dax_is_zero_entry(entry))) { | |
9973c98e | 652 | ret = -EIO; |
a6abc2c0 | 653 | goto put_unlocked; |
9973c98e RZ |
654 | } |
655 | ||
a6abc2c0 JK |
656 | /* Another fsync thread may have already written back this entry */ |
657 | if (!radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE)) | |
658 | goto put_unlocked; | |
659 | /* Lock the entry to serialize with page faults */ | |
660 | entry = lock_slot(mapping, slot); | |
661 | /* | |
662 | * We can clear the tag now but we have to be careful so that concurrent | |
663 | * dax_writeback_one() calls for the same index cannot finish before we | |
664 | * actually flush the caches. This is achieved as the calls will look | |
665 | * at the entry only under tree_lock and once they do that they will | |
666 | * see the entry locked and wait for it to unlock. | |
667 | */ | |
668 | radix_tree_tag_clear(page_tree, index, PAGECACHE_TAG_TOWRITE); | |
669 | spin_unlock_irq(&mapping->tree_lock); | |
670 | ||
642261ac RZ |
671 | /* |
672 | * Even if dax_writeback_mapping_range() was given a wbc->range_start | |
673 | * in the middle of a PMD, the 'index' we are given will be aligned to | |
674 | * the start index of the PMD, as will the sector we pull from | |
675 | * 'entry'. This allows us to flush for PMD_SIZE and not have to | |
676 | * worry about partial PMD writebacks. | |
677 | */ | |
cccbce67 DW |
678 | sector = dax_radix_sector(entry); |
679 | size = PAGE_SIZE << dax_radix_order(entry); | |
680 | ||
681 | id = dax_read_lock(); | |
682 | ret = bdev_dax_pgoff(bdev, sector, size, &pgoff); | |
683 | if (ret) | |
684 | goto dax_unlock; | |
9973c98e RZ |
685 | |
686 | /* | |
cccbce67 DW |
687 | * dax_direct_access() may sleep, so cannot hold tree_lock over |
688 | * its invocation. | |
9973c98e | 689 | */ |
cccbce67 DW |
690 | ret = dax_direct_access(dax_dev, pgoff, size / PAGE_SIZE, &kaddr, &pfn); |
691 | if (ret < 0) | |
692 | goto dax_unlock; | |
9973c98e | 693 | |
cccbce67 | 694 | if (WARN_ON_ONCE(ret < size / PAGE_SIZE)) { |
9973c98e | 695 | ret = -EIO; |
cccbce67 | 696 | goto dax_unlock; |
9973c98e RZ |
697 | } |
698 | ||
cccbce67 | 699 | dax_mapping_entry_mkclean(mapping, index, pfn_t_to_pfn(pfn)); |
6318770a | 700 | dax_flush(dax_dev, pgoff, kaddr, size); |
4b4bb46d JK |
701 | /* |
702 | * After we have flushed the cache, we can clear the dirty tag. There | |
703 | * cannot be new dirty data in the pfn after the flush has completed as | |
704 | * the pfn mappings are writeprotected and fault waits for mapping | |
705 | * entry lock. | |
706 | */ | |
707 | spin_lock_irq(&mapping->tree_lock); | |
708 | radix_tree_tag_clear(page_tree, index, PAGECACHE_TAG_DIRTY); | |
709 | spin_unlock_irq(&mapping->tree_lock); | |
f9bc3a07 | 710 | trace_dax_writeback_one(mapping->host, index, size >> PAGE_SHIFT); |
cccbce67 DW |
711 | dax_unlock: |
712 | dax_read_unlock(id); | |
54585e35 | 713 | put_locked_mapping_entry(mapping, index); |
9973c98e RZ |
714 | return ret; |
715 | ||
a6abc2c0 JK |
716 | put_unlocked: |
717 | put_unlocked_mapping_entry(mapping, index, entry2); | |
9973c98e RZ |
718 | spin_unlock_irq(&mapping->tree_lock); |
719 | return ret; | |
720 | } | |
721 | ||
722 | /* | |
723 | * Flush the mapping to the persistent domain within the byte range of [start, | |
724 | * end]. This is required by data integrity operations to ensure file data is | |
725 | * on persistent storage prior to completion of the operation. | |
726 | */ | |
7f6d5b52 RZ |
727 | int dax_writeback_mapping_range(struct address_space *mapping, |
728 | struct block_device *bdev, struct writeback_control *wbc) | |
9973c98e RZ |
729 | { |
730 | struct inode *inode = mapping->host; | |
642261ac | 731 | pgoff_t start_index, end_index; |
9973c98e | 732 | pgoff_t indices[PAGEVEC_SIZE]; |
cccbce67 | 733 | struct dax_device *dax_dev; |
9973c98e RZ |
734 | struct pagevec pvec; |
