Blame view
fs/direct-io.c
38.3 KB
1da177e4c
|
1 2 3 4 5 6 7 |
/* * fs/direct-io.c * * Copyright (C) 2002, Linus Torvalds. * * O_DIRECT * |
e1f8e8744
|
8 |
* 04Jul2002 Andrew Morton |
1da177e4c
|
9 10 11 |
* Initial version * 11Sep2002 janetinc@us.ibm.com * added readv/writev support. |
e1f8e8744
|
12 |
* 29Oct2002 Andrew Morton |
1da177e4c
|
13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 |
* rewrote bio_add_page() support. * 30Oct2002 pbadari@us.ibm.com * added support for non-aligned IO. * 06Nov2002 pbadari@us.ibm.com * added asynchronous IO support. * 21Jul2003 nathans@sgi.com * added IO completion notifier. */ #include <linux/kernel.h> #include <linux/module.h> #include <linux/types.h> #include <linux/fs.h> #include <linux/mm.h> #include <linux/slab.h> #include <linux/highmem.h> #include <linux/pagemap.h> |
98c4d57de
|
30 |
#include <linux/task_io_accounting_ops.h> |
1da177e4c
|
31 32 33 34 35 36 37 |
#include <linux/bio.h> #include <linux/wait.h> #include <linux/err.h> #include <linux/blkdev.h> #include <linux/buffer_head.h> #include <linux/rwsem.h> #include <linux/uio.h> |
60063497a
|
38 |
#include <linux/atomic.h> |
65dd2aa90
|
39 |
#include <linux/prefetch.h> |
1da177e4c
|
40 41 42 |
/* * How many user pages to map in one call to get_user_pages(). This determines |
cde1ecb32
|
43 |
* the size of a structure in the slab cache |
1da177e4c
|
44 45 46 47 48 49 50 51 52 53 54 55 56 |
*/ #define DIO_PAGES 64 /* * This code generally works in units of "dio_blocks". A dio_block is * somewhere between the hard sector size and the filesystem block size. it * is determined on a per-invocation basis. When talking to the filesystem * we need to convert dio_blocks to fs_blocks by scaling the dio_block quantity * down by dio->blkfactor. Similarly, fs-blocksize quantities are converted * to bio_block quantities by shifting left by blkfactor. * * If blkfactor is zero then the user's request was aligned to the filesystem's * blocksize. |
1da177e4c
|
57 |
*/ |
eb28be2b4
|
58 59 60 |
/* dio_state only used in the submission path */ struct dio_submit { |
1da177e4c
|
61 |
struct bio *bio; /* bio under assembly */ |
1da177e4c
|
62 63 64 65 66 67 68 69 70 71 |
unsigned blkbits; /* doesn't change */ unsigned blkfactor; /* When we're using an alignment which is finer than the filesystem's soft blocksize, this specifies how much finer. blkfactor=2 means 1/4-block alignment. Does not change */ unsigned start_zero_done; /* flag: sub-blocksize zeroing has been performed at the start of a write */ int pages_in_io; /* approximate total IO pages */ |
1da177e4c
|
72 73 74 |
sector_t block_in_file; /* Current offset into the underlying file in dio_block units. */ unsigned blocks_available; /* At block_in_file. changes */ |
0dc2bc49b
|
75 |
int reap_counter; /* rate limit reaping */ |
1da177e4c
|
76 |
sector_t final_block_in_request;/* doesn't change */ |
1da177e4c
|
77 |
int boundary; /* prev block is at a boundary */ |
1d8fa7a2b
|
78 |
get_block_t *get_block; /* block mapping function */ |
facd07b07
|
79 |
dio_submit_t *submit_io; /* IO submition function */ |
eb28be2b4
|
80 |
|
facd07b07
|
81 |
loff_t logical_offset_in_bio; /* current first logical block in bio */ |
1da177e4c
|
82 83 84 |
sector_t final_block_in_bio; /* current final block in bio + 1 */ sector_t next_block_for_io; /* next block to be put under IO, in dio_blocks units */ |
1da177e4c
|
85 86 87 88 89 90 91 92 93 94 |
/* * Deferred addition of a page to the dio. These variables are * private to dio_send_cur_page(), submit_page_section() and * dio_bio_add_page(). */ struct page *cur_page; /* The page */ unsigned cur_page_offset; /* Offset into it, in bytes */ unsigned cur_page_len; /* Nr of bytes at cur_page_offset */ sector_t cur_page_block; /* Where it starts */ |
facd07b07
|
95 |
loff_t cur_page_fs_offset; /* Offset in file */ |
1da177e4c
|
96 |
|
7b2c99d15
|
97 |
struct iov_iter *iter; |
1da177e4c
|
98 99 100 101 |
/* * Page queue. These variables belong to dio_refill_pages() and * dio_get_page(). */ |
1da177e4c
|
102 103 |
unsigned head; /* next page to process */ unsigned tail; /* last valid page + 1 */ |
7b2c99d15
|
104 |
size_t from, to; |
eb28be2b4
|
105 106 107 108 109 |
}; /* dio_state communicated between submission path and end_io */ struct dio { int flags; /* doesn't change */ |
eb28be2b4
|
110 |
int rw; |
15c4f638f
|
111 112 |
blk_qc_t bio_cookie; struct block_device *bio_bdev; |
0dc2bc49b
|
113 |
struct inode *inode; |
eb28be2b4
|
114 115 |
loff_t i_size; /* i_size when submitted */ dio_iodone_t *end_io; /* IO completion function */ |
eb28be2b4
|
116 |
|
18772641d
|
117 |
void *private; /* copy from map_bh.b_private */ |
eb28be2b4
|
118 119 120 |
/* BIO completion state */ spinlock_t bio_lock; /* protects BIO fields below */ |
0dc2bc49b
|
121 122 |
int page_errors; /* errno from get_user_pages() */ int is_async; /* is IO async ? */ |
7b7a8665e
|
123 |
bool defer_completion; /* defer AIO completion to workqueue? */ |
53cbf3b15
|
124 |
bool should_dirty; /* if pages should be dirtied */ |
0dc2bc49b
|
125 |
int io_error; /* IO error in completion path */ |
eb28be2b4
|
126 127 128 129 130 131 |
unsigned long refcount; /* direct_io_worker() and bios */ struct bio *bio_list; /* singly linked via bi_private */ struct task_struct *waiter; /* waiting task (NULL if none) */ /* AIO related stuff */ struct kiocb *iocb; /* kiocb */ |
eb28be2b4
|
132 |
ssize_t result; /* IO result */ |
23aee091d
|
133 134 135 136 137 |
/* * pages[] (and any fields placed after it) are not zeroed out at * allocation time. Don't add new fields after pages[] unless you * wish that they not be zeroed. */ |
7b7a8665e
|
138 139 140 141 |
union { struct page *pages[DIO_PAGES]; /* page buffer */ struct work_struct complete_work;/* deferred AIO completion */ }; |
6e8267f53
|
142 143 144 |
} ____cacheline_aligned_in_smp; static struct kmem_cache *dio_cache __read_mostly; |
1da177e4c
|
145 146 147 148 |
/* * How many pages are in the queue? */ |
eb28be2b4
|
149 |
static inline unsigned dio_pages_present(struct dio_submit *sdio) |
1da177e4c
|
150 |
{ |
eb28be2b4
|
151 |
return sdio->tail - sdio->head; |
1da177e4c
|
152 153 154 155 156 |
} /* * Go grab and pin some userspace pages. Typically we'll get 64 at a time. */ |
ba253fbf6
|
157 |
