]>
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> |
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 | ||
b2e0d162 DW |
57 | static long dax_map_atomic(struct block_device *bdev, struct blk_dax_ctl *dax) |
58 | { | |
59 | struct request_queue *q = bdev->bd_queue; | |
60 | long rc = -EIO; | |
61 | ||
7a9eb206 | 62 | dax->addr = ERR_PTR(-EIO); |
b2e0d162 DW |
63 | if (blk_queue_enter(q, true) != 0) |
64 | return rc; | |
65 | ||
66 | rc = bdev_direct_access(bdev, dax); | |
67 | if (rc < 0) { | |
7a9eb206 | 68 | dax->addr = ERR_PTR(rc); |
b2e0d162 DW |
69 | blk_queue_exit(q); |
70 | return rc; | |
71 | } | |
72 | return rc; | |
73 | } | |
74 | ||
75 | static void dax_unmap_atomic(struct block_device *bdev, | |
76 | const struct blk_dax_ctl *dax) | |
77 | { | |
78 | if (IS_ERR(dax->addr)) | |
79 | return; | |
80 | blk_queue_exit(bdev->bd_queue); | |
81 | } | |
82 | ||
642261ac | 83 | static int dax_is_pmd_entry(void *entry) |
d1a5f2b4 | 84 | { |
642261ac | 85 | return (unsigned long)entry & RADIX_DAX_PMD; |
d1a5f2b4 DW |
86 | } |
87 | ||
642261ac | 88 | static int dax_is_pte_entry(void *entry) |
d475c634 | 89 | { |
642261ac | 90 | return !((unsigned long)entry & RADIX_DAX_PMD); |
d475c634 MW |
91 | } |
92 | ||
642261ac | 93 | static int dax_is_zero_entry(void *entry) |
d475c634 | 94 | { |
642261ac | 95 | return (unsigned long)entry & RADIX_DAX_HZP; |
d475c634 MW |
96 | } |
97 | ||
642261ac | 98 | static int dax_is_empty_entry(void *entry) |
b2e0d162 | 99 | { |
642261ac | 100 | return (unsigned long)entry & RADIX_DAX_EMPTY; |
b2e0d162 DW |
101 | } |
102 | ||
d1a5f2b4 | 103 | struct page *read_dax_sector(struct block_device *bdev, sector_t n) |
d475c634 | 104 | { |
d1a5f2b4 | 105 | struct page *page = alloc_pages(GFP_KERNEL, 0); |
b2e0d162 | 106 | struct blk_dax_ctl dax = { |
d1a5f2b4 DW |
107 | .size = PAGE_SIZE, |
108 | .sector = n & ~((((int) PAGE_SIZE) / 512) - 1), | |
b2e0d162 | 109 | }; |
d1a5f2b4 | 110 | long rc; |
d475c634 | 111 | |
d1a5f2b4 DW |
112 | if (!page) |
113 | return ERR_PTR(-ENOMEM); | |
d475c634 | 114 | |
d1a5f2b4 DW |
115 | rc = dax_map_atomic(bdev, &dax); |
116 | if (rc < 0) | |
117 | return ERR_PTR(rc); | |
118 | memcpy_from_pmem(page_address(page), dax.addr, PAGE_SIZE); | |
b2e0d162 | 119 | dax_unmap_atomic(bdev, &dax); |
d1a5f2b4 | 120 | return page; |
d475c634 | 121 | } |
f7ca90b1 | 122 | |
ac401cc7 JK |
123 | /* |
124 | * DAX radix tree locking | |
125 | */ | |
126 | struct exceptional_entry_key { | |
127 | struct address_space *mapping; | |
63e95b5c | 128 | pgoff_t entry_start; |
ac401cc7 JK |
129 | }; |
130 | ||
131 | struct wait_exceptional_entry_queue { | |
132 | wait_queue_t wait; | |
133 | struct exceptional_entry_key key; | |
134 | }; | |
135 | ||
63e95b5c RZ |
136 | static wait_queue_head_t *dax_entry_waitqueue(struct address_space *mapping, |
137 | pgoff_t index, void *entry, struct exceptional_entry_key *key) | |
138 | { | |
139 | unsigned long hash; | |
140 | ||
141 | /* | |
142 | * If 'entry' is a PMD, align the 'index' that we use for the wait | |
143 | * queue to the start of that PMD. This ensures that all offsets in | |
144 | * the range covered by the PMD map to the same bit lock. | |
145 | */ | |
642261ac | 146 | if (dax_is_pmd_entry(entry)) |
63e95b5c RZ |
147 | index &= ~((1UL << (PMD_SHIFT - PAGE_SHIFT)) - 1); |
148 | ||
149 | key->mapping = mapping; | |
150 | key->entry_start = index; | |
151 | ||
152 | hash = hash_long((unsigned long)mapping ^ index, DAX_WAIT_TABLE_BITS); | |
153 | return wait_table + hash; | |
154 | } | |
155 | ||
ac401cc7 JK |
156 | static int wake_exceptional_entry_func(wait_queue_t *wait, unsigned int mode, |
157 | int sync, void *keyp) | |
158 | { | |
159 | struct exceptional_entry_key *key = keyp; | |
160 | struct wait_exceptional_entry_queue *ewait = | |
161 | container_of(wait, struct wait_exceptional_entry_queue, wait); | |
162 | ||
163 | if (key->mapping != ewait->key.mapping || | |
63e95b5c | 164 | key->entry_start != ewait->key.entry_start) |
ac401cc7 JK |
165 | return 0; |
166 | return autoremove_wake_function(wait, mode, sync, NULL); | |
167 | } | |
168 | ||
169 | /* | |
170 | * Check whether the given slot is locked. The function must be called with | |
171 | * mapping->tree_lock held | |
172 | */ | |
173 | static inline int slot_locked(struct address_space *mapping, void **slot) | |
174 | { | |
175 | unsigned long entry = (unsigned long) | |
176 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
177 | return entry & RADIX_DAX_ENTRY_LOCK; | |
178 | } | |
179 | ||
180 | /* | |
181 | * Mark the given slot is locked. The function must be called with | |
182 | * mapping->tree_lock held | |
183 | */ | |
184 | static inline void *lock_slot(struct address_space *mapping, void **slot) | |
185 | { | |
186 | unsigned long entry = (unsigned long) | |
187 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
188 | ||
189 | entry |= RADIX_DAX_ENTRY_LOCK; | |
6d75f366 | 190 | radix_tree_replace_slot(&mapping->page_tree, slot, (void *)entry); |
ac401cc7 JK |
191 | return (void *)entry; |
192 | } | |
193 | ||
194 | /* | |
195 | * Mark the given slot is unlocked. The function must be called with | |
196 | * mapping->tree_lock held | |
197 | */ | |
198 | static inline void *unlock_slot(struct address_space *mapping, void **slot) | |
199 | { | |
200 | unsigned long entry = (unsigned long) | |
201 | radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
202 | ||
203 | entry &= ~(unsigned long)RADIX_DAX_ENTRY_LOCK; | |
6d75f366 | 204 | radix_tree_replace_slot(&mapping->page_tree, slot, (void *)entry); |
ac401cc7 JK |
205 | return (void *)entry; |
206 | } | |
207 | ||
208 | /* | |
209 | * Lookup entry in radix tree, wait for it to become unlocked if it is | |
210 | * exceptional entry and return it. The caller must call | |
211 | * put_unlocked_mapping_entry() when he decided not to lock the entry or | |
212 | * put_locked_mapping_entry() when he locked the entry and now wants to | |
213 | * unlock it. | |
214 | * | |
215 | * The function must be called with mapping->tree_lock held. | |
216 | */ | |
217 | static void *get_unlocked_mapping_entry(struct address_space *mapping, | |
218 | pgoff_t index, void ***slotp) | |
219 | { | |
e3ad61c6 | 220 | void *entry, **slot; |
ac401cc7 | 221 | struct wait_exceptional_entry_queue ewait; |
63e95b5c | 222 | wait_queue_head_t *wq; |
ac401cc7 JK |
223 | |
224 | init_wait(&ewait.wait); | |
225 | ewait.wait.func = wake_exceptional_entry_func; | |
ac401cc7 JK |
226 | |
227 | for (;;) { | |
e3ad61c6 | 228 | entry = __radix_tree_lookup(&mapping->page_tree, index, NULL, |
ac401cc7 | 229 | &slot); |
e3ad61c6 | 230 | if (!entry || !radix_tree_exceptional_entry(entry) || |
ac401cc7 JK |
231 | !slot_locked(mapping, slot)) { |
232 | if (slotp) | |
233 | *slotp = slot; | |
e3ad61c6 | 234 | return entry; |
ac401cc7 | 235 | } |
63e95b5c RZ |
236 | |
237 | wq = dax_entry_waitqueue(mapping, index, entry, &ewait.key); | |
ac401cc7 JK |