735 | bool done = false; | |
736 | int i, ret = 0; | |
9973c98e RZ |
737 | |
738 | if (WARN_ON_ONCE(inode->i_blkbits != PAGE_SHIFT)) | |
739 | return -EIO; | |
740 | ||
7f6d5b52 RZ |
741 | if (!mapping->nrexceptional || wbc->sync_mode != WB_SYNC_ALL) |
742 | return 0; | |
743 | ||
cccbce67 DW |
744 | dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
745 | if (!dax_dev) | |
746 | return -EIO; | |
747 | ||
09cbfeaf KS |
748 | start_index = wbc->range_start >> PAGE_SHIFT; |
749 | end_index = wbc->range_end >> PAGE_SHIFT; | |
9973c98e | 750 | |
d14a3f48 RZ |
751 | trace_dax_writeback_range(inode, start_index, end_index); |
752 | ||
9973c98e RZ |
753 | tag_pages_for_writeback(mapping, start_index, end_index); |
754 | ||
755 | pagevec_init(&pvec, 0); | |
756 | while (!done) { | |
757 | pvec.nr = find_get_entries_tag(mapping, start_index, | |
758 | PAGECACHE_TAG_TOWRITE, PAGEVEC_SIZE, | |
759 | pvec.pages, indices); | |
760 | ||
761 | if (pvec.nr == 0) | |
762 | break; | |
763 | ||
764 | for (i = 0; i < pvec.nr; i++) { | |
765 | if (indices[i] > end_index) { | |
766 | done = true; | |
767 | break; | |
768 | } | |
769 | ||
cccbce67 DW |
770 | ret = dax_writeback_one(bdev, dax_dev, mapping, |
771 | indices[i], pvec.pages[i]); | |
819ec6b9 JL |
772 | if (ret < 0) { |
773 | mapping_set_error(mapping, ret); | |
d14a3f48 | 774 | goto out; |
819ec6b9 | 775 | } |
9973c98e | 776 | } |
1eb643d0 | 777 | start_index = indices[pvec.nr - 1] + 1; |
9973c98e | 778 | } |
d14a3f48 | 779 | out: |
cccbce67 | 780 | put_dax(dax_dev); |
d14a3f48 RZ |
781 | trace_dax_writeback_range_done(inode, start_index, end_index); |
782 | return (ret < 0 ? ret : 0); | |
9973c98e RZ |
783 | } |
784 | EXPORT_SYMBOL_GPL(dax_writeback_mapping_range); | |
785 | ||
ac401cc7 | 786 | static int dax_insert_mapping(struct address_space *mapping, |
cccbce67 | 787 | struct block_device *bdev, struct dax_device *dax_dev, |
54585e35 | 788 | sector_t sector, size_t size, void *entry, |
cccbce67 | 789 | struct vm_area_struct *vma, struct vm_fault *vmf) |
f7ca90b1 | 790 | { |
1a29d85e | 791 | unsigned long vaddr = vmf->address; |
cccbce67 DW |
792 | void *ret, *kaddr; |
793 | pgoff_t pgoff; | |
794 | int id, rc; | |
795 | pfn_t pfn; | |
f7ca90b1 | 796 | |
cccbce67 DW |
797 | rc = bdev_dax_pgoff(bdev, sector, size, &pgoff); |
798 | if (rc) | |
799 | return rc; | |
f7ca90b1 | 800 | |
cccbce67 DW |
801 | id = dax_read_lock(); |
802 | rc = dax_direct_access(dax_dev, pgoff, PHYS_PFN(size), &kaddr, &pfn); | |
803 | if (rc < 0) { | |
804 | dax_read_unlock(id); | |
805 | return rc; | |
806 | } | |
807 | dax_read_unlock(id); | |
808 | ||
809 | ret = dax_insert_mapping_entry(mapping, vmf, entry, sector, 0); | |
4d9a2c87 JK |
810 | if (IS_ERR(ret)) |
811 | return PTR_ERR(ret); | |
9973c98e | 812 | |
b4440734 | 813 | trace_dax_insert_mapping(mapping->host, vmf, ret); |
54585e35 RZ |
814 | if (vmf->flags & FAULT_FLAG_WRITE) |
815 | return vm_insert_mixed_mkwrite(vma, vaddr, pfn); | |
816 | else | |
817 | return vm_insert_mixed(vma, vaddr, pfn); | |
0e3b210c | 818 | } |
0e3b210c | 819 | |
7b6956df | 820 | /* |
54585e35 RZ |
821 | * The user has performed a load from a hole in the file. Allocating a new |
822 | * page in the file would cause excessive storage usage for workloads with | |
823 | * sparse files. Instead we insert a read-only mapping of the 4k zero page. | |
824 | * If this page is ever written to we will re-fault and change the mapping to | |
825 | * point to real DAX storage instead. | |
7b6956df | 826 | */ |
54585e35 | 827 | static int dax_load_hole(struct address_space *mapping, void *entry, |
7b6956df RZ |
828 | struct vm_fault *vmf) |
829 | { | |
830 | struct inode *inode = mapping->host; | |
54585e35 RZ |
831 | unsigned long vaddr = vmf->address; |
832 | int ret = VM_FAULT_NOPAGE; | |
833 | struct page *zero_page; | |
834 | void *entry2; | |
7b6956df | 835 | |