static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) |
1da177e4c
|
158 |
{ |
7b2c99d15
|
159 |
ssize_t ret; |
1da177e4c
|
160 |
|
2c80929c4
|
161 |
ret = iov_iter_get_pages(sdio->iter, dio->pages, LONG_MAX, DIO_PAGES, |
7b2c99d15
|
162 |
&sdio->from); |
1da177e4c
|
163 |
|
eb28be2b4
|
164 |
if (ret < 0 && sdio->blocks_available && (dio->rw & WRITE)) { |
557ed1fa2
|
165 |
struct page *page = ZERO_PAGE(0); |
1da177e4c
|
166 167 168 169 170 171 172 |
/* * A memory fault, but the filesystem has some outstanding * mapped blocks. We need to use those blocks up to avoid * leaking stale data in the file. */ if (dio->page_errors == 0) dio->page_errors = ret; |
09cbfeaf1
|
173 |
get_page(page); |
b5810039a
|
174 |
dio->pages[0] = page; |
eb28be2b4
|
175 176 |
sdio->head = 0; sdio->tail = 1; |
7b2c99d15
|
177 178 179 |
sdio->from = 0; sdio->to = PAGE_SIZE; return 0; |
1da177e4c
|
180 181 182 |
} if (ret >= 0) { |
7b2c99d15
|
183 184 |
iov_iter_advance(sdio->iter, ret); ret += sdio->from; |
eb28be2b4
|
185 |
sdio->head = 0; |
7b2c99d15
|
186 187 188 |
sdio->tail = (ret + PAGE_SIZE - 1) / PAGE_SIZE; sdio->to = ((ret - 1) & (PAGE_SIZE - 1)) + 1; return 0; |
1da177e4c
|
189 |
} |
1da177e4c
|
190 191 192 193 194 195 196 197 198 |
return ret; } /* * Get another userspace page. Returns an ERR_PTR on error. Pages are * buffered inside the dio so that we can call get_user_pages() against a * decent number of pages, less frequently. To provide nicer use of the * L1 cache. */ |
ba253fbf6
|
199 |
static inline struct page *dio_get_page(struct dio *dio, |
6fcc5420b
|
200 |
struct dio_submit *sdio) |
1da177e4c
|
201 |
{ |
eb28be2b4
|
202 |
if (dio_pages_present(sdio) == 0) { |
1da177e4c
|
203 |
int ret; |
eb28be2b4
|
204 |
ret = dio_refill_pages(dio, sdio); |
1da177e4c
|
205 206 |
if (ret) return ERR_PTR(ret); |
eb28be2b4
|
207 |
BUG_ON(dio_pages_present(sdio) == 0); |
1da177e4c
|
208 |
} |
6fcc5420b
|
209 |
return dio->pages[sdio->head]; |
1da177e4c
|
210 |
} |
6d544bb4d
|
211 212 213 214 |
/** * dio_complete() - called when all DIO BIO I/O has been completed * @offset: the byte offset in the file of the completed operation * |
7b7a8665e
|
215 216 |
* This drops i_dio_count, lets interested parties know that a DIO operation * has completed, and calculates the resulting return code for the operation. |
6d544bb4d
|
217 218 219 220 221 |
* * It lets the filesystem know if it registered an interest earlier via * get_block. Pass the private field of the map buffer_head so that * filesystems can use it to hold additional state between get_block calls and * dio_complete. |
1da177e4c
|
222 |
*/ |
7b7a8665e
|
223 224 |
static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is_async) |
1da177e4c
|
225 |
{ |
6d544bb4d
|
226 |
ssize_t transferred = 0; |
8459d86af
|
227 228 229 230 231 232 233 234 |
/* * AIO submission can race with bio completion to get here while * expecting to have the last io completed by bio completion. * In that case -EIOCBQUEUED is in fact not an error we want * to preserve through this call. */ if (ret == -EIOCBQUEUED) ret = 0; |
6d544bb4d
|
235 236 237 238 239 240 241 |
if (dio->result) { transferred = dio->result; /* Check for short read case */ if ((dio->rw == READ) && ((offset + transferred) > dio->i_size)) transferred = dio->i_size - offset; } |
6d544bb4d
|
242 243 244 245 246 247 |
if (ret == 0) ret = dio->page_errors; if (ret == 0) ret = dio->io_error; if (ret == 0) ret = transferred; |
187372a3b
|
248 249 250 251 252 253 254 |
if (dio->end_io) { int err; err = dio->end_io(dio->iocb, offset, ret, dio->private); if (err) ret = err; } |
7b7a8665e
|
255 |
|
fe0f07d08
|
256 257 |
if (!(dio->flags & DIO_SKIP_DIO_COUNT)) inode_dio_end(dio->inode); |
02afc27fa
|
258 259 260 261 262 263 264 265 266 |
if (is_async) { if (dio->rw & WRITE) { int err; err = generic_write_sync(dio->iocb->ki_filp, offset, transferred); if (err < 0 && ret > 0) ret = err; } |
04b2fa9f8
|
267 |
dio->iocb->ki_complete(dio->iocb, ret, 0); |
02afc27fa
|
268 |
} |
40e2e9731
|
269 |
|
7b7a8665e
|
270 |
kmem_cache_free(dio_cache, dio); |
6d544bb4d
|
271 |
return ret; |
1da177e4c
|
272 |
} |
7b7a8665e
|
273 274 275 276 277 278 |
static void dio_aio_complete_work(struct work_struct *work) { struct dio *dio = container_of(work, struct dio, complete_work); dio_complete(dio, dio->iocb->ki_pos, 0, true); } |
1da177e4c
|
279 |
static int dio_bio_complete(struct dio *dio, struct bio *bio); |
7b7a8665e
|
280 |
|
1da177e4c
|
281 282 283 |
/* * Asynchronous IO callback. */ |
4246a0b63
|
284 |
static void dio_bio_end_aio(struct bio *bio) |
1da177e4c
|
285 286 |
{ struct dio *dio = bio->bi_private; |
5eb6c7a2a
|
287 288 |
unsigned long remaining; unsigned long flags; |
1da177e4c
|
289 |
|
1da177e4c
|
290 291 |
/* cleanup the bio */ dio_bio_complete(dio, bio); |
0273201e6
|
292 |
|
5eb6c7a2a
|
293 294 295 |
spin_lock_irqsave(&dio->bio_lock, flags); remaining = --dio->refcount; if (remaining == 1 && dio->waiter) |
20258b2b3
|
296 |
wake_up_process(dio->waiter); |
5eb6c7a2a
|
297 |
spin_unlock_irqrestore(&dio->bio_lock, flags); |
20258b2b3
|
298 |
|
8459d86af
|
299 |
if (remaining == 0) { |
7b7a8665e
|
300 301 302 303 304 305 306 |
if (dio->result && dio->defer_completion) { INIT_WORK(&dio->complete_work, dio_aio_complete_work); queue_work(dio->inode->i_sb->s_dio_done_wq, &dio->complete_work); } else { dio_complete(dio, dio->iocb->ki_pos, 0, true); } |
8459d86af
|
307 |
} |
1da177e4c
|
308 309 310 311 312 313 314 315 316 |
} /* * The BIO completion handler simply queues the BIO up for the process-context * handler. * * During I/O bi_private points at the dio. After I/O, bi_private is used to * implement a singly-linked list of completed BIOs, at dio->bio_list. */ |
4246a0b63
|
317 |
static void dio_bio_end_io(struct bio *bio) |
1da177e4c
|
318 319 320 |
{ struct dio *dio = bio->bi_private; unsigned long flags; |
1da177e4c
|
321 322 323 |
spin_lock_irqsave(&dio->bio_lock, flags); bio->bi_private = dio->bio_list; dio->bio_list = bio; |
5eb6c7a2a
|
324 |
if (--dio->refcount == 1 && dio->waiter) |
1da177e4c
|
325 326 |
wake_up_process(dio->waiter); spin_unlock_irqrestore(&dio->bio_lock, flags); |
1da177e4c
|
327 |
} |
facd07b07
|
328 329 330 331 332 333 334 335 336 337 338 339 340 341 |
