Blame view
fs/direct-io.c
37.7 KB
1da177e4c Linux-2.6.12-rc2 |
1 2 3 4 5 6 7 |
/* * fs/direct-io.c * * Copyright (C) 2002, Linus Torvalds. * * O_DIRECT * |
e1f8e8744 Remove Andrew Mor... |
8 |
* 04Jul2002 Andrew Morton |
1da177e4c Linux-2.6.12-rc2 |
9 10 11 |
* Initial version * 11Sep2002 janetinc@us.ibm.com * added readv/writev support. |
e1f8e8744 Remove Andrew Mor... |
12 |
* 29Oct2002 Andrew Morton |
1da177e4c Linux-2.6.12-rc2 |
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 [PATCH] io-accoun... |
30 |
#include <linux/task_io_accounting_ops.h> |
1da177e4c Linux-2.6.12-rc2 |
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 atomic: use <linu... |
38 |
#include <linux/atomic.h> |
65dd2aa90 dio: optimize cac... |
39 |
#include <linux/prefetch.h> |
1da177e4c Linux-2.6.12-rc2 |
40 41 42 |
/* * How many user pages to map in one call to get_user_pages(). This determines |
cde1ecb32 direct-io: fix a ... |
43 |
* the size of a structure in the slab cache |
1da177e4c Linux-2.6.12-rc2 |
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 Linux-2.6.12-rc2 |
57 |
*/ |
eb28be2b4 direct-io: separa... |
58 59 60 |
/* dio_state only used in the submission path */ struct dio_submit { |
1da177e4c Linux-2.6.12-rc2 |
61 |
struct bio *bio; /* bio under assembly */ |
1da177e4c Linux-2.6.12-rc2 |
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 Linux-2.6.12-rc2 |
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 direct-io: rearra... |
75 |
int reap_counter; /* rate limit reaping */ |
1da177e4c Linux-2.6.12-rc2 |
76 |
sector_t final_block_in_request;/* doesn't change */ |
1da177e4c Linux-2.6.12-rc2 |
77 |
int boundary; /* prev block is at a boundary */ |
1d8fa7a2b [PATCH] remove ->... |
78 |
get_block_t *get_block; /* block mapping function */ |
facd07b07 direct-io: add a ... |
79 |
dio_submit_t *submit_io; /* IO submition function */ |
eb28be2b4 direct-io: separa... |
80 |
|
facd07b07 direct-io: add a ... |
81 |
loff_t logical_offset_in_bio; /* current first logical block in bio */ |
1da177e4c Linux-2.6.12-rc2 |
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 Linux-2.6.12-rc2 |
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 direct-io: add a ... |
95 |
loff_t cur_page_fs_offset; /* Offset in file */ |
1da177e4c Linux-2.6.12-rc2 |
96 |
|
7b2c99d15 new helper: iov_i... |
97 |
struct iov_iter *iter; |
1da177e4c Linux-2.6.12-rc2 |
98 99 100 101 |
/* * Page queue. These variables belong to dio_refill_pages() and * dio_get_page(). */ |
1da177e4c Linux-2.6.12-rc2 |
102 103 |
unsigned head; /* next page to process */ unsigned tail; /* last valid page + 1 */ |
7b2c99d15 new helper: iov_i... |
104 |
size_t from, to; |
eb28be2b4 direct-io: separa... |
105 106 107 108 109 |
}; /* dio_state communicated between submission path and end_io */ struct dio { int flags; /* doesn't change */ |
eb28be2b4 direct-io: separa... |
110 |
int rw; |
0dc2bc49b direct-io: rearra... |
111 |
struct inode *inode; |
eb28be2b4 direct-io: separa... |
112 113 |
loff_t i_size; /* i_size when submitted */ dio_iodone_t *end_io; /* IO completion function */ |
eb28be2b4 direct-io: separa... |
114 |
|
18772641d direct-io: separa... |
115 |
void *private; /* copy from map_bh.b_private */ |
eb28be2b4 direct-io: separa... |
116 117 118 |
/* BIO completion state */ spinlock_t bio_lock; /* protects BIO fields below */ |
0dc2bc49b direct-io: rearra... |
119 120 |
int page_errors; /* errno from get_user_pages() */ int is_async; /* is IO async ? */ |
7b7a8665e direct-io: Implem... |
121 |
bool defer_completion; /* defer AIO completion to workqueue? */ |
0dc2bc49b direct-io: rearra... |
122 |
int io_error; /* IO error in completion path */ |