54585e35 RZ |
836 | zero_page = ZERO_PAGE(0); |
837 | if (unlikely(!zero_page)) { | |
7b6956df RZ |
838 | ret = VM_FAULT_OOM; |
839 | goto out; | |
840 | } | |
841 | ||
54585e35 RZ |
842 | entry2 = dax_insert_mapping_entry(mapping, vmf, entry, 0, |
843 | RADIX_DAX_ZERO_PAGE); | |
844 | if (IS_ERR(entry2)) { | |
845 | ret = VM_FAULT_SIGBUS; | |
846 | goto out; | |
7b6956df | 847 | } |
54585e35 RZ |
848 | |
849 | vm_insert_mixed(vmf->vma, vaddr, page_to_pfn_t(zero_page)); | |
7b6956df RZ |
850 | out: |
851 | trace_dax_load_hole(inode, vmf, ret); | |
852 | return ret; | |
853 | } | |
854 | ||
4b0228fa VV |
855 | static bool dax_range_is_aligned(struct block_device *bdev, |
856 | unsigned int offset, unsigned int length) | |
857 | { | |
858 | unsigned short sector_size = bdev_logical_block_size(bdev); | |
859 | ||
860 | if (!IS_ALIGNED(offset, sector_size)) | |
861 | return false; | |
862 | if (!IS_ALIGNED(length, sector_size)) | |
863 | return false; | |
864 | ||
865 | return true; | |
866 | } | |
867 | ||
cccbce67 DW |
868 | int __dax_zero_page_range(struct block_device *bdev, |
869 | struct dax_device *dax_dev, sector_t sector, | |
870 | unsigned int offset, unsigned int size) | |
679c8bd3 | 871 | { |
cccbce67 DW |
872 | if (dax_range_is_aligned(bdev, offset, size)) { |
873 | sector_t start_sector = sector + (offset >> 9); | |
4b0228fa VV |
874 | |
875 | return blkdev_issue_zeroout(bdev, start_sector, | |
53ef7d0e | 876 | size >> 9, GFP_NOFS, 0); |
4b0228fa | 877 | } else { |
cccbce67 DW |
878 | pgoff_t pgoff; |
879 | long rc, id; | |
880 | void *kaddr; | |
881 | pfn_t pfn; | |
882 | ||
e84b83b9 | 883 | rc = bdev_dax_pgoff(bdev, sector, PAGE_SIZE, &pgoff); |
cccbce67 DW |
884 | if (rc) |
885 | return rc; | |
886 | ||
887 | id = dax_read_lock(); | |
e84b83b9 | 888 | rc = dax_direct_access(dax_dev, pgoff, 1, &kaddr, |
cccbce67 DW |
889 | &pfn); |
890 | if (rc < 0) { | |
891 | dax_read_unlock(id); | |
892 | return rc; | |
893 | } | |
81f55870 DW |
894 | memset(kaddr + offset, 0, size); |
895 | dax_flush(dax_dev, pgoff, kaddr + offset, size); | |
cccbce67 | 896 | dax_read_unlock(id); |
4b0228fa | 897 | } |
679c8bd3 CH |
898 | return 0; |
899 | } | |
900 | EXPORT_SYMBOL_GPL(__dax_zero_page_range); | |
901 | ||
333ccc97 | 902 | static sector_t dax_iomap_sector(struct iomap *iomap, loff_t pos) |
25726bc1 | 903 | { |
333ccc97 | 904 | return iomap->blkno + (((pos & PAGE_MASK) - iomap->offset) >> 9); |
25726bc1 | 905 | } |
a254e568 | 906 | |
a254e568 | 907 | static loff_t |
11c59c92 | 908 | dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data, |
a254e568 CH |
909 | struct iomap *iomap) |
910 | { | |
cccbce67 DW |
911 | struct block_device *bdev = iomap->bdev; |
912 | struct dax_device *dax_dev = iomap->dax_dev; | |
a254e568 CH |
913 | struct iov_iter *iter = data; |
914 | loff_t end = pos + length, done = 0; | |
915 | ssize_t ret = 0; | |
cccbce67 | 916 | int id; |
a254e568 CH |
917 | |
918 | if (iov_iter_rw(iter) == READ) { | |
919 | end = min(end, i_size_read(inode)); | |
920 | if (pos >= end) | |
921 | return 0; | |
922 | ||
923 | if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN) | |
924 | return iov_iter_zero(min(length, end - pos), iter); | |
925 | } | |
926 | ||
927 | if (WARN_ON_ONCE(iomap->type != IOMAP_MAPPED)) | |
928 | return -EIO; | |
929 | ||
e3fce68c JK |
930 | /* |
931 | * Write can allocate block for an area which has a hole page mapped | |
932 | * into page tables. We have to tear down these mappings so that data | |
933 | * written by write(2) is visible in mmap. | |
934 | */ | |
cd656375 | 935 | if (iomap->flags & IOMAP_F_NEW) { |
e3fce68c JK |
936 | invalidate_inode_pages2_range(inode->i_mapping, |
937 | pos >> PAGE_SHIFT, | |
938 | (end - 1) >> PAGE_SHIFT); | |
939 | } | |
940 | ||
cccbce67 | 941 | id = dax_read_lock(); |
a254e568 CH |
942 | while (pos < end) { |