/** * dio_end_io - handle the end io action for the given bio * @bio: The direct io bio thats being completed * @error: Error if there was one * * This is meant to be called by any filesystem that uses their own dio_submit_t * so that the DIO specific endio actions are dealt with after the filesystem * has done it's completion work. */ void dio_end_io(struct bio *bio, int error) { struct dio *dio = bio->bi_private; if (dio->is_async) |
4246a0b63
|
342 |
dio_bio_end_aio(bio); |
facd07b07
|
343 |
else |
4246a0b63
|
344 |
dio_bio_end_io(bio); |
facd07b07
|
345 346 |
} EXPORT_SYMBOL_GPL(dio_end_io); |
ba253fbf6
|
347 |
static inline void |
eb28be2b4
|
348 349 350 |
dio_bio_alloc(struct dio *dio, struct dio_submit *sdio, struct block_device *bdev, sector_t first_sector, int nr_vecs) |
1da177e4c
|
351 352 |
{ struct bio *bio; |
20d9600cb
|
353 354 |
/* * bio_alloc() is guaranteed to return a bio when called with |
71baba4b9
|
355 |
* __GFP_RECLAIM and we request a valid number of vectors. |
20d9600cb
|
356 |
*/ |
1da177e4c
|
357 |
bio = bio_alloc(GFP_KERNEL, nr_vecs); |
1da177e4c
|
358 359 |
bio->bi_bdev = bdev; |
4f024f379
|
360 |
bio->bi_iter.bi_sector = first_sector; |
1da177e4c
|
361 362 363 364 |
if (dio->is_async) bio->bi_end_io = dio_bio_end_aio; else bio->bi_end_io = dio_bio_end_io; |
eb28be2b4
|
365 366 |
sdio->bio = bio; sdio->logical_offset_in_bio = sdio->cur_page_fs_offset; |
1da177e4c
|
367 368 369 370 371 372 |
} /* * In the AIO read case we speculatively dirty the pages before starting IO. * During IO completion, any of these pages which happen to have been written * back will be redirtied by bio_check_pages_dirty(). |
0273201e6
|
373 374 |
* * bios hold a dio reference between submit_bio and ->end_io. |
1da177e4c
|
375 |
*/ |
ba253fbf6
|
376 |
static inline void dio_bio_submit(struct dio *dio, struct dio_submit *sdio) |
1da177e4c
|
377 |
{ |
eb28be2b4
|
378 |
struct bio *bio = sdio->bio; |
5eb6c7a2a
|
379 |
unsigned long flags; |
1da177e4c
|
380 381 |
bio->bi_private = dio; |
5eb6c7a2a
|
382 383 384 385 |
spin_lock_irqsave(&dio->bio_lock, flags); dio->refcount++; spin_unlock_irqrestore(&dio->bio_lock, flags); |
53cbf3b15
|
386 |
if (dio->is_async && dio->rw == READ && dio->should_dirty) |
1da177e4c
|
387 |
bio_set_pages_dirty(bio); |
5eb6c7a2a
|
388 |
|
c1c534609
|
389 |
dio->bio_bdev = bio->bi_bdev; |
15c4f638f
|
390 |
if (sdio->submit_io) { |
eb28be2b4
|
391 392 |
sdio->submit_io(dio->rw, bio, dio->inode, sdio->logical_offset_in_bio); |
15c4f638f
|
393 |
dio->bio_cookie = BLK_QC_T_NONE; |
c1c534609
|
394 |
} else |
15c4f638f
|
395 |
dio->bio_cookie = submit_bio(dio->rw, bio); |
1da177e4c
|
396 |
|
eb28be2b4
|
397 398 399 |
sdio->bio = NULL; sdio->boundary = 0; sdio->logical_offset_in_bio = 0; |
1da177e4c
|
400 401 402 403 404 |
} /* * Release any resources in case of a failure */ |
ba253fbf6
|
405 |
static inline void dio_cleanup(struct dio *dio, struct dio_submit *sdio) |
1da177e4c
|
406 |
{ |
7b2c99d15
|
407 |
while (sdio->head < sdio->tail) |
09cbfeaf1
|
408 |
put_page(dio->pages[sdio->head++]); |
1da177e4c
|
409 410 411 |
} /* |
0273201e6
|
412 413 414 415 |
* Wait for the next BIO to complete. Remove it and return it. NULL is * returned once all BIOs have been completed. This must only be called once * all bios have been issued so that dio->refcount can only decrease. This * requires that that the caller hold a reference on the dio. |
1da177e4c
|
416 417 418 419 |
*/ static struct bio *dio_await_one(struct dio *dio) { unsigned long flags; |
0273201e6
|
420 |
struct bio *bio = NULL; |
1da177e4c
|
421 422 |
spin_lock_irqsave(&dio->bio_lock, flags); |
5eb6c7a2a
|
423 424 425 426 427 428 429 430 431 432 433 |
/* * Wait as long as the list is empty and there are bios in flight. bio * completion drops the count, maybe adds to the list, and wakes while * holding the bio_lock so we don't need set_current_state()'s barrier * and can call it after testing our condition. */ while (dio->refcount > 1 && dio->bio_list == NULL) { __set_current_state(TASK_UNINTERRUPTIBLE); dio->waiter = current; spin_unlock_irqrestore(&dio->bio_lock, flags); |
c43c83a29
|
434 435 |
if (!(dio->iocb->ki_flags & IOCB_HIPRI) || !blk_poll(bdev_get_queue(dio->bio_bdev), dio->bio_cookie)) |
15c4f638f
|
436 |
io_schedule(); |
5eb6c7a2a
|
437 438 439 |
/* wake up sets us TASK_RUNNING */ spin_lock_irqsave(&dio->bio_lock, flags); dio->waiter = NULL; |
1da177e4c
|
440 |
} |
0273201e6
|
441 442 443 444 |
if (dio->bio_list) { bio = dio->bio_list; dio->bio_list = bio->bi_private; } |
1da177e4c
|
445 446 447 448 449 450 451 452 453 |
spin_unlock_irqrestore(&dio->bio_lock, flags); return bio; } /* * Process one completed BIO. No locks are held. */ static int dio_bio_complete(struct dio *dio, struct bio *bio) { |
cb34e057a
|
454 455 |
struct bio_vec *bvec; unsigned i; |
9b81c8423
|
456 |
int err; |
1da177e4c
|
457 |
|
4246a0b63
|
458 |
if (bio->bi_error) |
174e27c60
|
459 |
dio->io_error = -EIO; |
1da177e4c
|
460 |
|
53cbf3b15
|
461 |
if (dio->is_async && dio->rw == READ && dio->should_dirty) { |
9b81c8423
|
462 |
err = bio->bi_error; |
7ddc971f8
|
463 |
bio_check_pages_dirty(bio); /* transfers ownership */ |
1da177e4c
|
464 |
} else { |
cb34e057a
|
465 466 |
bio_for_each_segment_all(bvec, bio, i) { struct page *page = bvec->bv_page; |
1da177e4c
|
467 |
|
53cbf3b15
|
468 469 |
if (dio->rw == READ && !PageCompound(page) && dio->should_dirty) |
1da177e4c
|
470 |
set_page_dirty_lock(page); |
09cbfeaf1
|
471 |
put_page(page); |
1da177e4c
|
472 |
} |
9b81c8423
|
473 |
err = bio->bi_error; |
1da177e4c
|
474 475 |
bio_put(bio); } |
9b81c8423
|
476 |
return err; |
1da177e4c
|
477 478 479 |
} /* |
0273201e6
|
480 481 482 |
* Wait on and process all in-flight BIOs. This must only be called once * all bios have been issued so that the refcount can only decrease. * This just waits for all bios to make it through dio_bio_complete. IO |
beb7dd86a
|
483 |
* errors are propagated through dio->io_error and should be propagated via |
0273201e6
|
484 |
* dio_complete(). |
1da177e4c
|
485 |
*/ |
6d544bb4d
|
486 |
static void dio_await_completion(struct dio *dio) |
1da177e4c
|
487 |
{ |
0273201e6
|
488 489 490 491 492 493 |
struct bio *bio; do { bio = dio_await_one(dio); if (bio) dio_bio_complete(dio, bio); } while (bio); |
1da177e4c
|
494 495 496 497 498 499 500 501 502 |