eb28be2b4 direct-io: separa... |
123 124 125 126 127 128 |
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 direct-io: separa... |
129 |
ssize_t result; /* IO result */ |
23aee091d dio: don't zero o... |
130 131 132 133 134 |
/* * 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 direct-io: Implem... |
135 136 137 138 |
union { struct page *pages[DIO_PAGES]; /* page buffer */ struct work_struct complete_work;/* deferred AIO completion */ }; |
6e8267f53 direct-io: use a ... |
139 140 141 |
} ____cacheline_aligned_in_smp; static struct kmem_cache *dio_cache __read_mostly; |
1da177e4c Linux-2.6.12-rc2 |
142 143 144 145 |
/* * How many pages are in the queue? */ |
eb28be2b4 direct-io: separa... |
146 |
static inline unsigned dio_pages_present(struct dio_submit *sdio) |
1da177e4c Linux-2.6.12-rc2 |
147 |
{ |
eb28be2b4 direct-io: separa... |
148 |
return sdio->tail - sdio->head; |
1da177e4c Linux-2.6.12-rc2 |
149 150 151 152 153 |
} /* * Go grab and pin some userspace pages. Typically we'll get 64 at a time. */ |
ba253fbf6 direct-io: inline... |
154 |
static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) |
1da177e4c Linux-2.6.12-rc2 |
155 |
{ |
7b2c99d15 new helper: iov_i... |
156 |
ssize_t ret; |
1da177e4c Linux-2.6.12-rc2 |
157 |
|
2c80929c4 fuse: honour max_... |
158 |
ret = iov_iter_get_pages(sdio->iter, dio->pages, LONG_MAX, DIO_PAGES, |
7b2c99d15 new helper: iov_i... |
159 |
&sdio->from); |
1da177e4c Linux-2.6.12-rc2 |
160 |
|
eb28be2b4 direct-io: separa... |
161 |
if (ret < 0 && sdio->blocks_available && (dio->rw & WRITE)) { |
557ed1fa2 remove ZERO_PAGE |
162 |
struct page *page = ZERO_PAGE(0); |
1da177e4c Linux-2.6.12-rc2 |
163 164 165 166 167 168 169 |
/* * 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; |
b5810039a [PATCH] core remo... |
170 171 |
page_cache_get(page); dio->pages[0] = page; |
eb28be2b4 direct-io: separa... |
172 173 |
sdio->head = 0; sdio->tail = 1; |
7b2c99d15 new helper: iov_i... |
174 175 176 |
sdio->from = 0; sdio->to = PAGE_SIZE; return 0; |
1da177e4c Linux-2.6.12-rc2 |
177 178 179 |
} if (ret >= 0) { |
7b2c99d15 new helper: iov_i... |
180 181 |
iov_iter_advance(sdio->iter, ret); ret += sdio->from; |
eb28be2b4 direct-io: separa... |
182 |
sdio->head = 0; |
7b2c99d15 new helper: iov_i... |
183 184 185 |
sdio->tail = (ret + PAGE_SIZE - 1) / PAGE_SIZE; sdio->to = ((ret - 1) & (PAGE_SIZE - 1)) + 1; return 0; |
1da177e4c Linux-2.6.12-rc2 |
186 |
} |
1da177e4c Linux-2.6.12-rc2 |
187 188 189 190 191 192 193 194 195 |
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 direct-io: inline... |
196 |
static inline struct page *dio_get_page(struct dio *dio, |
6fcc5420b direct-io: fix un... |
197 |
struct dio_submit *sdio) |
1da177e4c Linux-2.6.12-rc2 |
198 |
{ |
eb28be2b4 direct-io: separa... |
199 |
if (dio_pages_present(sdio) == 0) { |
1da177e4c Linux-2.6.12-rc2 |
200 |
int ret; |
eb28be2b4 direct-io: separa... |
201 |
ret = dio_refill_pages(dio, sdio); |
1da177e4c Linux-2.6.12-rc2 |
202 203 |
if (ret) return ERR_PTR(ret); |
eb28be2b4 direct-io: separa... |
204 |
BUG_ON(dio_pages_present(sdio) == 0); |
1da177e4c Linux-2.6.12-rc2 |
205 |
} |
6fcc5420b direct-io: fix un... |
206 |
return dio->pages[sdio->head]; |
1da177e4c Linux-2.6.12-rc2 |
207 |
} |
6d544bb4d [PATCH] dio: cent... |
208 209 210 211 |
/** * dio_complete() - called when all DIO BIO I/O has been completed * @offset: the byte offset in the file of the completed operation * |
7b7a8665e direct-io: Implem... |
212 213 |
* 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 [PATCH] dio: cent... |
214 215 216 217 218 |
* * 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 Linux-2.6.12-rc2 |
219 |
*/ |
7b7a8665e direct-io: Implem... |