943 | unsigned offset = pos & (PAGE_SIZE - 1); | |
cccbce67 DW |
944 | const size_t size = ALIGN(length + offset, PAGE_SIZE); |
945 | const sector_t sector = dax_iomap_sector(iomap, pos); | |
a254e568 | 946 | ssize_t map_len; |
cccbce67 DW |
947 | pgoff_t pgoff; |
948 | void *kaddr; | |
949 | pfn_t pfn; | |
a254e568 | 950 | |
d1908f52 MH |
951 | if (fatal_signal_pending(current)) { |
952 | ret = -EINTR; | |
953 | break; | |
954 | } | |
955 | ||
cccbce67 DW |
956 | ret = bdev_dax_pgoff(bdev, sector, size, &pgoff); |
957 | if (ret) | |
958 | break; | |
959 | ||
960 | map_len = dax_direct_access(dax_dev, pgoff, PHYS_PFN(size), | |
961 | &kaddr, &pfn); | |
a254e568 CH |
962 | if (map_len < 0) { |
963 | ret = map_len; | |
964 | break; | |
965 | } | |
966 | ||
cccbce67 DW |
967 | map_len = PFN_PHYS(map_len); |
968 | kaddr += offset; | |
a254e568 CH |
969 | map_len -= offset; |
970 | if (map_len > end - pos) | |
971 | map_len = end - pos; | |
972 | ||
973 | if (iov_iter_rw(iter) == WRITE) | |
fec53774 DW |
974 | map_len = dax_copy_from_iter(dax_dev, pgoff, kaddr, |
975 | map_len, iter); | |
a254e568 | 976 | else |
cccbce67 | 977 | map_len = copy_to_iter(kaddr, map_len, iter); |
a254e568 CH |
978 | if (map_len <= 0) { |
979 | ret = map_len ? map_len : -EFAULT; | |
980 | break; | |
981 | } | |
982 | ||
983 | pos += map_len; | |
984 | length -= map_len; | |
985 | done += map_len; | |
986 | } | |
cccbce67 | 987 | dax_read_unlock(id); |
a254e568 CH |
988 | |
989 | return done ? done : ret; | |
990 | } | |
991 | ||
992 | /** | |
11c59c92 | 993 | * dax_iomap_rw - Perform I/O to a DAX file |
a254e568 CH |
994 | * @iocb: The control block for this I/O |
995 | * @iter: The addresses to do I/O from or to | |
996 | * @ops: iomap ops passed from the file system | |
997 | * | |
998 | * This function performs read and write operations to directly mapped | |
999 | * persistent memory. The callers needs to take care of read/write exclusion | |
1000 | * and evicting any page cache pages in the region under I/O. | |
1001 | */ | |
1002 | ssize_t | |
11c59c92 | 1003 | dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter, |
8ff6daa1 | 1004 | const struct iomap_ops *ops) |
a254e568 CH |
1005 | { |
1006 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
1007 | struct inode *inode = mapping->host; | |
1008 | loff_t pos = iocb->ki_pos, ret = 0, done = 0; | |
1009 | unsigned flags = 0; | |
1010 | ||
168316db CH |
1011 | if (iov_iter_rw(iter) == WRITE) { |
1012 | lockdep_assert_held_exclusive(&inode->i_rwsem); | |
a254e568 | 1013 | flags |= IOMAP_WRITE; |
168316db CH |
1014 | } else { |
1015 | lockdep_assert_held(&inode->i_rwsem); | |
1016 | } | |
a254e568 | 1017 | |
a254e568 CH |
1018 | while (iov_iter_count(iter)) { |
1019 | ret = iomap_apply(inode, pos, iov_iter_count(iter), flags, ops, | |
11c59c92 | 1020 | iter, dax_iomap_actor); |
a254e568 CH |
1021 | if (ret <= 0) |
1022 | break; | |
1023 | pos += ret; | |
1024 | done += ret; | |
1025 | } | |
1026 | ||
1027 | iocb->ki_pos += done; | |
1028 | return done ? done : ret; | |
1029 | } | |
11c59c92 | 1030 | EXPORT_SYMBOL_GPL(dax_iomap_rw); |
a7d73fe6 | 1031 | |
9f141d6e JK |
1032 | static int dax_fault_return(int error) |
1033 | { | |
1034 | if (error == 0) | |
1035 | return VM_FAULT_NOPAGE; | |
1036 | if (error == -ENOMEM) | |
1037 | return VM_FAULT_OOM; | |
1038 | return VM_FAULT_SIGBUS; | |
1039 | } | |
1040 | ||
a2d58167 DJ |
1041 | static int dax_iomap_pte_fault(struct vm_fault *vmf, |
1042 | const struct iomap_ops *ops) | |
a7d73fe6 | 1043 | { |
11bac800 | 1044 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
a7d73fe6 | 1045 | struct inode *inode = mapping->host; |
1a29d85e | 1046 | unsigned long vaddr = vmf->address; |
a7d73fe6 CH |
1047 | loff_t pos = (loff_t)vmf->pgoff << PAGE_SHIFT; |
1048 | sector_t sector; | |
1049 | struct iomap iomap = { 0 }; | |
9484ab1b | 1050 | unsigned flags = IOMAP_FAULT; |