238 | prepare_to_wait_exclusive(wq, &ewait.wait, |
239 | TASK_UNINTERRUPTIBLE); | |
240 | spin_unlock_irq(&mapping->tree_lock); | |
241 | schedule(); | |
242 | finish_wait(wq, &ewait.wait); | |
243 | spin_lock_irq(&mapping->tree_lock); | |
244 | } | |
245 | } | |
246 | ||
b1aa812b JK |
247 | static void dax_unlock_mapping_entry(struct address_space *mapping, |
248 | pgoff_t index) | |
249 | { | |
250 | void *entry, **slot; | |
251 | ||
252 | spin_lock_irq(&mapping->tree_lock); | |
253 | entry = __radix_tree_lookup(&mapping->page_tree, index, NULL, &slot); | |
254 | if (WARN_ON_ONCE(!entry || !radix_tree_exceptional_entry(entry) || | |
255 | !slot_locked(mapping, slot))) { | |
256 | spin_unlock_irq(&mapping->tree_lock); | |
257 | return; | |
258 | } | |
259 | unlock_slot(mapping, slot); | |
260 | spin_unlock_irq(&mapping->tree_lock); | |
261 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); | |
262 | } | |
263 | ||
422476c4 RZ |
264 | static void put_locked_mapping_entry(struct address_space *mapping, |
265 | pgoff_t index, void *entry) | |
266 | { | |
267 | if (!radix_tree_exceptional_entry(entry)) { | |
268 | unlock_page(entry); | |
269 | put_page(entry); | |
270 | } else { | |
271 | dax_unlock_mapping_entry(mapping, index); | |
272 | } | |
273 | } | |
274 | ||
275 | /* | |
276 | * Called when we are done with radix tree entry we looked up via | |
277 | * get_unlocked_mapping_entry() and which we didn't lock in the end. | |
278 | */ | |
279 | static void put_unlocked_mapping_entry(struct address_space *mapping, | |
280 | pgoff_t index, void *entry) | |
281 | { | |
282 | if (!radix_tree_exceptional_entry(entry)) | |
283 | return; | |
284 | ||
285 | /* We have to wake up next waiter for the radix tree entry lock */ | |
286 | dax_wake_mapping_entry_waiter(mapping, index, entry, false); | |
287 | } | |
288 | ||
ac401cc7 JK |
289 | /* |
290 | * Find radix tree entry at given index. If it points to a page, return with | |
291 | * the page locked. If it points to the exceptional entry, return with the | |
292 | * radix tree entry locked. If the radix tree doesn't contain given index, | |
293 | * create empty exceptional entry for the index and return with it locked. | |
294 | * | |
642261ac RZ |
295 | * When requesting an entry with size RADIX_DAX_PMD, grab_mapping_entry() will |
296 | * either return that locked entry or will return an error. This error will | |
297 | * happen if there are any 4k entries (either zero pages or DAX entries) | |
298 | * within the 2MiB range that we are requesting. | |
299 | * | |
300 | * We always favor 4k entries over 2MiB entries. There isn't a flow where we | |
301 | * evict 4k entries in order to 'upgrade' them to a 2MiB entry. A 2MiB | |
302 | * insertion will fail if it finds any 4k entries already in the tree, and a | |
303 | * 4k insertion will cause an existing 2MiB entry to be unmapped and | |
304 | * downgraded to 4k entries. This happens for both 2MiB huge zero pages as | |
305 | * well as 2MiB empty entries. | |
306 | * | |
307 | * The exception to this downgrade path is for 2MiB DAX PMD entries that have | |
308 | * real storage backing them. We will leave these real 2MiB DAX entries in | |
309 | * the tree, and PTE writes will simply dirty the entire 2MiB DAX entry. | |
310 | * | |
ac401cc7 JK |
311 | * Note: Unlike filemap_fault() we don't honor FAULT_FLAG_RETRY flags. For |
312 | * persistent memory the benefit is doubtful. We can add that later if we can | |
313 | * show it helps. | |
314 | */ | |
642261ac RZ |
315 | static void *grab_mapping_entry(struct address_space *mapping, pgoff_t index, |
316 | unsigned long size_flag) | |
ac401cc7 | 317 | { |
642261ac | 318 | bool pmd_downgrade = false; /* splitting 2MiB entry into 4k entries? */ |
e3ad61c6 | 319 | void *entry, **slot; |
ac401cc7 JK |
320 | |
321 | restart: | |
322 | spin_lock_irq(&mapping->tree_lock); | |
e3ad61c6 | 323 | entry = get_unlocked_mapping_entry(mapping, index, &slot); |
642261ac RZ |
324 | |
325 | if (entry) { | |
326 | if (size_flag & RADIX_DAX_PMD) { | |
327 | if (!radix_tree_exceptional_entry(entry) || | |
328 | dax_is_pte_entry(entry)) { | |
329 | put_unlocked_mapping_entry(mapping, index, | |
330 | entry); | |
331 | entry = ERR_PTR(-EEXIST); | |
332 | goto out_unlock; | |
333 | } | |
334 | } else { /* trying to grab a PTE entry */ | |
335 | if (radix_tree_exceptional_entry(entry) && | |
336 | dax_is_pmd_entry(entry) && | |
337 | (dax_is_zero_entry(entry) || | |
338 | dax_is_empty_entry(entry))) { | |
339 | pmd_downgrade = true; | |
340 | } | |
341 | } | |
342 | } | |
343 | ||
ac401cc7 | 344 | /* No entry for given index? Make sure radix tree is big enough. */ |
642261ac | 345 | if (!entry || pmd_downgrade) { |
ac401cc7 JK |
346 | int err; |
347 | ||
642261ac RZ |
348 | if (pmd_downgrade) { |
349 | /* | |
350 | * Make sure 'entry' remains valid while we drop | |
351 | * mapping->tree_lock. | |
352 | */ | |
353 | entry = lock_slot(mapping, slot); | |
354 | } | |
355 | ||
ac401cc7 | 356 | spin_unlock_irq(&mapping->tree_lock); |
642261ac RZ |
357 | /* |
358 | * Besides huge zero pages the only other thing that gets | |
359 | * downgraded are empty entries which don't need to be | |
360 | * unmapped. | |
361 | */ | |
362 | if (pmd_downgrade && dax_is_zero_entry(entry)) | |
363 | unmap_mapping_range(mapping, | |
364 | (index << PAGE_SHIFT) & PMD_MASK, PMD_SIZE, 0); | |
365 | ||
ac401cc7 JK |
366 | err = radix_tree_preload( |
367 | mapping_gfp_mask(mapping) & ~__GFP_HIGHMEM); | |
0cb80b48 JK |
368 | if (err) { |
369 | if (pmd_downgrade) | |
370 | put_locked_mapping_entry(mapping, index, entry); | |
ac401cc7 | 371 | return ERR_PTR(err); |
0cb80b48 | 372 | } |
ac401cc7 | 373 | spin_lock_irq(&mapping->tree_lock); |
642261ac RZ |
374 | |
375 | if (pmd_downgrade) { | |
376 | radix_tree_delete(&mapping->page_tree, index); | |
377 | mapping->nrexceptional--; | |
378 | dax_wake_mapping_entry_waiter(mapping, index, entry, | |
379 | true); | |
380 | } | |
381 | ||
382 | entry = dax_radix_locked_entry(0, size_flag | RADIX_DAX_EMPTY); | |
383 | ||
384 | err = __radix_tree_insert(&mapping->page_tree, index, | |
385 | dax_radix_order(entry), entry); | |
ac401cc7 JK |
386 | radix_tree_preload_end(); |
387 | if (err) { | |
388 | spin_unlock_irq(&mapping->tree_lock); | |
642261ac RZ |
389 | /* |
390 | * Someone already created the entry? This is a | |
391 | * normal failure when inserting PMDs in a range | |
392 | * that already contains PTEs. In that case we want | |
393 | * to return -EEXIST immediately. | |
394 | */ | |
395 | if (err == -EEXIST && !(size_flag & RADIX_DAX_PMD)) | |
ac401cc7 | 396 | goto restart; |
642261ac RZ |
397 | /* |
398 | * Our insertion of a DAX PMD entry failed, most | |
399 | * likely because it collided with a PTE sized entry | |
400 | * at a different index in the PMD range. We haven't | |
401 | * inserted anything into the radix tree and have no | |
402 | * waiters to wake. | |
403 | */ | |
ac401cc7 JK |
404 | return ERR_PTR(err); |