} /* * A really large O_DIRECT read or write can generate a lot of BIOs. So * to keep the memory consumption sane we periodically reap any completed BIOs * during the BIO generation phase. * * This also helps to limit the peak amount of pinned userspace memory. */ |
ba253fbf6
|
503 |
static inline int dio_bio_reap(struct dio *dio, struct dio_submit *sdio) |
1da177e4c
|
504 505 |
{ int ret = 0; |
eb28be2b4
|
506 |
if (sdio->reap_counter++ >= 64) { |
1da177e4c
|
507 508 509 510 511 512 513 514 515 516 517 518 519 |
while (dio->bio_list) { unsigned long flags; struct bio *bio; int ret2; spin_lock_irqsave(&dio->bio_lock, flags); bio = dio->bio_list; dio->bio_list = bio->bi_private; spin_unlock_irqrestore(&dio->bio_lock, flags); ret2 = dio_bio_complete(dio, bio); if (ret == 0) ret = ret2; } |
eb28be2b4
|
520 |
sdio->reap_counter = 0; |
1da177e4c
|
521 522 523 524 525 |
} return ret; } /* |
7b7a8665e
|
526 527 528 529 530 531 532 |
* Create workqueue for deferred direct IO completions. We allocate the * workqueue when it's first needed. This avoids creating workqueue for * filesystems that don't need it and also allows us to create the workqueue * late enough so the we can include s_id in the name of the workqueue. */ static int sb_init_dio_done_wq(struct super_block *sb) { |
45150c43b
|
533 |
struct workqueue_struct *old; |
7b7a8665e
|
534 535 536 537 538 539 540 541 |
struct workqueue_struct *wq = alloc_workqueue("dio/%s", WQ_MEM_RECLAIM, 0, sb->s_id); if (!wq) return -ENOMEM; /* * This has to be atomic as more DIOs can race to create the workqueue */ |
45150c43b
|
542 |
old = cmpxchg(&sb->s_dio_done_wq, NULL, wq); |
7b7a8665e
|
543 |
/* Someone created workqueue before us? Free ours... */ |
45150c43b
|
544 |
if (old) |
7b7a8665e
|
545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 |
destroy_workqueue(wq); return 0; } static int dio_set_defer_completion(struct dio *dio) { struct super_block *sb = dio->inode->i_sb; if (dio->defer_completion) return 0; dio->defer_completion = true; if (!sb->s_dio_done_wq) return sb_init_dio_done_wq(sb); return 0; } /* |
1da177e4c
|
562 |
* Call into the fs to map some more disk blocks. We record the current number |
eb28be2b4
|
563 |
* of available blocks at sdio->blocks_available. These are in units of the |
1da177e4c
|
564 565 566 567 568 |
* fs blocksize, (1 << inode->i_blkbits). * * The fs is allowed to map lots of blocks at once. If it wants to do that, * it uses the passed inode-relative block number as the file offset, as usual. * |
1d8fa7a2b
|
569 |
* get_block() is passed the number of i_blkbits-sized blocks which direct_io |
1da177e4c
|
570 571 572 573 574 575 576 577 578 579 580 581 |
* has remaining to do. The fs should not map more than this number of blocks. * * If the fs has mapped a lot of blocks, it should populate bh->b_size to * indicate how much contiguous disk space has been made available at * bh->b_blocknr. * * If *any* of the mapped blocks are new, then the fs must set buffer_new(). * This isn't very efficient... * * In the case of filesystem holes: the fs may return an arbitrarily-large * hole by returning an appropriate value in b_size and by clearing * buffer_mapped(). However the direct-io code will only process holes one |
1d8fa7a2b
|
582 |
* block at a time - it will repeatedly call get_block() as it walks the hole. |
1da177e4c
|
583 |
*/ |
18772641d
|
584 585 |
static int get_more_blocks(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) |
1da177e4c
|
586 587 |
{ int ret; |
1da177e4c
|
588 |
sector_t fs_startblk; /* Into file, in filesystem-sized blocks */ |
ae55e1aaa
|
589 |
sector_t fs_endblk; /* Into file, in filesystem-sized blocks */ |
1da177e4c
|
590 |
unsigned long fs_count; /* Number of filesystem-sized blocks */ |
1da177e4c
|
591 |
int create; |
ab73857e3
|
592 |
unsigned int i_blkbits = sdio->blkbits + sdio->blkfactor; |
1da177e4c
|
593 594 595 596 597 598 599 |
/* * If there was a memory error and we've overwritten all the * mapped blocks then we can now return that memory error */ ret = dio->page_errors; if (ret == 0) { |
eb28be2b4
|
600 601 |
BUG_ON(sdio->block_in_file >= sdio->final_block_in_request); fs_startblk = sdio->block_in_file >> sdio->blkfactor; |
ae55e1aaa
|
602 603 604 |
fs_endblk = (sdio->final_block_in_request - 1) >> sdio->blkfactor; fs_count = fs_endblk - fs_startblk + 1; |
1da177e4c
|
605 |
|
3c674e742
|
606 |
map_bh->b_state = 0; |
ab73857e3
|
607 |
map_bh->b_size = fs_count << i_blkbits; |
3c674e742
|
608 |
|
5fe878ae7
|
609 610 611 612 613 614 615 616 617 618 619 |
/* * For writes inside i_size on a DIO_SKIP_HOLES filesystem we * forbid block creations: only overwrites are permitted. * We will return early to the caller once we see an * unmapped buffer head returned, and the caller will fall * back to buffered I/O. * * Otherwise the decision is left to the get_blocks method, * which may decide to handle it or also return an unmapped * buffer head. */ |
b31dc66a5
|
620 |
create = dio->rw & WRITE; |
5fe878ae7
|
621 |
if (dio->flags & DIO_SKIP_HOLES) { |
eb28be2b4
|
622 623 |
if (sdio->block_in_file < (i_size_read(dio->inode) >> sdio->blkbits)) |
1da177e4c
|
624 |
create = 0; |
1da177e4c
|
625 |
} |
3c674e742
|
626 |
|
eb28be2b4
|
627 |
ret = (*sdio->get_block)(dio->inode, fs_startblk, |
1da177e4c
|
628 |
map_bh, create); |
18772641d
|
629 630 631 |
/* Store for completion */ dio->private = map_bh->b_private; |
7b7a8665e
|
632 633 634 |
if (ret == 0 && buffer_defer_completion(map_bh)) ret = dio_set_defer_completion(dio); |
1da177e4c
|
635 636 637 638 639 640 641 |
} return ret; } /* * There is no bio. Make one now. */ |
ba253fbf6
|
642 643 |
static inline int dio_new_bio(struct dio *dio, struct dio_submit *sdio, sector_t start_sector, struct buffer_head *map_bh) |
1da177e4c
|
644 645 646 |
{ sector_t sector; int ret, nr_pages; |
eb28be2b4
|
647 |
ret = dio_bio_reap(dio, sdio); |
1da177e4c
|
648 649 |
if (ret) goto out; |
eb28be2b4
|
650 |
sector = start_sector << (sdio->blkbits - 9); |
b54ffb73c
|
651 |
nr_pages = min(sdio->pages_in_io, BIO_MAX_PAGES); |
1da177e4c
|
652 |
BUG_ON(nr_pages <= 0); |
18772641d
|
653 |
dio_bio_alloc(dio, sdio, map_bh->b_bdev, sector, nr_pages); |
eb28be2b4
|
654 |
sdio->boundary = 0; |
1da177e4c
|
655 656 657 658 659 660 661 662 663 664 665 |