a7d73fe6 | 1051 | int error, major = 0; |
b1aa812b | 1052 | int vmf_ret = 0; |
a7d73fe6 CH |
1053 | void *entry; |
1054 | ||
a9c42b33 | 1055 | trace_dax_pte_fault(inode, vmf, vmf_ret); |
a7d73fe6 CH |
1056 | /* |
1057 | * Check whether offset isn't beyond end of file now. Caller is supposed | |
1058 | * to hold locks serializing us with truncate / punch hole so this is | |
1059 | * a reliable test. | |
1060 | */ | |
a9c42b33 RZ |
1061 | if (pos >= i_size_read(inode)) { |
1062 | vmf_ret = VM_FAULT_SIGBUS; | |
1063 | goto out; | |
1064 | } | |
a7d73fe6 | 1065 | |
a7d73fe6 CH |
1066 | if ((vmf->flags & FAULT_FLAG_WRITE) && !vmf->cow_page) |
1067 | flags |= IOMAP_WRITE; | |
1068 | ||
13e451fd JK |
1069 | entry = grab_mapping_entry(mapping, vmf->pgoff, 0); |
1070 | if (IS_ERR(entry)) { | |
1071 | vmf_ret = dax_fault_return(PTR_ERR(entry)); | |
1072 | goto out; | |
1073 | } | |
1074 | ||
e2093926 RZ |
1075 | /* |
1076 | * It is possible, particularly with mixed reads & writes to private | |
1077 | * mappings, that we have raced with a PMD fault that overlaps with | |
1078 | * the PTE we need to set up. If so just return and the fault will be | |
1079 | * retried. | |
1080 | */ | |
1081 | if (pmd_trans_huge(*vmf->pmd) || pmd_devmap(*vmf->pmd)) { | |
1082 | vmf_ret = VM_FAULT_NOPAGE; | |
1083 | goto unlock_entry; | |
1084 | } | |
1085 | ||
a7d73fe6 CH |
1086 | /* |
1087 | * Note that we don't bother to use iomap_apply here: DAX required | |
1088 | * the file system block size to be equal the page size, which means | |
1089 | * that we never have to deal with more than a single extent here. | |
1090 | */ | |
1091 | error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap); | |
a9c42b33 RZ |
1092 | if (error) { |
1093 | vmf_ret = dax_fault_return(error); | |
13e451fd | 1094 | goto unlock_entry; |
a9c42b33 | 1095 | } |
a7d73fe6 | 1096 | if (WARN_ON_ONCE(iomap.offset + iomap.length < pos + PAGE_SIZE)) { |
13e451fd JK |
1097 | error = -EIO; /* fs corruption? */ |
1098 | goto error_finish_iomap; | |
a7d73fe6 CH |
1099 | } |
1100 | ||
333ccc97 | 1101 | sector = dax_iomap_sector(&iomap, pos); |
a7d73fe6 CH |
1102 | |
1103 | if (vmf->cow_page) { | |
1104 | switch (iomap.type) { | |
1105 | case IOMAP_HOLE: | |
1106 | case IOMAP_UNWRITTEN: | |
1107 | clear_user_highpage(vmf->cow_page, vaddr); | |
1108 | break; | |
1109 | case IOMAP_MAPPED: | |
cccbce67 DW |
1110 | error = copy_user_dax(iomap.bdev, iomap.dax_dev, |
1111 | sector, PAGE_SIZE, vmf->cow_page, vaddr); | |
a7d73fe6 CH |
1112 | break; |
1113 | default: | |
1114 | WARN_ON_ONCE(1); | |
1115 | error = -EIO; | |
1116 | break; | |
1117 | } | |
1118 | ||
1119 | if (error) | |
13e451fd | 1120 | goto error_finish_iomap; |
b1aa812b JK |
1121 | |
1122 | __SetPageUptodate(vmf->cow_page); | |
1123 | vmf_ret = finish_fault(vmf); | |
1124 | if (!vmf_ret) | |
1125 | vmf_ret = VM_FAULT_DONE_COW; | |
13e451fd | 1126 | goto finish_iomap; |
a7d73fe6 CH |
1127 | } |
1128 | ||
1129 | switch (iomap.type) { | |
1130 | case IOMAP_MAPPED: | |
1131 | if (iomap.flags & IOMAP_F_NEW) { | |
1132 | count_vm_event(PGMAJFAULT); | |
2262185c | 1133 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
a7d73fe6 CH |
1134 | major = VM_FAULT_MAJOR; |
1135 | } | |
cccbce67 | 1136 | error = dax_insert_mapping(mapping, iomap.bdev, iomap.dax_dev, |
54585e35 | 1137 | sector, PAGE_SIZE, entry, vmf->vma, vmf); |
9f141d6e JK |
1138 | /* -EBUSY is fine, somebody else faulted on the same PTE */ |
1139 | if (error == -EBUSY) | |
1140 | error = 0; | |
a7d73fe6 CH |
1141 | break; |
1142 | case IOMAP_UNWRITTEN: | |
1143 | case IOMAP_HOLE: | |
1550290b | 1144 | if (!(vmf->flags & FAULT_FLAG_WRITE)) { |
54585e35 | 1145 | vmf_ret = dax_load_hole(mapping, entry, vmf); |
13e451fd | 1146 | goto finish_iomap; |
1550290b | 1147 | } |
a7d73fe6 CH |
1148 | /*FALLTHRU*/ |
1149 | default: | |
1150 | WARN_ON_ONCE(1); | |