405 | } | |
406 | /* Good, we have inserted empty locked entry into the tree. */ | |
407 | mapping->nrexceptional++; | |
408 | spin_unlock_irq(&mapping->tree_lock); | |
e3ad61c6 | 409 | return entry; |
ac401cc7 JK |
410 | } |
411 | /* Normal page in radix tree? */ | |
e3ad61c6 RZ |
412 | if (!radix_tree_exceptional_entry(entry)) { |
413 | struct page *page = entry; | |
ac401cc7 JK |
414 | |
415 | get_page(page); | |
416 | spin_unlock_irq(&mapping->tree_lock); | |
417 | lock_page(page); | |
418 | /* Page got truncated? Retry... */ | |
419 | if (unlikely(page->mapping != mapping)) { | |
420 | unlock_page(page); | |
421 | put_page(page); | |
422 | goto restart; | |
423 | } | |
424 | return page; | |
425 | } | |
e3ad61c6 | 426 | entry = lock_slot(mapping, slot); |
642261ac | 427 | out_unlock: |
ac401cc7 | 428 | spin_unlock_irq(&mapping->tree_lock); |
e3ad61c6 | 429 | return entry; |
ac401cc7 JK |
430 | } |
431 | ||
63e95b5c RZ |
432 | /* |
433 | * We do not necessarily hold the mapping->tree_lock when we call this | |
434 | * function so it is possible that 'entry' is no longer a valid item in the | |
642261ac RZ |
435 | * radix tree. This is okay because all we really need to do is to find the |
436 | * correct waitqueue where tasks might be waiting for that old 'entry' and | |
437 | * wake them. | |
63e95b5c | 438 | */ |
ac401cc7 | 439 | void dax_wake_mapping_entry_waiter(struct address_space *mapping, |
63e95b5c | 440 | pgoff_t index, void *entry, bool wake_all) |
ac401cc7 | 441 | { |
63e95b5c RZ |
442 | struct exceptional_entry_key key; |
443 | wait_queue_head_t *wq; | |
444 | ||
445 | wq = dax_entry_waitqueue(mapping, index, entry, &key); | |
ac401cc7 JK |
446 | |
447 | /* | |
448 | * Checking for locked entry and prepare_to_wait_exclusive() happens | |
449 | * under mapping->tree_lock, ditto for entry handling in our callers. | |
450 | * So at this point all tasks that could have seen our entry locked | |
451 | * must be in the waitqueue and the following check will see them. | |
452 | */ | |
63e95b5c | 453 | if (waitqueue_active(wq)) |
ac401cc7 | 454 | __wake_up(wq, TASK_NORMAL, wake_all ? 0 : 1, &key); |
ac401cc7 JK |
455 | } |
456 | ||
c6dcf52c JK |
457 | static int __dax_invalidate_mapping_entry(struct address_space *mapping, |
458 | pgoff_t index, bool trunc) | |
459 | { | |
460 | int ret = 0; | |
461 | void *entry; | |
462 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
463 | ||
464 | spin_lock_irq(&mapping->tree_lock); | |
465 | entry = get_unlocked_mapping_entry(mapping, index, NULL); | |
466 | if (!entry || !radix_tree_exceptional_entry(entry)) | |
467 | goto out; | |
468 | if (!trunc && | |
469 | (radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_DIRTY) || | |
470 | radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE))) | |
471 | goto out; | |
472 | radix_tree_delete(page_tree, index); | |
473 | mapping->nrexceptional--; | |
474 | ret = 1; | |
475 | out: | |
476 | put_unlocked_mapping_entry(mapping, index, entry); | |
477 | spin_unlock_irq(&mapping->tree_lock); | |
478 | return ret; | |
479 | } | |
ac401cc7 JK |
480 | /* |
481 | * Delete exceptional DAX entry at @index from @mapping. Wait for radix tree | |
482 | * entry to get unlocked before deleting it. | |
483 | */ | |
484 | int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index) | |
485 | { | |
c6dcf52c | 486 | int ret = __dax_invalidate_mapping_entry(mapping, index, true); |
ac401cc7 | 487 | |
ac401cc7 JK |
488 | /* |
489 | * This gets called from truncate / punch_hole path. As such, the caller | |
490 | * must hold locks protecting against concurrent modifications of the | |
491 | * radix tree (usually fs-private i_mmap_sem for writing). Since the | |
492 | * caller has seen exceptional entry for this index, we better find it | |
493 | * at that index as well... | |
494 | */ | |
c6dcf52c JK |
495 | WARN_ON_ONCE(!ret); |
496 | return ret; | |
497 | } | |
498 | ||
499 | /* | |
500 | * Invalidate exceptional DAX entry if easily possible. This handles DAX | |
501 | * entries for invalidate_inode_pages() so we evict the entry only if we can | |
502 | * do so without blocking. | |
503 | */ | |
504 | int dax_invalidate_mapping_entry(struct address_space *mapping, pgoff_t index) | |
505 | { | |
506 | int ret = 0; | |
507 | void *entry, **slot; | |
508 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
509 | ||
510 | spin_lock_irq(&mapping->tree_lock); | |
511 | entry = __radix_tree_lookup(page_tree, index, NULL, &slot); | |
512 | if (!entry || !radix_tree_exceptional_entry(entry) || | |
513 | slot_locked(mapping, slot)) | |
514 | goto out; | |
515 | if (radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_DIRTY) || | |
516 | radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE)) | |
517 | goto out; | |
518 | radix_tree_delete(page_tree, index); | |
ac401cc7 | 519 | mapping->nrexceptional--; |
c6dcf52c JK |
520 | ret = 1; |
521 | out: | |
ac401cc7 | 522 | spin_unlock_irq(&mapping->tree_lock); |
c6dcf52c JK |
523 | if (ret) |
524 | dax_wake_mapping_entry_waiter(mapping, index, entry, true); | |
525 | return ret; | |
526 | } | |
ac401cc7 | 527 | |
c6dcf52c JK |
528 | /* |
529 | * Invalidate exceptional DAX entry if it is clean. | |
530 | */ | |
531 | int dax_invalidate_mapping_entry_sync(struct address_space *mapping, | |
532 | pgoff_t index) | |
533 | { | |
534 | return __dax_invalidate_mapping_entry(mapping, index, false); | |
ac401cc7 JK |
535 | } |
536 | ||
f7ca90b1 MW |
537 | /* |
538 | * The user has performed a load from a hole in the file. Allocating | |
539 | * a new page in the file would cause excessive storage usage for | |
540 | * workloads with sparse files. We allocate a page cache page instead. | |
541 | * We'll kick it out of the page cache if it's ever written to, | |
542 | * otherwise it will simply fall out of the page cache under memory | |
543 | * pressure without ever having been dirtied. | |
544 | */ | |
f449b936 | 545 | static int dax_load_hole(struct address_space *mapping, void **entry, |
ac401cc7 | 546 | struct vm_fault *vmf) |
f7ca90b1 | 547 | { |
ac401cc7 | 548 | struct page *page; |
f449b936 | 549 | int ret; |
f7ca90b1 | 550 | |
ac401cc7 | 551 | /* Hole page already exists? Return it... */ |
f449b936 JK |
552 | if (!radix_tree_exceptional_entry(*entry)) { |
553 | page = *entry; | |
554 | goto out; | |
ac401cc7 | 555 | } |
f7ca90b1 | 556 | |
ac401cc7 JK |
557 | /* This will replace locked radix tree entry with a hole page */ |
558 | page = find_or_create_page(mapping, vmf->pgoff, | |
559 | vmf->gfp_mask | __GFP_ZERO); | |
b1aa812b | 560 | if (!page) |
ac401cc7 | 561 | return VM_FAULT_OOM; |
f449b936 | 562 | out: |
f7ca90b1 | 563 | vmf->page = page; |
f449b936 JK |
564 | ret = finish_fault(vmf); |
565 | vmf->page = NULL; | |
566 | *entry = page; | |
567 | if (!ret) { | |
568 | /* Grab reference for PTE that is now referencing the page */ | |
569 | get_page(page); | |
570 | return VM_FAULT_NOPAGE; | |
571 | } | |