out: return ret; } /* * Attempt to put the current chunk of 'cur_page' into the current BIO. If * that was successful then update final_block_in_bio and take a ref against * the just-added page. * * Return zero on success. Non-zero means the caller needs to start a new BIO. */ |
ba253fbf6
|
666 |
static inline int dio_bio_add_page(struct dio_submit *sdio) |
1da177e4c
|
667 668 |
{ int ret; |
eb28be2b4
|
669 670 671 |
ret = bio_add_page(sdio->bio, sdio->cur_page, sdio->cur_page_len, sdio->cur_page_offset); if (ret == sdio->cur_page_len) { |
1da177e4c
|
672 673 674 |
/* * Decrement count only, if we are done with this page */ |
eb28be2b4
|
675 676 |
if ((sdio->cur_page_len + sdio->cur_page_offset) == PAGE_SIZE) sdio->pages_in_io--; |
09cbfeaf1
|
677 |
get_page(sdio->cur_page); |
eb28be2b4
|
678 679 |
sdio->final_block_in_bio = sdio->cur_page_block + (sdio->cur_page_len >> sdio->blkbits); |
1da177e4c
|
680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 |
ret = 0; } else { ret = 1; } return ret; } /* * Put cur_page under IO. The section of cur_page which is described by * cur_page_offset,cur_page_len is put into a BIO. The section of cur_page * starts on-disk at cur_page_block. * * We take a ref against the page here (on behalf of its presence in the bio). * * The caller of this function is responsible for removing cur_page from the * dio, and for dropping the refcount which came from that presence. */ |
ba253fbf6
|
697 698 |
static inline int dio_send_cur_page(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) |
1da177e4c
|
699 700 |
{ int ret = 0; |
eb28be2b4
|
701 702 703 |
if (sdio->bio) { loff_t cur_offset = sdio->cur_page_fs_offset; loff_t bio_next_offset = sdio->logical_offset_in_bio + |
4f024f379
|
704 |
sdio->bio->bi_iter.bi_size; |
c2c6ca417
|
705 |
|
1da177e4c
|
706 |
/* |
c2c6ca417
|
707 708 |
* See whether this new request is contiguous with the old. * |
f0940cee2
|
709 710 |
* Btrfs cannot handle having logically non-contiguous requests * submitted. For example if you have |
c2c6ca417
|
711 712 |
* * Logical: [0-4095][HOLE][8192-12287] |
f0940cee2
|
713 |
* Physical: [0-4095] [4096-8191] |
c2c6ca417
|
714 715 716 717 718 |
* * We cannot submit those pages together as one BIO. So if our * current logical offset in the file does not equal what would * be the next logical offset in the bio, submit the bio we * have. |
1da177e4c
|
719 |
*/ |
eb28be2b4
|
720 |
if (sdio->final_block_in_bio != sdio->cur_page_block || |
c2c6ca417
|
721 |
cur_offset != bio_next_offset) |
eb28be2b4
|
722 |
dio_bio_submit(dio, sdio); |
1da177e4c
|
723 |
} |
eb28be2b4
|
724 |
if (sdio->bio == NULL) { |
18772641d
|
725 |
ret = dio_new_bio(dio, sdio, sdio->cur_page_block, map_bh); |
1da177e4c
|
726 727 728 |
if (ret) goto out; } |
eb28be2b4
|
729 730 |
if (dio_bio_add_page(sdio) != 0) { dio_bio_submit(dio, sdio); |
18772641d
|
731 |
ret = dio_new_bio(dio, sdio, sdio->cur_page_block, map_bh); |
1da177e4c
|
732 |
if (ret == 0) { |
eb28be2b4
|
733 |
ret = dio_bio_add_page(sdio); |
1da177e4c
|
734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 |
BUG_ON(ret != 0); } } out: return ret; } /* * An autonomous function to put a chunk of a page under deferred IO. * * The caller doesn't actually know (or care) whether this piece of page is in * a BIO, or is under IO or whatever. We just take care of all possible * situations here. The separation between the logic of do_direct_IO() and * that of submit_page_section() is important for clarity. Please don't break. * * The chunk of page starts on-disk at blocknr. * * We perform deferred IO, by recording the last-submitted page inside our * private part of the dio structure. If possible, we just expand the IO * across that page here. * * If that doesn't work out then we put the old page into the bio and add this * page to the dio instead. */ |
ba253fbf6
|
758 |
static inline int |
eb28be2b4
|
759 |
submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page, |
18772641d
|
760 761 |
unsigned offset, unsigned len, sector_t blocknr, struct buffer_head *map_bh) |
1da177e4c
|
762 763 |
{ int ret = 0; |
98c4d57de
|
764 765 766 767 768 769 |
if (dio->rw & WRITE) { /* * Read accounting is performed in submit_bio() */ task_io_account_write(len); } |
1da177e4c
|
770 771 772 |
/* * Can we just grow the current page's presence in the dio? */ |
eb28be2b4
|
773 774 775 776 777 |
if (sdio->cur_page == page && sdio->cur_page_offset + sdio->cur_page_len == offset && sdio->cur_page_block + (sdio->cur_page_len >> sdio->blkbits) == blocknr) { sdio->cur_page_len += len; |
1da177e4c
|
778 779 780 781 782 783 |
goto out; } /* * If there's a deferred page already there then send it. */ |
eb28be2b4
|
784 |
if (sdio->cur_page) { |
18772641d
|
785 |
ret = dio_send_cur_page(dio, sdio, map_bh); |
09cbfeaf1
|
786 |
put_page(sdio->cur_page); |
eb28be2b4
|
787 |
sdio->cur_page = NULL; |
1da177e4c
|
788 |
if (ret) |
b1058b981
|
789 |
return ret; |
1da177e4c
|
790 |
} |
09cbfeaf1
|
791 |
get_page(page); /* It is in dio */ |
eb28be2b4
|
792 793 794 795 796 |
sdio->cur_page = page; sdio->cur_page_offset = offset; sdio->cur_page_len = len; sdio->cur_page_block = blocknr; sdio->cur_page_fs_offset = sdio->block_in_file << sdio->blkbits; |
1da177e4c
|
797 |
out: |
b1058b981
|
798 799 800 801 802 803 804 |
/* * If sdio->boundary then we want to schedule the IO now to * avoid metadata seeks. */ if (sdio->boundary) { ret = dio_send_cur_page(dio, sdio, map_bh); dio_bio_submit(dio, sdio); |
09cbfeaf1
|
805 |
put_page(sdio->cur_page); |
b1058b981
|
806 807 |
sdio->cur_page = NULL; } |
1da177e4c
|
808 809 810 811 812 813 814 815 |
return ret; } /* * Clean any dirty buffers in the blockdev mapping which alias newly-created * file blocks. Only called for S_ISREG files - blockdevs do not set * buffer_new */ |
18772641d
|
816 |
static void clean_blockdev_aliases(struct dio *dio, struct buffer_head *map_bh) |
1da177e4c
|
817 818 819 |
{ unsigned i; unsigned nblocks; |
18772641d
|
820 |
nblocks = map_bh->b_size >> dio->inode->i_blkbits; |
1da177e4c
|
821 822 |
for (i = 0; i < nblocks; i++) { |
18772641d
|
823 824 |
unmap_underlying_metadata(map_bh->b_bdev, map_bh->b_blocknr + i); |
1da177e4c
|
825 826 827 828 829 830 831 832 833 834 835 836 |