1151 | error = -EIO; | |
1152 | break; | |
1153 | } | |
1154 | ||
13e451fd | 1155 | error_finish_iomap: |
9f141d6e | 1156 | vmf_ret = dax_fault_return(error) | major; |
9f141d6e JK |
1157 | finish_iomap: |
1158 | if (ops->iomap_end) { | |
1159 | int copied = PAGE_SIZE; | |
1160 | ||
1161 | if (vmf_ret & VM_FAULT_ERROR) | |
1162 | copied = 0; | |
1163 | /* | |
1164 | * The fault is done by now and there's no way back (other | |
1165 | * thread may be already happily using PTE we have installed). | |
1166 | * Just ignore error from ->iomap_end since we cannot do much | |
1167 | * with it. | |
1168 | */ | |
1169 | ops->iomap_end(inode, pos, PAGE_SIZE, copied, flags, &iomap); | |
1550290b | 1170 | } |
13e451fd | 1171 | unlock_entry: |
54585e35 | 1172 | put_locked_mapping_entry(mapping, vmf->pgoff); |
13e451fd | 1173 | out: |
a9c42b33 | 1174 | trace_dax_pte_fault_done(inode, vmf, vmf_ret); |
9f141d6e | 1175 | return vmf_ret; |
a7d73fe6 | 1176 | } |
642261ac RZ |
1177 | |
1178 | #ifdef CONFIG_FS_DAX_PMD | |
1179 | /* | |
1180 | * The 'colour' (ie low bits) within a PMD of a page offset. This comes up | |
1181 | * more often than one might expect in the below functions. | |
1182 | */ | |
1183 | #define PG_PMD_COLOUR ((PMD_SIZE >> PAGE_SHIFT) - 1) | |
1184 | ||
f4200391 | 1185 | static int dax_pmd_insert_mapping(struct vm_fault *vmf, struct iomap *iomap, |
54585e35 | 1186 | loff_t pos, void *entry) |
642261ac | 1187 | { |
f4200391 | 1188 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
cccbce67 DW |
1189 | const sector_t sector = dax_iomap_sector(iomap, pos); |
1190 | struct dax_device *dax_dev = iomap->dax_dev; | |
642261ac | 1191 | struct block_device *bdev = iomap->bdev; |
27a7ffac | 1192 | struct inode *inode = mapping->host; |
cccbce67 DW |
1193 | const size_t size = PMD_SIZE; |
1194 | void *ret = NULL, *kaddr; | |
1195 | long length = 0; | |
1196 | pgoff_t pgoff; | |
1197 | pfn_t pfn; | |
1198 | int id; | |
1199 | ||
1200 | if (bdev_dax_pgoff(bdev, sector, size, &pgoff) != 0) | |
27a7ffac | 1201 | goto fallback; |
642261ac | 1202 | |
cccbce67 DW |
1203 | id = dax_read_lock(); |
1204 | length = dax_direct_access(dax_dev, pgoff, PHYS_PFN(size), &kaddr, &pfn); | |
1205 | if (length < 0) | |
1206 | goto unlock_fallback; | |
1207 | length = PFN_PHYS(length); | |
1208 | ||
1209 | if (length < size) | |
1210 | goto unlock_fallback; | |
1211 | if (pfn_t_to_pfn(pfn) & PG_PMD_COLOUR) | |
1212 | goto unlock_fallback; | |
1213 | if (!pfn_t_devmap(pfn)) | |
1214 | goto unlock_fallback; | |
1215 | dax_read_unlock(id); | |
1216 | ||
54585e35 | 1217 | ret = dax_insert_mapping_entry(mapping, vmf, entry, sector, |
642261ac RZ |
1218 | RADIX_DAX_PMD); |
1219 | if (IS_ERR(ret)) | |
27a7ffac | 1220 | goto fallback; |
642261ac | 1221 | |
cccbce67 | 1222 | trace_dax_pmd_insert_mapping(inode, vmf, length, pfn, ret); |
f4200391 | 1223 | return vmf_insert_pfn_pmd(vmf->vma, vmf->address, vmf->pmd, |
cccbce67 | 1224 | pfn, vmf->flags & FAULT_FLAG_WRITE); |
642261ac | 1225 | |
cccbce67 DW |
1226 | unlock_fallback: |
1227 | dax_read_unlock(id); | |
27a7ffac | 1228 | fallback: |
cccbce67 | 1229 | trace_dax_pmd_insert_mapping_fallback(inode, vmf, length, pfn, ret); |
642261ac RZ |
1230 | return VM_FAULT_FALLBACK; |
1231 | } | |
1232 | ||
f4200391 | 1233 | static int dax_pmd_load_hole(struct vm_fault *vmf, struct iomap *iomap, |
54585e35 | 1234 | void *entry) |
642261ac | 1235 | { |
f4200391 DJ |
1236 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
1237 | unsigned long pmd_addr = vmf->address & PMD_MASK; | |
653b2ea3 | 1238 | struct inode *inode = mapping->host; |
642261ac | 1239 | struct page *zero_page; |
653b2ea3 | 1240 | void *ret = NULL; |
642261ac RZ |
1241 | spinlock_t *ptl; |
1242 | pmd_t pmd_entry; | |
642261ac | 1243 | |
f4200391 | 1244 | zero_page = mm_get_huge_zero_page(vmf->vma->vm_mm); |
642261ac RZ |