572 | return ret; | |
f7ca90b1 MW |
573 | } |
574 | ||
b0d5e82f CH |
575 | static int copy_user_dax(struct block_device *bdev, sector_t sector, size_t size, |
576 | struct page *to, unsigned long vaddr) | |
f7ca90b1 | 577 | { |
b2e0d162 | 578 | struct blk_dax_ctl dax = { |
b0d5e82f CH |
579 | .sector = sector, |
580 | .size = size, | |
b2e0d162 | 581 | }; |
e2e05394 RZ |
582 | void *vto; |
583 | ||
b2e0d162 DW |
584 | if (dax_map_atomic(bdev, &dax) < 0) |
585 | return PTR_ERR(dax.addr); | |
f7ca90b1 | 586 | vto = kmap_atomic(to); |
b2e0d162 | 587 | copy_user_page(vto, (void __force *)dax.addr, vaddr, to); |
f7ca90b1 | 588 | kunmap_atomic(vto); |
b2e0d162 | 589 | dax_unmap_atomic(bdev, &dax); |
f7ca90b1 MW |
590 | return 0; |
591 | } | |
592 | ||
642261ac RZ |
593 | /* |
594 | * By this point grab_mapping_entry() has ensured that we have a locked entry | |
595 | * of the appropriate size so we don't have to worry about downgrading PMDs to | |
596 | * PTEs. If we happen to be trying to insert a PTE and there is a PMD | |
597 | * already in the tree, we will skip the insertion and just dirty the PMD as | |
598 | * appropriate. | |
599 | */ | |
ac401cc7 JK |
600 | static void *dax_insert_mapping_entry(struct address_space *mapping, |
601 | struct vm_fault *vmf, | |
642261ac RZ |
602 | void *entry, sector_t sector, |
603 | unsigned long flags) | |
9973c98e RZ |
604 | { |
605 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
ac401cc7 JK |
606 | int error = 0; |
607 | bool hole_fill = false; | |
608 | void *new_entry; | |
609 | pgoff_t index = vmf->pgoff; | |
9973c98e | 610 | |
ac401cc7 | 611 | if (vmf->flags & FAULT_FLAG_WRITE) |
d2b2a28e | 612 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
9973c98e | 613 | |
ac401cc7 JK |
614 | /* Replacing hole page with block mapping? */ |
615 | if (!radix_tree_exceptional_entry(entry)) { | |
616 | hole_fill = true; | |
617 | /* | |
618 | * Unmap the page now before we remove it from page cache below. | |
619 | * The page is locked so it cannot be faulted in again. | |
620 | */ | |
621 | unmap_mapping_range(mapping, vmf->pgoff << PAGE_SHIFT, | |
622 | PAGE_SIZE, 0); | |
623 | error = radix_tree_preload(vmf->gfp_mask & ~__GFP_HIGHMEM); | |
624 | if (error) | |
625 | return ERR_PTR(error); | |
642261ac RZ |
626 | } else if (dax_is_zero_entry(entry) && !(flags & RADIX_DAX_HZP)) { |
627 | /* replacing huge zero page with PMD block mapping */ | |
628 | unmap_mapping_range(mapping, | |
629 | (vmf->pgoff << PAGE_SHIFT) & PMD_MASK, PMD_SIZE, 0); | |
9973c98e RZ |
630 | } |
631 | ||
ac401cc7 | 632 | spin_lock_irq(&mapping->tree_lock); |
642261ac RZ |
633 | new_entry = dax_radix_locked_entry(sector, flags); |
634 | ||
ac401cc7 JK |
635 | if (hole_fill) { |
636 | __delete_from_page_cache(entry, NULL); | |
637 | /* Drop pagecache reference */ | |
638 | put_page(entry); | |
642261ac RZ |
639 | error = __radix_tree_insert(page_tree, index, |
640 | dax_radix_order(new_entry), new_entry); | |
ac401cc7 JK |
641 | if (error) { |
642 | new_entry = ERR_PTR(error); | |
9973c98e RZ |
643 | goto unlock; |
644 | } | |
ac401cc7 | 645 | mapping->nrexceptional++; |
642261ac RZ |
646 | } else if (dax_is_zero_entry(entry) || dax_is_empty_entry(entry)) { |
647 | /* | |
648 | * Only swap our new entry into the radix tree if the current | |
649 | * entry is a zero page or an empty entry. If a normal PTE or | |
650 | * PMD entry is already in the tree, we leave it alone. This | |
651 | * means that if we are trying to insert a PTE and the | |
652 | * existing entry is a PMD, we will just leave the PMD in the | |
653 | * tree and dirty it if necessary. | |
654 | */ | |
f7942430 | 655 | struct radix_tree_node *node; |
ac401cc7 JK |
656 | void **slot; |
657 | void *ret; | |
9973c98e | 658 | |
f7942430 | 659 | ret = __radix_tree_lookup(page_tree, index, &node, &slot); |
ac401cc7 | 660 | WARN_ON_ONCE(ret != entry); |
4d693d08 JW |
661 | __radix_tree_replace(page_tree, node, slot, |
662 | new_entry, NULL, NULL); | |
9973c98e | 663 | } |
ac401cc7 | 664 | if (vmf->flags & FAULT_FLAG_WRITE) |
9973c98e RZ |
665 | radix_tree_tag_set(page_tree, index, PAGECACHE_TAG_DIRTY); |
666 | unlock: | |
667 | spin_unlock_irq(&mapping->tree_lock); | |
ac401cc7 JK |
668 | if (hole_fill) { |
669 | radix_tree_preload_end(); | |
670 | /* | |
671 | * We don't need hole page anymore, it has been replaced with | |
672 | * locked radix tree entry now. | |
673 | */ | |
674 | if (mapping->a_ops->freepage) | |
675 | mapping->a_ops->freepage(entry); | |
676 | unlock_page(entry); | |
677 | put_page(entry); | |
678 | } | |
679 | return new_entry; | |
9973c98e RZ |
680 | } |
681 | ||
4b4bb46d JK |
682 | static inline unsigned long |
683 | pgoff_address(pgoff_t pgoff, struct vm_area_struct *vma) | |
684 | { | |
685 | unsigned long address; | |
686 | ||
687 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
688 | VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); | |
689 | return address; | |
690 | } | |
691 | ||
692 | /* Walk all mappings of a given index of a file and writeprotect them */ | |
693 | static void dax_mapping_entry_mkclean(struct address_space *mapping, | |
694 | pgoff_t index, unsigned long pfn) | |
695 | { | |
696 | struct vm_area_struct *vma; | |
f729c8c9 RZ |
697 | pte_t pte, *ptep = NULL; |
698 | pmd_t *pmdp = NULL; | |
4b4bb46d JK |
699 | spinlock_t *ptl; |
700 | bool changed; | |
701 | ||
702 | i_mmap_lock_read(mapping); | |
703 | vma_interval_tree_foreach(vma, &mapping->i_mmap, index, index) { | |
704 | unsigned long address; | |
705 | ||
706 | cond_resched(); | |
707 | ||
708 | if (!(vma->vm_flags & VM_SHARED)) | |
709 | continue; | |
710 | ||
711 | address = pgoff_address(index, vma); | |
712 | changed = false; | |
f729c8c9 | 713 | if (follow_pte_pmd(vma->vm_mm, address, &ptep, &pmdp, &ptl)) |
4b4bb46d | 714 | continue; |
4b4bb46d | 715 | |
f729c8c9 RZ |
716 | if (pmdp) { |
717 | #ifdef CONFIG_FS_DAX_PMD | |
718 | pmd_t pmd; | |
719 | ||
720 | if (pfn != pmd_pfn(*pmdp)) | |
721 | goto unlock_pmd; | |
722 | if (!pmd_dirty(*pmdp) && !pmd_write(*pmdp)) | |
723 | goto unlock_pmd; | |
724 | ||
725 | flush_cache_page(vma, address, pfn); | |
726 | pmd = pmdp_huge_clear_flush(vma, address, pmdp); | |
727 | pmd = pmd_wrprotect(pmd); | |
728 | pmd = pmd_mkclean(pmd); | |
729 | set_pmd_at(vma->vm_mm, address, pmdp, pmd); | |
730 | changed = true; | |
731 | unlock_pmd: | |
732 | spin_unlock(ptl); | |
733 | #endif | |
734 | } else { | |
735 | if (pfn != pte_pfn(*ptep)) | |
736 | goto unlock_pte; | |
737 | if (!pte_dirty(*ptep) && !pte_write(*ptep)) | |
738 | goto unlock_pte; | |
739 | ||
740 | flush_cache_page(vma, address, pfn); | |
741 | pte = ptep_clear_flush(vma, address, ptep); | |
742 | pte = pte_wrprotect(pte); | |
743 | pte = pte_mkclean(pte); | |
744 | set_pte_at(vma->vm_mm, address, ptep, pte); | |
745 | changed = true; | |
746 | unlock_pte: | |
747 | pte_unmap_unlock(ptep, ptl); | |
748 | } | |
4b4bb46d JK |
749 | |
750 | if (changed) | |