} } /* * If we are not writing the entire block and get_block() allocated * the block for us, we need to fill-in the unused portion of the * block with zeros. This happens only if user-buffer, fileoffset or * io length is not filesystem block-size multiple. * * `end' is zero if we're doing the start of the IO, 1 at the end of the * IO. */ |
ba253fbf6
|
837 838 |
static inline void dio_zero_block(struct dio *dio, struct dio_submit *sdio, int end, struct buffer_head *map_bh) |
1da177e4c
|
839 840 841 842 843 |
{ unsigned dio_blocks_per_fs_block; unsigned this_chunk_blocks; /* In dio_blocks */ unsigned this_chunk_bytes; struct page *page; |
eb28be2b4
|
844 |
sdio->start_zero_done = 1; |
18772641d
|
845 |
if (!sdio->blkfactor || !buffer_new(map_bh)) |
1da177e4c
|
846 |
return; |
eb28be2b4
|
847 848 |
dio_blocks_per_fs_block = 1 << sdio->blkfactor; this_chunk_blocks = sdio->block_in_file & (dio_blocks_per_fs_block - 1); |
1da177e4c
|
849 850 851 852 853 854 855 856 857 858 |
if (!this_chunk_blocks) return; /* * We need to zero out part of an fs block. It is either at the * beginning or the end of the fs block. */ if (end) this_chunk_blocks = dio_blocks_per_fs_block - this_chunk_blocks; |
eb28be2b4
|
859 |
this_chunk_bytes = this_chunk_blocks << sdio->blkbits; |
1da177e4c
|
860 |
|
557ed1fa2
|
861 |
page = ZERO_PAGE(0); |
eb28be2b4
|
862 |
if (submit_page_section(dio, sdio, page, 0, this_chunk_bytes, |
18772641d
|
863 |
sdio->next_block_for_io, map_bh)) |
1da177e4c
|
864 |
return; |
eb28be2b4
|
865 |
sdio->next_block_for_io += this_chunk_blocks; |
1da177e4c
|
866 867 868 869 870 871 872 873 874 875 876 |
} /* * Walk the user pages, and the file, mapping blocks to disk and generating * a sequence of (page,offset,len,block) mappings. These mappings are injected * into submit_page_section(), which takes care of the next stage of submission * * Direct IO against a blockdev is different from a file. Because we can * happily perform page-sized but 512-byte aligned IOs. It is important that * blockdev IO be able to have fine alignment and large sizes. * |
1d8fa7a2b
|
877 |
* So what we do is to permit the ->get_block function to populate bh.b_size |
1da177e4c
|
878 879 880 |
* with the size of IO which is permitted at this offset and this i_blkbits. * * For best results, the blockdev should be set up with 512-byte i_blkbits and |
1d8fa7a2b
|
881 |
* it should set b_size to PAGE_SIZE or more inside get_block(). This gives |
1da177e4c
|
882 883 |
* fine alignment but still allows this function to work in PAGE_SIZE units. */ |
18772641d
|
884 885 |
static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) |
1da177e4c
|
886 |
{ |
eb28be2b4
|
887 |
const unsigned blkbits = sdio->blkbits; |
1da177e4c
|
888 |
int ret = 0; |
eb28be2b4
|
889 |
while (sdio->block_in_file < sdio->final_block_in_request) { |
7b2c99d15
|
890 891 |
struct page *page; size_t from, to; |
6fcc5420b
|
892 893 |
page = dio_get_page(dio, sdio); |
1da177e4c
|
894 895 896 897 |
if (IS_ERR(page)) { ret = PTR_ERR(page); goto out; } |
6fcc5420b
|
898 899 900 |
from = sdio->head ? 0 : sdio->from; to = (sdio->head == sdio->tail - 1) ? sdio->to : PAGE_SIZE; sdio->head++; |
1da177e4c
|
901 |
|
7b2c99d15
|
902 |
while (from < to) { |
1da177e4c
|
903 904 905 |
unsigned this_chunk_bytes; /* # of bytes mapped */ unsigned this_chunk_blocks; /* # of blocks */ unsigned u; |
eb28be2b4
|
906 |
if (sdio->blocks_available == 0) { |
1da177e4c
|
907 908 909 910 911 |
/* * Need to go and map some more disk */ unsigned long blkmask; unsigned long dio_remainder; |
18772641d
|
912 |
ret = get_more_blocks(dio, sdio, map_bh); |
1da177e4c
|
913 |
if (ret) { |
09cbfeaf1
|
914 |
put_page(page); |
1da177e4c
|
915 916 917 918 |
goto out; } if (!buffer_mapped(map_bh)) goto do_holes; |
eb28be2b4
|
919 920 921 922 |
sdio->blocks_available = map_bh->b_size >> sdio->blkbits; sdio->next_block_for_io = map_bh->b_blocknr << sdio->blkfactor; |
1da177e4c
|
923 |
if (buffer_new(map_bh)) |
18772641d
|
924 |
clean_blockdev_aliases(dio, map_bh); |
1da177e4c
|
925 |
|
eb28be2b4
|
926 |
if (!sdio->blkfactor) |
1da177e4c
|
927 |
goto do_holes; |
eb28be2b4
|
928 929 |
blkmask = (1 << sdio->blkfactor) - 1; dio_remainder = (sdio->block_in_file & blkmask); |
1da177e4c
|
930 931 932 933 934 935 936 937 938 939 940 941 942 |
/* * If we are at the start of IO and that IO * starts partway into a fs-block, * dio_remainder will be non-zero. If the IO * is a read then we can simply advance the IO * cursor to the first block which is to be * read. But if the IO is a write and the * block was newly allocated we cannot do that; * the start of the fs block must be zeroed out * on-disk */ if (!buffer_new(map_bh)) |
eb28be2b4
|
943 944 |
sdio->next_block_for_io += dio_remainder; sdio->blocks_available -= dio_remainder; |
1da177e4c
|
945 946 947 948 |
} do_holes: /* Handle holes */ if (!buffer_mapped(map_bh)) { |
35dc8161d
|
949 |
loff_t i_size_aligned; |
1da177e4c
|
950 951 |
/* AKPM: eargh, -ENOTBLK is a hack */ |
b31dc66a5
|
952 |
if (dio->rw & WRITE) { |
09cbfeaf1
|
953 |
put_page(page); |
1da177e4c
|
954 955 |
return -ENOTBLK; } |
35dc8161d
|
956 957 958 959 960 961 |
/* * Be sure to account for a partial block as the * last block in the file */ i_size_aligned = ALIGN(i_size_read(dio->inode), 1 << blkbits); |
eb28be2b4
|
962 |
if (sdio->block_in_file >= |
35dc8161d
|
963 |
i_size_aligned >> blkbits) { |
1da177e4c
|
964 |
/* We hit eof */ |
09cbfeaf1
|
965 |
put_page(page); |
1da177e4c
|
966 967 |
goto out; } |
7b2c99d15
|
968 |
zero_user(page, from, 1 << blkbits); |
eb28be2b4
|
969 |
sdio->block_in_file++; |
7b2c99d15
|
970 |
from += 1 << blkbits; |
3320c60b3
|
971 |
dio->result += 1 << blkbits; |
1da177e4c
|
972 973 974 975 976 977 978 979 |
goto next_block; } /* * If we're performing IO which has an alignment which * is finer than the underlying fs, go check to see if * we must zero out the start of this block. */ |
eb28be2b4
|
980 |
if (unlikely(sdio->blkfactor && !sdio->start_zero_done)) |
18772641d
|
981 |
dio_zero_block(dio, sdio, 0, map_bh); |
1da177e4c
|
982 983 984 985 986 |
/* * Work out, in this_chunk_blocks, how much disk we * can add to this page */ |
eb28be2b4
|
987 |
this_chunk_blocks = sdio->blocks_available; |
7b2c99d15
|
988 |
u = (to - from) >> blkbits; |
1da177e4c
|
989 990 |
if (this_chunk_blocks > u) this_chunk_blocks = u; |