1245 | |
1246 | if (unlikely(!zero_page)) | |
653b2ea3 | 1247 | goto fallback; |
642261ac | 1248 | |
54585e35 RZ |
1249 | ret = dax_insert_mapping_entry(mapping, vmf, entry, 0, |
1250 | RADIX_DAX_PMD | RADIX_DAX_ZERO_PAGE); | |
642261ac | 1251 | if (IS_ERR(ret)) |
653b2ea3 | 1252 | goto fallback; |
642261ac | 1253 | |
f4200391 DJ |
1254 | ptl = pmd_lock(vmf->vma->vm_mm, vmf->pmd); |
1255 | if (!pmd_none(*(vmf->pmd))) { | |
642261ac | 1256 | spin_unlock(ptl); |
653b2ea3 | 1257 | goto fallback; |
642261ac RZ |
1258 | } |
1259 | ||
f4200391 | 1260 | pmd_entry = mk_pmd(zero_page, vmf->vma->vm_page_prot); |
642261ac | 1261 | pmd_entry = pmd_mkhuge(pmd_entry); |
f4200391 | 1262 | set_pmd_at(vmf->vma->vm_mm, pmd_addr, vmf->pmd, pmd_entry); |
642261ac | 1263 | spin_unlock(ptl); |
f4200391 | 1264 | trace_dax_pmd_load_hole(inode, vmf, zero_page, ret); |
642261ac | 1265 | return VM_FAULT_NOPAGE; |
653b2ea3 RZ |
1266 | |
1267 | fallback: | |
f4200391 | 1268 | trace_dax_pmd_load_hole_fallback(inode, vmf, zero_page, ret); |
653b2ea3 | 1269 | return VM_FAULT_FALLBACK; |
642261ac RZ |
1270 | } |
1271 | ||
a2d58167 DJ |
1272 | static int dax_iomap_pmd_fault(struct vm_fault *vmf, |
1273 | const struct iomap_ops *ops) | |
642261ac | 1274 | { |
f4200391 | 1275 | struct vm_area_struct *vma = vmf->vma; |
642261ac | 1276 | struct address_space *mapping = vma->vm_file->f_mapping; |
d8a849e1 DJ |
1277 | unsigned long pmd_addr = vmf->address & PMD_MASK; |
1278 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
9484ab1b | 1279 | unsigned int iomap_flags = (write ? IOMAP_WRITE : 0) | IOMAP_FAULT; |
642261ac RZ |
1280 | struct inode *inode = mapping->host; |
1281 | int result = VM_FAULT_FALLBACK; | |
1282 | struct iomap iomap = { 0 }; | |
1283 | pgoff_t max_pgoff, pgoff; | |
642261ac RZ |
1284 | void *entry; |
1285 | loff_t pos; | |
1286 | int error; | |
1287 | ||
282a8e03 RZ |
1288 | /* |
1289 | * Check whether offset isn't beyond end of file now. Caller is | |
1290 | * supposed to hold locks serializing us with truncate / punch hole so | |
1291 | * this is a reliable test. | |
1292 | */ | |
1293 | pgoff = linear_page_index(vma, pmd_addr); | |
1294 | max_pgoff = (i_size_read(inode) - 1) >> PAGE_SHIFT; | |
1295 | ||
f4200391 | 1296 | trace_dax_pmd_fault(inode, vmf, max_pgoff, 0); |
282a8e03 | 1297 | |
fffa281b RZ |
1298 | /* |
1299 | * Make sure that the faulting address's PMD offset (color) matches | |
1300 | * the PMD offset from the start of the file. This is necessary so | |
1301 | * that a PMD range in the page table overlaps exactly with a PMD | |
1302 | * range in the radix tree. | |
1303 | */ | |
1304 | if ((vmf->pgoff & PG_PMD_COLOUR) != | |
1305 | ((vmf->address >> PAGE_SHIFT) & PG_PMD_COLOUR)) | |
1306 | goto fallback; | |
1307 | ||
642261ac RZ |
1308 | /* Fall back to PTEs if we're going to COW */ |
1309 | if (write && !(vma->vm_flags & VM_SHARED)) | |
1310 | goto fallback; | |
1311 | ||
1312 | /* If the PMD would extend outside the VMA */ | |
1313 | if (pmd_addr < vma->vm_start) | |
1314 | goto fallback; | |
1315 | if ((pmd_addr + PMD_SIZE) > vma->vm_end) | |
1316 | goto fallback; | |
1317 | ||
282a8e03 RZ |
1318 | if (pgoff > max_pgoff) { |
1319 | result = VM_FAULT_SIGBUS; | |
1320 | goto out; | |
1321 | } | |
642261ac RZ |
1322 | |
1323 | /* If the PMD would extend beyond the file size */ | |
1324 | if ((pgoff | PG_PMD_COLOUR) > max_pgoff) | |
1325 | goto fallback; | |
1326 | ||
876f2946 | 1327 | /* |
54585e35 RZ |
1328 | * grab_mapping_entry() will make sure we get a 2MiB empty entry, a |
1329 | * 2MiB zero page entry or a DAX PMD. If it can't (because a 4k page | |
1330 | * is already in the tree, for instance), it will return -EEXIST and | |
1331 | * we just fall back to 4k entries. | |
876f2946 RZ |
1332 | */ |
1333 | entry = grab_mapping_entry(mapping, pgoff, RADIX_DAX_PMD); | |
1334 | if (IS_ERR(entry)) | |