751 | mmu_notifier_invalidate_page(vma->vm_mm, address); | |
752 | } | |
753 | i_mmap_unlock_read(mapping); | |
754 | } | |
755 | ||
9973c98e RZ |
756 | static int dax_writeback_one(struct block_device *bdev, |
757 | struct address_space *mapping, pgoff_t index, void *entry) | |
758 | { | |
759 | struct radix_tree_root *page_tree = &mapping->page_tree; | |
9973c98e | 760 | struct blk_dax_ctl dax; |
a6abc2c0 | 761 | void *entry2, **slot; |
9973c98e RZ |
762 | int ret = 0; |
763 | ||
9973c98e | 764 | /* |
a6abc2c0 JK |
765 | * A page got tagged dirty in DAX mapping? Something is seriously |
766 | * wrong. | |
9973c98e | 767 | */ |
a6abc2c0 JK |
768 | if (WARN_ON(!radix_tree_exceptional_entry(entry))) |
769 | return -EIO; | |
9973c98e | 770 | |
a6abc2c0 JK |
771 | spin_lock_irq(&mapping->tree_lock); |
772 | entry2 = get_unlocked_mapping_entry(mapping, index, &slot); | |
773 | /* Entry got punched out / reallocated? */ | |
774 | if (!entry2 || !radix_tree_exceptional_entry(entry2)) | |
775 | goto put_unlocked; | |
776 | /* | |
777 | * Entry got reallocated elsewhere? No need to writeback. We have to | |
778 | * compare sectors as we must not bail out due to difference in lockbit | |
779 | * or entry type. | |
780 | */ | |
781 | if (dax_radix_sector(entry2) != dax_radix_sector(entry)) | |
782 | goto put_unlocked; | |
642261ac RZ |
783 | if (WARN_ON_ONCE(dax_is_empty_entry(entry) || |
784 | dax_is_zero_entry(entry))) { | |
9973c98e | 785 | ret = -EIO; |
a6abc2c0 | 786 | goto put_unlocked; |
9973c98e RZ |
787 | } |
788 | ||
a6abc2c0 JK |
789 | /* Another fsync thread may have already written back this entry */ |
790 | if (!radix_tree_tag_get(page_tree, index, PAGECACHE_TAG_TOWRITE)) | |
791 | goto put_unlocked; | |
792 | /* Lock the entry to serialize with page faults */ | |
793 | entry = lock_slot(mapping, slot); | |
794 | /* | |
795 | * We can clear the tag now but we have to be careful so that concurrent | |
796 | * dax_writeback_one() calls for the same index cannot finish before we | |
797 | * actually flush the caches. This is achieved as the calls will look | |
798 | * at the entry only under tree_lock and once they do that they will | |
799 | * see the entry locked and wait for it to unlock. | |
800 | */ | |
801 | radix_tree_tag_clear(page_tree, index, PAGECACHE_TAG_TOWRITE); | |
802 | spin_unlock_irq(&mapping->tree_lock); | |
803 | ||
642261ac RZ |
804 | /* |
805 | * Even if dax_writeback_mapping_range() was given a wbc->range_start | |
806 | * in the middle of a PMD, the 'index' we are given will be aligned to | |
807 | * the start index of the PMD, as will the sector we pull from | |
808 | * 'entry'. This allows us to flush for PMD_SIZE and not have to | |
809 | * worry about partial PMD writebacks. | |
810 | */ | |
811 | dax.sector = dax_radix_sector(entry); | |
812 | dax.size = PAGE_SIZE << dax_radix_order(entry); | |
9973c98e RZ |
813 | |
814 | /* | |
815 | * We cannot hold tree_lock while calling dax_map_atomic() because it | |
816 | * eventually calls cond_resched(). | |
817 | */ | |
818 | ret = dax_map_atomic(bdev, &dax); | |
a6abc2c0 JK |
819 | if (ret < 0) { |
820 | put_locked_mapping_entry(mapping, index, entry); | |
9973c98e | 821 | return ret; |
a6abc2c0 | 822 | } |
9973c98e RZ |
823 | |
824 | if (WARN_ON_ONCE(ret < dax.size)) { | |
825 | ret = -EIO; | |
826 | goto unmap; | |
827 | } | |
828 | ||
4b4bb46d | 829 | dax_mapping_entry_mkclean(mapping, index, pfn_t_to_pfn(dax.pfn)); |
9973c98e | 830 | wb_cache_pmem(dax.addr, dax.size); |
4b4bb46d JK |
831 | /* |
832 | * After we have flushed the cache, we can clear the dirty tag. There | |
833 | * cannot be new dirty data in the pfn after the flush has completed as | |
834 | * the pfn mappings are writeprotected and fault waits for mapping | |
835 | * entry lock. | |
836 | */ | |
837 | spin_lock_irq(&mapping->tree_lock); | |
838 | radix_tree_tag_clear(page_tree, index, PAGECACHE_TAG_DIRTY); | |
839 | spin_unlock_irq(&mapping->tree_lock); | |
9973c98e RZ |
840 | unmap: |
841 | dax_unmap_atomic(bdev, &dax); | |
a6abc2c0 | 842 | put_locked_mapping_entry(mapping, index, entry); |
9973c98e RZ |
843 | return ret; |
844 | ||
a6abc2c0 JK |
845 | put_unlocked: |
846 | put_unlocked_mapping_entry(mapping, index, entry2); | |
9973c98e RZ |
847 | spin_unlock_irq(&mapping->tree_lock); |
848 | return ret; | |
849 | } | |
850 | ||
851 | /* | |
852 | * Flush the mapping to the persistent domain within the byte range of [start, | |
853 | * end]. This is required by data integrity operations to ensure file data is | |
854 | * on persistent storage prior to completion of the operation. | |
855 | */ | |
7f6d5b52 RZ |
856 | int dax_writeback_mapping_range(struct address_space *mapping, |
857 | struct block_device *bdev, struct writeback_control *wbc) | |
9973c98e RZ |
858 | { |
859 | struct inode *inode = mapping->host; | |
642261ac | 860 | pgoff_t start_index, end_index; |
9973c98e RZ |
861 | pgoff_t indices[PAGEVEC_SIZE]; |
862 | struct pagevec pvec; | |
863 | bool done = false; | |
864 | int i, ret = 0; | |
9973c98e RZ |
865 | |
866 | if (WARN_ON_ONCE(inode->i_blkbits != PAGE_SHIFT)) | |
867 | return -EIO; | |
868 | ||
7f6d5b52 RZ |
869 | if (!mapping->nrexceptional || wbc->sync_mode != WB_SYNC_ALL) |
870 | return 0; | |
871 | ||
09cbfeaf KS |
872 | start_index = wbc->range_start >> PAGE_SHIFT; |
873 | end_index = wbc->range_end >> PAGE_SHIFT; | |
9973c98e RZ |
874 | |
875 | tag_pages_for_writeback(mapping, start_index, end_index); | |
876 | ||
877 | pagevec_init(&pvec, 0); | |
878 | while (!done) { | |
879 | pvec.nr = find_get_entries_tag(mapping, start_index, | |
880 | PAGECACHE_TAG_TOWRITE, PAGEVEC_SIZE, | |
881 | pvec.pages, indices); | |
882 | ||
883 | if (pvec.nr == 0) | |
884 | break; | |
885 | ||
886 | for (i = 0; i < pvec.nr; i++) { | |
887 | if (indices[i] > end_index) { | |
888 | done = true; | |
889 | break; | |
890 | } | |
891 | ||
892 | ret = dax_writeback_one(bdev, mapping, indices[i], | |
893 | pvec.pages[i]); | |
894 | if (ret < 0) | |
895 | return ret; | |
896 | } | |
897 | } | |
9973c98e RZ |
898 | return 0; |
899 | } | |
900 | EXPORT_SYMBOL_GPL(dax_writeback_mapping_range); | |
901 | ||
ac401cc7 | 902 | static int dax_insert_mapping(struct address_space *mapping, |
1aaba095 CH |
903 | struct block_device *bdev, sector_t sector, size_t size, |
904 | void **entryp, struct vm_area_struct *vma, struct vm_fault *vmf) | |
f7ca90b1 | 905 | { |
1a29d85e | 906 | unsigned long vaddr = vmf->address; |
b2e0d162 | 907 | struct blk_dax_ctl dax = { |
1aaba095 CH |
908 | .sector = sector, |
909 | .size = size, | |
b2e0d162 | 910 | }; |
ac401cc7 JK |
911 | void *ret; |
912 | void *entry = *entryp; | |
f7ca90b1 | 913 | |
4d9a2c87 JK |
914 | if (dax_map_atomic(bdev, &dax) < 0) |
915 | return PTR_ERR(dax.addr); | |
b2e0d162 | 916 | dax_unmap_atomic(bdev, &dax); |
f7ca90b1 | 917 | |
642261ac | 918 | ret = dax_insert_mapping_entry(mapping, vmf, entry, dax.sector, 0); |
4d9a2c87 JK |
919 | if (IS_ERR(ret)) |