eb28be2b4
|
991 |
u = sdio->final_block_in_request - sdio->block_in_file; |
1da177e4c
|
992 993 994 995 |
if (this_chunk_blocks > u) this_chunk_blocks = u; this_chunk_bytes = this_chunk_blocks << blkbits; BUG_ON(this_chunk_bytes == 0); |
092c8d46e
|
996 997 |
if (this_chunk_blocks == sdio->blocks_available) sdio->boundary = buffer_boundary(map_bh); |
eb28be2b4
|
998 |
ret = submit_page_section(dio, sdio, page, |
7b2c99d15
|
999 |
from, |
eb28be2b4
|
1000 |
this_chunk_bytes, |
18772641d
|
1001 1002 |
sdio->next_block_for_io, map_bh); |
1da177e4c
|
1003 |
if (ret) { |
09cbfeaf1
|
1004 |
put_page(page); |
1da177e4c
|
1005 1006 |
goto out; } |
eb28be2b4
|
1007 |
sdio->next_block_for_io += this_chunk_blocks; |
1da177e4c
|
1008 |
|
eb28be2b4
|
1009 |
sdio->block_in_file += this_chunk_blocks; |
7b2c99d15
|
1010 1011 |
from += this_chunk_bytes; dio->result += this_chunk_bytes; |
eb28be2b4
|
1012 |
sdio->blocks_available -= this_chunk_blocks; |
1da177e4c
|
1013 |
next_block: |
eb28be2b4
|
1014 1015 |
BUG_ON(sdio->block_in_file > sdio->final_block_in_request); if (sdio->block_in_file == sdio->final_block_in_request) |
1da177e4c
|
1016 1017 1018 1019 |
break; } /* Drop the ref which was taken in get_user_pages() */ |
09cbfeaf1
|
1020 |
put_page(page); |
1da177e4c
|
1021 1022 1023 1024 |
} out: return ret; } |
847cc6371
|
1025 |
static inline int drop_refcount(struct dio *dio) |
1da177e4c
|
1026 |
{ |
847cc6371
|
1027 |
int ret2; |
5eb6c7a2a
|
1028 |
unsigned long flags; |
1da177e4c
|
1029 |
|
8459d86af
|
1030 1031 |
/* * Sync will always be dropping the final ref and completing the |
5eb6c7a2a
|
1032 1033 1034 |
* operation. AIO can if it was a broken operation described above or * in fact if all the bios race to complete before we get here. In * that case dio_complete() translates the EIOCBQUEUED into the proper |
04b2fa9f8
|
1035 |
* return code that the caller will hand to ->complete(). |
5eb6c7a2a
|
1036 1037 1038 1039 |
* * This is managed by the bio_lock instead of being an atomic_t so that * completion paths can drop their ref and use the remaining count to * decide to wake the submission path atomically. |
8459d86af
|
1040 |
*/ |
5eb6c7a2a
|
1041 1042 1043 |
spin_lock_irqsave(&dio->bio_lock, flags); ret2 = --dio->refcount; spin_unlock_irqrestore(&dio->bio_lock, flags); |
847cc6371
|
1044 |
return ret2; |
1da177e4c
|
1045 |
} |
eafdc7d19
|
1046 1047 1048 1049 1050 1051 1052 1053 1054 |
/* * This is a library function for use by filesystem drivers. * * The locking rules are governed by the flags parameter: * - if the flags value contains DIO_LOCKING we use a fancy locking * scheme for dumb filesystems. * For writes this function is called under i_mutex and returns with * i_mutex held, for reads, i_mutex is not held on entry, but it is * taken and dropped again before returning. |
eafdc7d19
|
1055 1056 1057 |
* - if the flags value does NOT contain DIO_LOCKING we don't use any * internal locking but rather rely on the filesystem to synchronize * direct I/O reads/writes versus each other and truncate. |
df2d6f265
|
1058 1059 1060 1061 1062 1063 1064 |
* * To help with locking against truncate we incremented the i_dio_count * counter before starting direct I/O, and decrement it once we are done. * Truncate can wait for it to reach zero to provide exclusion. It is * expected that filesystem provide exclusion between new direct I/O * and truncates. For DIO_LOCKING filesystems this is done by i_mutex, * but other filesystems need to take care of this on their own. |
ba253fbf6
|
1065 1066 1067 1068 1069 |
* * NOTE: if you pass "sdio" to anything by pointer make sure that function * is always inlined. Otherwise gcc is unable to split the structure into * individual fields and will generate much worse code. This is important * for the whole file. |
eafdc7d19
|
1070 |
*/ |
65dd2aa90
|
1071 |
static inline ssize_t |
17f8c842d
|
1072 1073 1074 1075 |
do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode, struct block_device *bdev, struct iov_iter *iter, loff_t offset, get_block_t get_block, dio_iodone_t end_io, dio_submit_t submit_io, int flags) |
1da177e4c
|
1076 |
{ |
ab73857e3
|
1077 1078 |
unsigned i_blkbits = ACCESS_ONCE(inode->i_blkbits); unsigned blkbits = i_blkbits; |
1da177e4c
|
1079 1080 |
unsigned blocksize_mask = (1 << blkbits) - 1; ssize_t retval = -EINVAL; |
af4364727
|
1081 1082 |
size_t count = iov_iter_count(iter); loff_t end = offset + count; |
1da177e4c
|
1083 |
struct dio *dio; |
eb28be2b4
|
1084 |
struct dio_submit sdio = { 0, }; |
847cc6371
|
1085 |
struct buffer_head map_bh = { 0, }; |
647d1e4c5
|
1086 |
struct blk_plug plug; |
886a39115
|
1087 |
unsigned long align = offset | iov_iter_alignment(iter); |
1da177e4c
|
1088 |
|
65dd2aa90
|
1089 1090 1091 1092 |
/* * Avoid references to bdev if not absolutely needed to give * the early prefetch in the caller enough time. */ |
1da177e4c
|
1093 |
|
886a39115
|
1094 |
if (align & blocksize_mask) { |
1da177e4c
|
1095 |
if (bdev) |
65dd2aa90
|
1096 |
blkbits = blksize_bits(bdev_logical_block_size(bdev)); |
1da177e4c
|
1097 |
blocksize_mask = (1 << blkbits) - 1; |
886a39115
|
1098 |
if (align & blocksize_mask) |
1da177e4c
|
1099 1100 |
goto out; } |
f9b5570d7
|
1101 |
/* watch out for a 0 len io from a tricksy fs */ |
17f8c842d
|
1102 |
if (iov_iter_rw(iter) == READ && !iov_iter_count(iter)) |
f9b5570d7
|
1103 |
return 0; |
6e8267f53
|
1104 |
dio = kmem_cache_alloc(dio_cache, GFP_KERNEL); |
1da177e4c
|
1105 1106 1107 |
retval = -ENOMEM; if (!dio) goto out; |
23aee091d
|
1108 1109 1110 1111 1112 1113 |
/* * Believe it or not, zeroing out the page array caused a .5% * performance regression in a database benchmark. So, we take * care to only zero out what's needed. */ memset(dio, 0, offsetof(struct dio, pages)); |
1da177e4c
|
1114 |
|
5fe878ae7
|
1115 1116 |
dio->flags = flags; if (dio->flags & DIO_LOCKING) { |
17f8c842d
|
1117 |
if (iov_iter_rw(iter) == READ) { |
5fe878ae7
|
1118 1119 |
struct address_space *mapping = iocb->ki_filp->f_mapping; |
1da177e4c
|
1120 |
|
5fe878ae7
|
1121 |
/* will be released by direct_io_worker */ |
5955102c9
|
1122 |
inode_lock(inode); |
1da177e4c
|
1123 1124 1125 1126 |
retval = filemap_write_and_wait_range(mapping, offset, end - 1); if (retval) { |
5955102c9
|