1335 | goto fallback; | |
1336 | ||
e2093926 RZ |
1337 | /* |
1338 | * It is possible, particularly with mixed reads & writes to private | |
1339 | * mappings, that we have raced with a PTE fault that overlaps with | |
1340 | * the PMD we need to set up. If so just return and the fault will be | |
1341 | * retried. | |
1342 | */ | |
1343 | if (!pmd_none(*vmf->pmd) && !pmd_trans_huge(*vmf->pmd) && | |
1344 | !pmd_devmap(*vmf->pmd)) { | |
1345 | result = 0; | |
1346 | goto unlock_entry; | |
1347 | } | |
1348 | ||
642261ac RZ |
1349 | /* |
1350 | * Note that we don't use iomap_apply here. We aren't doing I/O, only | |
1351 | * setting up a mapping, so really we're using iomap_begin() as a way | |
1352 | * to look up our filesystem block. | |
1353 | */ | |
1354 | pos = (loff_t)pgoff << PAGE_SHIFT; | |
1355 | error = ops->iomap_begin(inode, pos, PMD_SIZE, iomap_flags, &iomap); | |
1356 | if (error) | |
876f2946 | 1357 | goto unlock_entry; |
9f141d6e | 1358 | |
642261ac RZ |
1359 | if (iomap.offset + iomap.length < pos + PMD_SIZE) |
1360 | goto finish_iomap; | |
1361 | ||
642261ac RZ |
1362 | switch (iomap.type) { |
1363 | case IOMAP_MAPPED: | |
54585e35 | 1364 | result = dax_pmd_insert_mapping(vmf, &iomap, pos, entry); |
642261ac RZ |
1365 | break; |
1366 | case IOMAP_UNWRITTEN: | |
1367 | case IOMAP_HOLE: | |
1368 | if (WARN_ON_ONCE(write)) | |
876f2946 | 1369 | break; |
54585e35 | 1370 | result = dax_pmd_load_hole(vmf, &iomap, entry); |
642261ac RZ |
1371 | break; |
1372 | default: | |
1373 | WARN_ON_ONCE(1); | |
1374 | break; | |
1375 | } | |
1376 | ||
1377 | finish_iomap: | |
1378 | if (ops->iomap_end) { | |
9f141d6e JK |
1379 | int copied = PMD_SIZE; |
1380 | ||
1381 | if (result == VM_FAULT_FALLBACK) | |
1382 | copied = 0; | |
1383 | /* | |
1384 | * The fault is done by now and there's no way back (other | |
1385 | * thread may be already happily using PMD we have installed). | |
1386 | * Just ignore error from ->iomap_end since we cannot do much | |
1387 | * with it. | |
1388 | */ | |
1389 | ops->iomap_end(inode, pos, PMD_SIZE, copied, iomap_flags, | |
1390 | &iomap); | |
642261ac | 1391 | } |
876f2946 | 1392 | unlock_entry: |
54585e35 | 1393 | put_locked_mapping_entry(mapping, pgoff); |
642261ac RZ |
1394 | fallback: |
1395 | if (result == VM_FAULT_FALLBACK) { | |
d8a849e1 | 1396 | split_huge_pmd(vma, vmf->pmd, vmf->address); |
642261ac RZ |
1397 | count_vm_event(THP_FAULT_FALLBACK); |
1398 | } | |
282a8e03 | 1399 | out: |
f4200391 | 1400 | trace_dax_pmd_fault_done(inode, vmf, max_pgoff, result); |
642261ac RZ |
1401 | return result; |
1402 | } | |
a2d58167 | 1403 | #else |
01cddfe9 AB |
1404 | static int dax_iomap_pmd_fault(struct vm_fault *vmf, |
1405 | const struct iomap_ops *ops) | |
a2d58167 DJ |
1406 | { |
1407 | return VM_FAULT_FALLBACK; | |
1408 | } | |
642261ac | 1409 | #endif /* CONFIG_FS_DAX_PMD */ |
a2d58167 DJ |
1410 | |
1411 | /** | |
1412 | * dax_iomap_fault - handle a page fault on a DAX file | |
1413 | * @vmf: The description of the fault | |
1414 | * @ops: iomap ops passed from the file system | |
1415 | * | |
1416 | * When a page fault occurs, filesystems may call this helper in | |
1417 | * their fault handler for DAX files. dax_iomap_fault() assumes the caller | |
1418 | * has done all the necessary locking for page fault to proceed | |
1419 | * successfully. | |
1420 | */ | |
c791ace1 DJ |
1421 | int dax_iomap_fault(struct vm_fault *vmf, enum page_entry_size pe_size, |
1422 | const struct iomap_ops *ops) | |
a2d58167 | 1423 | { |
c791ace1 DJ |
1424 | switch (pe_size) { |
1425 | case PE_SIZE_PTE: | |
a2d58167 | 1426 | return dax_iomap_pte_fault(vmf, ops); |
c791ace1 | 1427 | case PE_SIZE_PMD: |
a2d58167 DJ |
1428 | return dax_iomap_pmd_fault(vmf, ops); |
1429 | default: | |
1430 | return VM_FAULT_FALLBACK; | |
1431 | } | |
1432 | } | |
1433 | EXPORT_SYMBOL_GPL(dax_iomap_fault); |