920 | return PTR_ERR(ret); | |
ac401cc7 | 921 | *entryp = ret; |
9973c98e | 922 | |
4d9a2c87 | 923 | return vm_insert_mixed(vma, vaddr, dax.pfn); |
f7ca90b1 MW |
924 | } |
925 | ||
0e3b210c BH |
926 | /** |
927 | * dax_pfn_mkwrite - handle first write to DAX page | |
928 | * @vma: The virtual memory area where the fault occurred | |
929 | * @vmf: The description of the fault | |
0e3b210c BH |
930 | */ |
931 | int dax_pfn_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) | |
932 | { | |
9973c98e | 933 | struct file *file = vma->vm_file; |
ac401cc7 | 934 | struct address_space *mapping = file->f_mapping; |
2f89dc12 | 935 | void *entry, **slot; |
ac401cc7 | 936 | pgoff_t index = vmf->pgoff; |
30f471fd | 937 | |
ac401cc7 | 938 | spin_lock_irq(&mapping->tree_lock); |
2f89dc12 JK |
939 | entry = get_unlocked_mapping_entry(mapping, index, &slot); |
940 | if (!entry || !radix_tree_exceptional_entry(entry)) { | |
941 | if (entry) | |
942 | put_unlocked_mapping_entry(mapping, index, entry); | |
943 | spin_unlock_irq(&mapping->tree_lock); | |
944 | return VM_FAULT_NOPAGE; | |
945 | } | |
ac401cc7 | 946 | radix_tree_tag_set(&mapping->page_tree, index, PAGECACHE_TAG_DIRTY); |
2f89dc12 | 947 | entry = lock_slot(mapping, slot); |
ac401cc7 | 948 | spin_unlock_irq(&mapping->tree_lock); |
2f89dc12 JK |
949 | /* |
950 | * If we race with somebody updating the PTE and finish_mkwrite_fault() | |
951 | * fails, we don't care. We need to return VM_FAULT_NOPAGE and retry | |
952 | * the fault in either case. | |
953 | */ | |
954 | finish_mkwrite_fault(vmf); | |
955 | put_locked_mapping_entry(mapping, index, entry); | |
0e3b210c BH |
956 | return VM_FAULT_NOPAGE; |
957 | } | |
958 | EXPORT_SYMBOL_GPL(dax_pfn_mkwrite); | |
959 | ||
4b0228fa VV |
960 | static bool dax_range_is_aligned(struct block_device *bdev, |
961 | unsigned int offset, unsigned int length) | |
962 | { | |
963 | unsigned short sector_size = bdev_logical_block_size(bdev); | |
964 | ||
965 | if (!IS_ALIGNED(offset, sector_size)) | |
966 | return false; | |
967 | if (!IS_ALIGNED(length, sector_size)) | |
968 | return false; | |
969 | ||
970 | return true; | |
971 | } | |
972 | ||
679c8bd3 CH |
973 | int __dax_zero_page_range(struct block_device *bdev, sector_t sector, |
974 | unsigned int offset, unsigned int length) | |
975 | { | |
976 | struct blk_dax_ctl dax = { | |
977 | .sector = sector, | |
978 | .size = PAGE_SIZE, | |
979 | }; | |
980 | ||
4b0228fa VV |
981 | if (dax_range_is_aligned(bdev, offset, length)) { |
982 | sector_t start_sector = dax.sector + (offset >> 9); | |
983 | ||
984 | return blkdev_issue_zeroout(bdev, start_sector, | |
985 | length >> 9, GFP_NOFS, true); | |
986 | } else { | |
987 | if (dax_map_atomic(bdev, &dax) < 0) | |
988 | return PTR_ERR(dax.addr); | |
989 | clear_pmem(dax.addr + offset, length); | |
4b0228fa VV |
990 | dax_unmap_atomic(bdev, &dax); |
991 | } | |
679c8bd3 CH |
992 | return 0; |
993 | } | |
994 | EXPORT_SYMBOL_GPL(__dax_zero_page_range); | |
995 | ||
333ccc97 | 996 | static sector_t dax_iomap_sector(struct iomap *iomap, loff_t pos) |
25726bc1 | 997 | { |
333ccc97 | 998 | return iomap->blkno + (((pos & PAGE_MASK) - iomap->offset) >> 9); |
25726bc1 | 999 | } |
a254e568 | 1000 | |
a254e568 | 1001 | static loff_t |
11c59c92 | 1002 | dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data, |
a254e568 CH |
1003 | struct iomap *iomap) |
1004 | { | |
1005 | struct iov_iter *iter = data; | |
1006 | loff_t end = pos + length, done = 0; | |
1007 | ssize_t ret = 0; | |
1008 | ||
1009 | if (iov_iter_rw(iter) == READ) { | |
1010 | end = min(end, i_size_read(inode)); | |
1011 | if (pos >= end) | |
1012 | return 0; | |
1013 | ||
1014 | if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN) | |
1015 | return iov_iter_zero(min(length, end - pos), iter); | |
1016 | } | |
1017 | ||
1018 | if (WARN_ON_ONCE(iomap->type != IOMAP_MAPPED)) | |
1019 | return -EIO; | |
1020 | ||
e3fce68c JK |
1021 | /* |
1022 | * Write can allocate block for an area which has a hole page mapped | |
1023 | * into page tables. We have to tear down these mappings so that data | |
1024 | * written by write(2) is visible in mmap. | |
1025 | */ | |
1026 | if ((iomap->flags & IOMAP_F_NEW) && inode->i_mapping->nrpages) { | |
1027 | invalidate_inode_pages2_range(inode->i_mapping, | |
1028 | pos >> PAGE_SHIFT, | |
1029 | (end - 1) >> PAGE_SHIFT); | |
1030 | } | |
1031 | ||
a254e568 CH |
1032 | while (pos < end) { |
1033 | unsigned offset = pos & (PAGE_SIZE - 1); | |
1034 | struct blk_dax_ctl dax = { 0 }; | |
1035 | ssize_t map_len; | |
1036 | ||
d1908f52 MH |
1037 | if (fatal_signal_pending(current)) { |
1038 | ret = -EINTR; | |
1039 | break; | |
1040 | } | |
1041 | ||
333ccc97 | 1042 | dax.sector = dax_iomap_sector(iomap, pos); |
a254e568 CH |
1043 | dax.size = (length + offset + PAGE_SIZE - 1) & PAGE_MASK; |
1044 | map_len = dax_map_atomic(iomap->bdev, &dax); | |
1045 | if (map_len < 0) { | |
1046 | ret = map_len; | |
1047 | break; | |
1048 | } | |
1049 | ||
1050 | dax.addr += offset; | |
1051 | map_len -= offset; | |
1052 | if (map_len > end - pos) | |
1053 | map_len = end - pos; | |
1054 | ||
1055 | if (iov_iter_rw(iter) == WRITE) | |
1056 | map_len = copy_from_iter_pmem(dax.addr, map_len, iter); | |
1057 | else | |
1058 | map_len = copy_to_iter(dax.addr, map_len, iter); | |
1059 | dax_unmap_atomic(iomap->bdev, &dax); | |
1060 | if (map_len <= 0) { | |
1061 | ret = map_len ? map_len : -EFAULT; | |
1062 | break; | |
1063 | } | |
1064 | ||
1065 | pos += map_len; | |
1066 | length -= map_len; | |
1067 | done += map_len; | |
1068 | } | |
1069 | ||
1070 | return done ? done : ret; | |
1071 | } | |
1072 | ||
1073 | /** | |
11c59c92 | 1074 | * dax_iomap_rw - Perform I/O to a DAX file |
a254e568 CH |
1075 | * @iocb: The control block for this I/O |
1076 | * @iter: The addresses to do I/O from or to | |
1077 | * @ops: iomap ops passed from the file system | |
1078 | * | |
1079 | * This function performs read and write operations to directly mapped | |
1080 | * persistent memory. The callers needs to take care of read/write exclusion | |
1081 | * and evicting any page cache pages in the region under I/O. | |
1082 | */ | |
1083 | ssize_t | |
11c59c92 | 1084 | dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter, |
a254e568 CH |
1085 | struct iomap_ops *ops) |
1086 | { | |
1087 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
1088 | struct inode *inode = mapping->host; | |
1089 | loff_t pos = iocb->ki_pos, ret = 0, done = 0; | |
1090 | unsigned flags = 0; | |
1091 | ||
168316db CH |
1092 | if (iov_iter_rw(iter) == WRITE) { |
1093 | lockdep_assert_held_exclusive(&inode->i_rwsem); | |
a254e568 | 1094 | flags |= IOMAP_WRITE; |
168316db CH |
1095 | } else { |
1096 | lockdep_assert_held(&inode->i_rwsem); | |
1097 | } | |
a254e568 | 1098 | |
a254e568 CH |
1099 | while (iov_iter_count(iter)) { |
1100 | ret = iomap_apply(inode, pos, iov_iter_count(iter), flags, ops, | |
11c59c92 | 1101 | iter, dax_iomap_actor); |