1127 |
inode_unlock(inode); |
6e8267f53
|
1128 |
kmem_cache_free(dio_cache, dio); |
1da177e4c
|
1129 1130 |
goto out; } |
1da177e4c
|
1131 |
} |
1da177e4c
|
1132 |
} |
74cedf9b6
|
1133 1134 1135 1136 |
/* Once we sampled i_size check for reads beyond EOF */ dio->i_size = i_size_read(inode); if (iov_iter_rw(iter) == READ && offset >= dio->i_size) { if (dio->flags & DIO_LOCKING) |
5955102c9
|
1137 |
inode_unlock(inode); |
74cedf9b6
|
1138 |
kmem_cache_free(dio_cache, dio); |
2d4594acb
|
1139 |
retval = 0; |
74cedf9b6
|
1140 1141 |
goto out; } |
1da177e4c
|
1142 |
/* |
603925737
|
1143 1144 1145 1146 |
* For file extending writes updating i_size before data writeouts * complete can expose uninitialized blocks in dumb filesystems. * In that case we need to wait for I/O completion even if asked * for an asynchronous write. |
1da177e4c
|
1147 |
*/ |
603925737
|
1148 1149 1150 |
if (is_sync_kiocb(iocb)) dio->is_async = false; else if (!(dio->flags & DIO_ASYNC_EXTEND) && |
17f8c842d
|
1151 |
iov_iter_rw(iter) == WRITE && end > i_size_read(inode)) |
603925737
|
1152 1153 1154 |
dio->is_async = false; else dio->is_async = true; |
847cc6371
|
1155 |
dio->inode = inode; |
17f8c842d
|
1156 |
dio->rw = iov_iter_rw(iter) == WRITE ? WRITE_ODIRECT : READ; |
02afc27fa
|
1157 1158 1159 1160 1161 |
/* * For AIO O_(D)SYNC writes we need to defer completions to a workqueue * so that we can call ->fsync. */ |
17f8c842d
|
1162 |
if (dio->is_async && iov_iter_rw(iter) == WRITE && |
02afc27fa
|
1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 |
((iocb->ki_filp->f_flags & O_DSYNC) || IS_SYNC(iocb->ki_filp->f_mapping->host))) { retval = dio_set_defer_completion(dio); if (retval) { /* * We grab i_mutex only for reads so we don't have * to release it here */ kmem_cache_free(dio_cache, dio); goto out; } } /* * Will be decremented at I/O completion time. */ |
fe0f07d08
|
1179 1180 |
if (!(dio->flags & DIO_SKIP_DIO_COUNT)) inode_dio_begin(inode); |
02afc27fa
|
1181 1182 |
retval = 0; |
847cc6371
|
1183 |
sdio.blkbits = blkbits; |
ab73857e3
|
1184 |
sdio.blkfactor = i_blkbits - blkbits; |
847cc6371
|
1185 1186 1187 1188 1189 1190 1191 1192 1193 |
sdio.block_in_file = offset >> blkbits; sdio.get_block = get_block; dio->end_io = end_io; sdio.submit_io = submit_io; sdio.final_block_in_bio = -1; sdio.next_block_for_io = -1; dio->iocb = iocb; |
847cc6371
|
1194 1195 1196 |
spin_lock_init(&dio->bio_lock); dio->refcount = 1; |
53cbf3b15
|
1197 |
dio->should_dirty = (iter->type == ITER_IOVEC); |
7b2c99d15
|
1198 1199 1200 |
sdio.iter = iter; sdio.final_block_in_request = (offset + iov_iter_count(iter)) >> blkbits; |
847cc6371
|
1201 1202 1203 1204 1205 1206 |
/* * In case of non-aligned buffers, we may need 2 more * pages since we need to zero out first and last block. */ if (unlikely(sdio.blkfactor)) sdio.pages_in_io = 2; |
f67da30c1
|
1207 |
sdio.pages_in_io += iov_iter_npages(iter, INT_MAX); |
847cc6371
|
1208 |
|
647d1e4c5
|
1209 |
blk_start_plug(&plug); |
7b2c99d15
|
1210 1211 1212 |
retval = do_direct_IO(dio, &sdio, &map_bh); if (retval) dio_cleanup(dio, &sdio); |
847cc6371
|
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 |
if (retval == -ENOTBLK) { /* * The remaining part of the request will be * be handled by buffered I/O when we return */ retval = 0; } /* * There may be some unwritten disk at the end of a part-written * fs-block-sized block. Go zero that now. */ dio_zero_block(dio, &sdio, 1, &map_bh); if (sdio.cur_page) { ssize_t ret2; ret2 = dio_send_cur_page(dio, &sdio, &map_bh); if (retval == 0) retval = ret2; |
09cbfeaf1
|
1233 |
put_page(sdio.cur_page); |
847cc6371
|
1234 1235 1236 1237 |
sdio.cur_page = NULL; } if (sdio.bio) dio_bio_submit(dio, &sdio); |
647d1e4c5
|
1238 |
blk_finish_plug(&plug); |
847cc6371
|
1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 |
/* * It is possible that, we return short IO due to end of file. * In that case, we need to release all the pages we got hold on. */ dio_cleanup(dio, &sdio); /* * All block lookups have been performed. For READ requests * we can let i_mutex go now that its achieved its purpose * of protecting us from looking up uninitialized blocks. */ |
17f8c842d
|
1250 |
if (iov_iter_rw(iter) == READ && (dio->flags & DIO_LOCKING)) |
5955102c9
|
1251 |
inode_unlock(dio->inode); |
847cc6371
|
1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 |
/* * The only time we want to leave bios in flight is when a successful * partial aio read or full aio write have been setup. In that case * bio completion will call aio_complete. The only time it's safe to * call aio_complete is when we return -EIOCBQUEUED, so we key on that. * This had *better* be the only place that raises -EIOCBQUEUED. */ BUG_ON(retval == -EIOCBQUEUED); if (dio->is_async && retval == 0 && dio->result && |
17f8c842d
|
1262 |
(iov_iter_rw(iter) == READ || dio->result == count)) |
847cc6371
|
1263 |
retval = -EIOCBQUEUED; |
af4364727
|
1264 |
else |
847cc6371
|
1265 1266 1267 1268 |
dio_await_completion(dio); if (drop_refcount(dio) == 0) { retval = dio_complete(dio, offset, retval, false); |
847cc6371
|
1269 1270 |
} else BUG_ON(retval != -EIOCBQUEUED); |
1da177e4c
|
1271 |
|
7bb46a673
|
1272 1273 1274 |
out: return retval; } |
65dd2aa90
|
1275 |
|
17f8c842d
|
1276 1277 1278 1279 1280 |
ssize_t __blockdev_direct_IO(struct kiocb *iocb, struct inode *inode, struct block_device *bdev, struct iov_iter *iter, loff_t offset, get_block_t get_block, dio_iodone_t end_io, dio_submit_t submit_io, int flags) |
65dd2aa90
|
1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 |
{ /* * The block device state is needed in the end to finally * submit everything. Since it's likely to be cache cold * prefetch it here as first thing to hide some of the * latency. * * Attempt to prefetch the pieces we likely need later. */ prefetch(&bdev->bd_disk->part_tbl); prefetch(bdev->bd_queue); prefetch((char *)bdev->bd_queue + SMP_CACHE_BYTES); |
17f8c842d
|
1293 1294 |
return do_blockdev_direct_IO(iocb, inode, bdev, iter, offset, get_block, end_io, submit_io, flags); |
65dd2aa90
|
1295 |
} |
1da177e4c
|
1296 |
EXPORT_SYMBOL(__blockdev_direct_IO); |
6e8267f53
|
1297 1298 1299 1300 1301 1302 1303 |
static __init int dio_init(void) { dio_cache = KMEM_CACHE(dio, SLAB_PANIC); return 0; } module_init(dio_init) |