a254e568 CH |
1102 | if (ret <= 0) |
1103 | break; | |
1104 | pos += ret; | |
1105 | done += ret; | |
1106 | } | |
1107 | ||
1108 | iocb->ki_pos += done; | |
1109 | return done ? done : ret; | |
1110 | } | |
11c59c92 | 1111 | EXPORT_SYMBOL_GPL(dax_iomap_rw); |
a7d73fe6 | 1112 | |
9f141d6e JK |
1113 | static int dax_fault_return(int error) |
1114 | { | |
1115 | if (error == 0) | |
1116 | return VM_FAULT_NOPAGE; | |
1117 | if (error == -ENOMEM) | |
1118 | return VM_FAULT_OOM; | |
1119 | return VM_FAULT_SIGBUS; | |
1120 | } | |
1121 | ||
a7d73fe6 | 1122 | /** |
11c59c92 | 1123 | * dax_iomap_fault - handle a page fault on a DAX file |
a7d73fe6 CH |
1124 | * @vma: The virtual memory area where the fault occurred |
1125 | * @vmf: The description of the fault | |
1126 | * @ops: iomap ops passed from the file system | |
1127 | * | |
1128 | * When a page fault occurs, filesystems may call this helper in their fault | |
1129 | * or mkwrite handler for DAX files. Assumes the caller has done all the | |
1130 | * necessary locking for the page fault to proceed successfully. | |
1131 | */ | |
11c59c92 | 1132 | int dax_iomap_fault(struct vm_area_struct *vma, struct vm_fault *vmf, |
a7d73fe6 CH |
1133 | struct iomap_ops *ops) |
1134 | { | |
1135 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1136 | struct inode *inode = mapping->host; | |
1a29d85e | 1137 | unsigned long vaddr = vmf->address; |
a7d73fe6 CH |
1138 | loff_t pos = (loff_t)vmf->pgoff << PAGE_SHIFT; |
1139 | sector_t sector; | |
1140 | struct iomap iomap = { 0 }; | |
9484ab1b | 1141 | unsigned flags = IOMAP_FAULT; |
a7d73fe6 | 1142 | int error, major = 0; |
b1aa812b | 1143 | int vmf_ret = 0; |
a7d73fe6 CH |
1144 | void *entry; |
1145 | ||
1146 | /* | |
1147 | * Check whether offset isn't beyond end of file now. Caller is supposed | |
1148 | * to hold locks serializing us with truncate / punch hole so this is | |
1149 | * a reliable test. | |
1150 | */ | |
1151 | if (pos >= i_size_read(inode)) | |
1152 | return VM_FAULT_SIGBUS; | |
1153 | ||
a7d73fe6 CH |
1154 | if ((vmf->flags & FAULT_FLAG_WRITE) && !vmf->cow_page) |
1155 | flags |= IOMAP_WRITE; | |
1156 | ||
1157 | /* | |
1158 | * Note that we don't bother to use iomap_apply here: DAX required | |
1159 | * the file system block size to be equal the page size, which means | |
1160 | * that we never have to deal with more than a single extent here. | |
1161 | */ | |
1162 | error = ops->iomap_begin(inode, pos, PAGE_SIZE, flags, &iomap); | |
1163 | if (error) | |
9f141d6e | 1164 | return dax_fault_return(error); |
a7d73fe6 | 1165 | if (WARN_ON_ONCE(iomap.offset + iomap.length < pos + PAGE_SIZE)) { |
9f141d6e JK |
1166 | vmf_ret = dax_fault_return(-EIO); /* fs corruption? */ |
1167 | goto finish_iomap; | |
1168 | } | |
1169 | ||
1170 | entry = grab_mapping_entry(mapping, vmf->pgoff, 0); | |
1171 | if (IS_ERR(entry)) { | |
1172 | vmf_ret = dax_fault_return(PTR_ERR(entry)); | |
1550290b | 1173 | goto finish_iomap; |
a7d73fe6 CH |
1174 | } |
1175 | ||
333ccc97 | 1176 | sector = dax_iomap_sector(&iomap, pos); |
a7d73fe6 CH |
1177 | |
1178 | if (vmf->cow_page) { | |
1179 | switch (iomap.type) { | |
1180 | case IOMAP_HOLE: | |
1181 | case IOMAP_UNWRITTEN: | |
1182 | clear_user_highpage(vmf->cow_page, vaddr); | |
1183 | break; | |
1184 | case IOMAP_MAPPED: | |
1185 | error = copy_user_dax(iomap.bdev, sector, PAGE_SIZE, | |
1186 | vmf->cow_page, vaddr); | |
1187 | break; | |
1188 | default: | |
1189 | WARN_ON_ONCE(1); | |
1190 | error = -EIO; | |
1191 | break; | |
1192 | } | |
1193 | ||
1194 | if (error) | |
9f141d6e | 1195 | goto error_unlock_entry; |
b1aa812b JK |
1196 | |
1197 | __SetPageUptodate(vmf->cow_page); | |
1198 | vmf_ret = finish_fault(vmf); | |
1199 | if (!vmf_ret) | |
1200 | vmf_ret = VM_FAULT_DONE_COW; | |
9f141d6e | 1201 | goto unlock_entry; |
a7d73fe6 CH |
1202 | } |
1203 | ||
1204 | switch (iomap.type) { | |
1205 | case IOMAP_MAPPED: | |
1206 | if (iomap.flags & IOMAP_F_NEW) { | |
1207 | count_vm_event(PGMAJFAULT); | |
1208 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1209 | major = VM_FAULT_MAJOR; | |
1210 | } | |
1211 | error = dax_insert_mapping(mapping, iomap.bdev, sector, | |
1212 | PAGE_SIZE, &entry, vma, vmf); | |
9f141d6e JK |
1213 | /* -EBUSY is fine, somebody else faulted on the same PTE */ |
1214 | if (error == -EBUSY) | |
1215 | error = 0; | |
a7d73fe6 CH |
1216 | break; |
1217 | case IOMAP_UNWRITTEN: | |
1218 | case IOMAP_HOLE: | |
1550290b | 1219 | if (!(vmf->flags & FAULT_FLAG_WRITE)) { |
f449b936 | 1220 | vmf_ret = dax_load_hole(mapping, &entry, vmf); |
9f141d6e | 1221 | goto unlock_entry; |
1550290b | 1222 | } |
a7d73fe6 CH |
1223 | /*FALLTHRU*/ |
1224 | default: | |
1225 | WARN_ON_ONCE(1); | |
1226 | error = -EIO; | |
1227 | break; | |
1228 | } | |
1229 | ||
9f141d6e JK |
1230 | error_unlock_entry: |
1231 | vmf_ret = dax_fault_return(error) | major; | |
a7d73fe6 | 1232 | unlock_entry: |
f449b936 | 1233 | put_locked_mapping_entry(mapping, vmf->pgoff, entry); |
9f141d6e JK |
1234 | finish_iomap: |
1235 | if (ops->iomap_end) { | |
1236 | int copied = PAGE_SIZE; | |
1237 | ||
1238 | if (vmf_ret & VM_FAULT_ERROR) | |
1239 | copied = 0; | |
1240 | /* | |
1241 | * The fault is done by now and there's no way back (other | |
1242 | * thread may be already happily using PTE we have installed). | |
1243 | * Just ignore error from ->iomap_end since we cannot do much | |
1244 | * with it. | |
1245 | */ | |
1246 | ops->iomap_end(inode, pos, PAGE_SIZE, copied, flags, &iomap); | |
1550290b | 1247 | } |
9f141d6e | 1248 | return vmf_ret; |
a7d73fe6 | 1249 | } |
11c59c92 | 1250 | EXPORT_SYMBOL_GPL(dax_iomap_fault); |
642261ac RZ |
1251 | |
1252 | #ifdef CONFIG_FS_DAX_PMD | |
1253 | /* | |
1254 | * The 'colour' (ie low bits) within a PMD of a page offset. This comes up | |
1255 | * more often than one might expect in the below functions. | |
1256 | */ | |
1257 | #define PG_PMD_COLOUR ((PMD_SIZE >> PAGE_SHIFT) - 1) | |
1258 | ||
1259 | static int dax_pmd_insert_mapping(struct vm_area_struct *vma, pmd_t *pmd, | |
1260 | struct vm_fault *vmf, unsigned long address, | |
1261 | struct iomap *iomap, loff_t pos, bool write, void **entryp) | |
1262 | { | |
1263 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1264 | struct block_device *bdev = iomap->bdev; | |
27a7ffac | 1265 | struct inode *inode = mapping->host; |
642261ac RZ |
1266 | struct blk_dax_ctl dax = { |
1267 | .sector = dax_iomap_sector(iomap, pos), | |
1268 | .size = PMD_SIZE, | |
1269 | }; | |
1270 | long length = dax_map_atomic(bdev, &dax); | |
27a7ffac | 1271 | void *ret = NULL; |
642261ac RZ |
1272 | |
1273 | if (length < 0) /* dax_map_atomic() failed */ | |
27a7ffac | 1274 | goto fallback; |
642261ac RZ |
1275 | if (length < PMD_SIZE) |
1276 | goto unmap_fallback; | |
1277 | if (pfn_t_to_pfn(dax.pfn) & PG_PMD_COLOUR) | |
1278 | goto unmap_fallback; | |
1279 | if (!pfn_t_devmap(dax.pfn)) | |
1280 | goto unmap_fallback; | |
1281 | ||
1282 | dax_unmap_atomic(bdev, &dax); | |
1283 | ||
1284 | ret = dax_insert_mapping_entry(mapping, vmf, *entryp, dax.sector, | |
1285 | RADIX_DAX_PMD); | |
1286 | if (IS_ERR(ret)) | |
27a7ffac | 1287 | goto fallback; |
642261ac RZ |
1288 | *entryp = ret; |
1289 | ||
27a7ffac RZ |
1290 | trace_dax_pmd_insert_mapping(inode, vma, address, write, length, |
1291 | dax.pfn, ret); | |
642261ac RZ |
1292 | return vmf_insert_pfn_pmd(vma, address, pmd, dax.pfn, write); |
1293 | ||
1294 | unmap_fallback: | |
1295 | dax_unmap_atomic(bdev, &dax); | |
27a7ffac RZ |
1296 | fallback: |
1297 | trace_dax_pmd_insert_mapping_fallback(inode, vma, address, write, | |
1298 | length, dax.pfn, ret); | |
642261ac RZ |
1299 | return VM_FAULT_FALLBACK; |
1300 | } | |
1301 | ||
1302 | static int dax_pmd_load_hole(struct vm_area_struct *vma, pmd_t *pmd, | |
1303 | struct vm_fault *vmf, unsigned long address, | |
1304 | struct iomap *iomap, void **entryp) | |
1305 | { | |
1306 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1307 | unsigned long pmd_addr = address & PMD_MASK; | |
653b2ea3 | 1308 | struct inode *inode = mapping->host; |
642261ac | 1309 | struct page *zero_page; |
653b2ea3 | 1310 | void *ret = NULL; |
642261ac RZ |
1311 | spinlock_t *ptl; |
1312 | pmd_t pmd_entry; | |
642261ac RZ |
1313 | |
1314 | zero_page = mm_get_huge_zero_page(vma->vm_mm); | |
1315 | ||
1316 | if (unlikely(!zero_page)) | |
653b2ea3 | 1317 | goto fallback; |
642261ac RZ |
1318 | |
1319 | ret = dax_insert_mapping_entry(mapping, vmf, *entryp, 0, | |
1320 | RADIX_DAX_PMD | RADIX_DAX_HZP); | |
1321 | if (IS_ERR(ret)) | |
653b2ea3 | 1322 | goto fallback; |
642261ac RZ |
1323 | *entryp = ret; |
1324 | ||
1325 | ptl = pmd_lock(vma->vm_mm, pmd); | |
1326 | if (!pmd_none(*pmd)) { | |
1327 | spin_unlock(ptl); | |
653b2ea3 | 1328 | goto fallback; |
642261ac RZ |
1329 | } |
1330 | ||
1331 | pmd_entry = mk_pmd(zero_page, vma->vm_page_prot); | |
1332 | pmd_entry = pmd_mkhuge(pmd_entry); | |
1333 | set_pmd_at(vma->vm_mm, pmd_addr, pmd, pmd_entry); | |
1334 | spin_unlock(ptl); | |
653b2ea3 | 1335 | trace_dax_pmd_load_hole(inode, vma, address, zero_page, ret); |
642261ac | 1336 | return VM_FAULT_NOPAGE; |
653b2ea3 RZ |
1337 | |
1338 | fallback: | |
1339 | trace_dax_pmd_load_hole_fallback(inode, vma, address, zero_page, ret); | |
1340 | return VM_FAULT_FALLBACK; | |
642261ac RZ |
1341 | } |
1342 | ||
1343 | int dax_iomap_pmd_fault(struct vm_area_struct *vma, unsigned long address, | |
1344 | pmd_t *pmd, unsigned int flags, struct iomap_ops *ops) | |
1345 | { | |
1346 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1347 | unsigned long pmd_addr = address & PMD_MASK; | |
1348 | bool write = flags & FAULT_FLAG_WRITE; | |
9484ab1b | 1349 | unsigned int iomap_flags = (write ? IOMAP_WRITE : 0) | IOMAP_FAULT; |
642261ac RZ |
1350 | struct inode *inode = mapping->host; |
1351 | int result = VM_FAULT_FALLBACK; | |
1352 | struct iomap iomap = { 0 }; | |
1353 | pgoff_t max_pgoff, pgoff; | |
1354 | struct vm_fault vmf; | |
1355 | void *entry; | |
1356 | loff_t pos; | |
1357 | int error; | |
1358 | ||
282a8e03 RZ |
1359 | /* |
1360 | * Check whether offset isn't beyond end of file now. Caller is | |
1361 | * supposed to hold locks serializing us with truncate / punch hole so | |
1362 | * this is a reliable test. | |
1363 | */ | |
1364 | pgoff = linear_page_index(vma, pmd_addr); | |
1365 | max_pgoff = (i_size_read(inode) - 1) >> PAGE_SHIFT; | |
1366 | ||
1367 | trace_dax_pmd_fault(inode, vma, address, flags, pgoff, max_pgoff, 0); | |
1368 | ||
642261ac RZ |
1369 | /* Fall back to PTEs if we're going to COW */ |
1370 | if (write && !(vma->vm_flags & VM_SHARED)) | |
1371 | goto fallback; | |
1372 | ||
1373 | /* If the PMD would extend outside the VMA */ | |
1374 | if (pmd_addr < vma->vm_start) | |
1375 | goto fallback; | |
1376 | if ((pmd_addr + PMD_SIZE) > vma->vm_end) | |
1377 | goto fallback; | |
1378 | ||
282a8e03 RZ |
1379 | if (pgoff > max_pgoff) { |
1380 | result = VM_FAULT_SIGBUS; | |
1381 | goto out; | |
1382 | } | |
642261ac RZ |
1383 | |
1384 | /* If the PMD would extend beyond the file size */ | |
1385 | if ((pgoff | PG_PMD_COLOUR) > max_pgoff) | |
1386 | goto fallback; | |
1387 | ||
642261ac RZ |
1388 | /* |
1389 | * Note that we don't use iomap_apply here. We aren't doing I/O, only | |
1390 | * setting up a mapping, so really we're using iomap_begin() as a way | |
1391 | * to look up our filesystem block. | |
1392 | */ | |
1393 | pos = (loff_t)pgoff << PAGE_SHIFT; | |
1394 | error = ops->iomap_begin(inode, pos, PMD_SIZE, iomap_flags, &iomap); | |
1395 | if (error) | |
9f141d6e JK |
1396 | goto fallback; |
1397 | ||
642261ac RZ |
1398 | if (iomap.offset + iomap.length < pos + PMD_SIZE) |
1399 | goto finish_iomap; | |
1400 | ||
9f141d6e JK |
1401 | /* |
1402 | * grab_mapping_entry() will make sure we get a 2M empty entry, a DAX | |
1403 | * PMD or a HZP entry. If it can't (because a 4k page is already in | |
1404 | * the tree, for instance), it will return -EEXIST and we just fall | |
1405 | * back to 4k entries. | |
1406 | */ | |
1407 | entry = grab_mapping_entry(mapping, pgoff, RADIX_DAX_PMD); | |
1408 | if (IS_ERR(entry)) | |
1409 | goto finish_iomap; | |
1410 | ||
642261ac RZ |
1411 | vmf.pgoff = pgoff; |
1412 | vmf.flags = flags; | |
1413 | vmf.gfp_mask = mapping_gfp_mask(mapping) | __GFP_IO; | |
1414 | ||
1415 | switch (iomap.type) { | |
1416 | case IOMAP_MAPPED: | |
1417 | result = dax_pmd_insert_mapping(vma, pmd, &vmf, address, | |
1418 | &iomap, pos, write, &entry); | |
1419 | break; | |
1420 | case IOMAP_UNWRITTEN: | |
1421 | case IOMAP_HOLE: | |
1422 | if (WARN_ON_ONCE(write)) | |
9f141d6e | 1423 | goto unlock_entry; |
642261ac RZ |
1424 | result = dax_pmd_load_hole(vma, pmd, &vmf, address, &iomap, |
1425 | &entry); | |
1426 | break; | |
1427 | default: | |
1428 | WARN_ON_ONCE(1); | |
1429 | break; | |
1430 | } | |
1431 | ||
9f141d6e JK |
1432 | unlock_entry: |
1433 | put_locked_mapping_entry(mapping, pgoff, entry); | |
642261ac RZ |
1434 | finish_iomap: |
1435 | if (ops->iomap_end) { | |
9f141d6e JK |
1436 | int copied = PMD_SIZE; |
1437 | ||
1438 | if (result == VM_FAULT_FALLBACK) | |
1439 | copied = 0; | |
1440 | /* | |
1441 | * The fault is done by now and there's no way back (other | |
1442 | * thread may be already happily using PMD we have installed). | |
1443 | * Just ignore error from ->iomap_end since we cannot do much | |
1444 | * with it. | |
1445 | */ | |
1446 | ops->iomap_end(inode, pos, PMD_SIZE, copied, iomap_flags, | |
1447 | &iomap); | |
642261ac | 1448 | } |
642261ac RZ |
1449 | fallback: |
1450 | if (result == VM_FAULT_FALLBACK) { | |
1451 | split_huge_pmd(vma, pmd, address); | |
1452 | count_vm_event(THP_FAULT_FALLBACK); | |
1453 | } | |
282a8e03 RZ |
1454 | out: |
1455 | trace_dax_pmd_fault_done(inode, vma, address, flags, pgoff, max_pgoff, | |
1456 | result); | |
642261ac RZ |
1457 | return result; |
1458 | } | |
1459 | EXPORT_SYMBOL_GPL(dax_iomap_pmd_fault); | |
1460 | #endif /* CONFIG_FS_DAX_PMD */ |