220 221 |
static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is_async) |
1da177e4c Linux-2.6.12-rc2 |
222 |
{ |
6d544bb4d [PATCH] dio: cent... |
223 |
ssize_t transferred = 0; |
8459d86af [PATCH] dio: only... |
224 225 226 227 228 229 230 231 |
/* * 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 [PATCH] dio: cent... |
232 233 234 235 236 237 238 |
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 [PATCH] dio: cent... |
239 240 241 242 243 244 |
if (ret == 0) ret = dio->page_errors; if (ret == 0) ret = dio->io_error; if (ret == 0) ret = transferred; |
7b7a8665e direct-io: Implem... |
245 246 |
if (dio->end_io && dio->result) dio->end_io(dio->iocb, offset, transferred, dio->private); |
fe0f07d08 direct-io: only i... |
247 248 |
if (!(dio->flags & DIO_SKIP_DIO_COUNT)) inode_dio_end(dio->inode); |
02afc27fa direct-io: Handle... |
249 250 251 252 253 254 255 256 257 |
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 fs: split generic... |
258 |
dio->iocb->ki_complete(dio->iocb, ret, 0); |
02afc27fa direct-io: Handle... |
259 |
} |
40e2e9731 direct-io: move a... |
260 |
|
7b7a8665e direct-io: Implem... |
261 |
kmem_cache_free(dio_cache, dio); |
6d544bb4d [PATCH] dio: cent... |
262 |
return ret; |
1da177e4c Linux-2.6.12-rc2 |
263 |
} |
7b7a8665e direct-io: Implem... |
264 265 266 267 268 269 |
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 Linux-2.6.12-rc2 |
270 |
static int dio_bio_complete(struct dio *dio, struct bio *bio); |
7b7a8665e direct-io: Implem... |
271 |
|
1da177e4c Linux-2.6.12-rc2 |
272 273 274 |
/* * Asynchronous IO callback. */ |
4246a0b63 block: add a bi_e... |
275 |
static void dio_bio_end_aio(struct bio *bio) |
1da177e4c Linux-2.6.12-rc2 |
276 277 |
{ struct dio *dio = bio->bi_private; |
5eb6c7a2a [PATCH] dio: lock... |
278 279 |
unsigned long remaining; unsigned long flags; |
1da177e4c Linux-2.6.12-rc2 |
280 |
|
1da177e4c Linux-2.6.12-rc2 |
281 282 |
/* cleanup the bio */ dio_bio_complete(dio, bio); |
0273201e6 [PATCH] dio: form... |
283 |
|
5eb6c7a2a [PATCH] dio: lock... |
284 285 286 |
spin_lock_irqsave(&dio->bio_lock, flags); remaining = --dio->refcount; if (remaining == 1 && dio->waiter) |
20258b2b3 [PATCH] dio: remo... |
287 |
wake_up_process(dio->waiter); |
5eb6c7a2a [PATCH] dio: lock... |
288 |
spin_unlock_irqrestore(&dio->bio_lock, flags); |
20258b2b3 [PATCH] dio: remo... |
289 |
|
8459d86af [PATCH] dio: only... |
290 |
if (remaining == 0) { |
7b7a8665e direct-io: Implem... |
291 292 293 294 295 296 297 |
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 [PATCH] dio: only... |
298 |
} |
1da177e4c Linux-2.6.12-rc2 |
299 300 301 302 303 304 305 306 307 |
} /* * 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 block: add a bi_e... |
308 |
static void dio_bio_end_io(struct bio *bio) |
1da177e4c Linux-2.6.12-rc2 |
309 310 311 |
{ struct dio *dio = bio->bi_private; unsigned long flags; |
1da177e4c Linux-2.6.12-rc2 |
312 313 314 |
spin_lock_irqsave(&dio->bio_lock, flags); bio->bi_private = dio->bio_list; dio->bio_list = bio; |
5eb6c7a2a [PATCH] dio: lock... |
315 |
if (--dio->refcount == 1 && dio->waiter) |
1da177e4c Linux-2.6.12-rc2 |
316 317 |
wake_up_process(dio->waiter); spin_unlock_irqrestore(&dio->bio_lock, flags); |
1da177e4c Linux-2.6.12-rc2 |
318 |
} |
facd07b07 direct-io: add a ... |
319 320 321 322 323 324 325 326 327 328 329 330 331 332 |
/** * 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 block: add a bi_e... |
333 |
dio_bio_end_aio(bio); |
facd07b07 direct-io: add a ... |
334 |
else |
4246a0b63 block: add a bi_e... |
335 |
dio_bio_end_io(bio); |
facd07b07 direct-io: add a ... |
336 337 |
} EXPORT_SYMBOL_GPL(dio_end_io); |
ba253fbf6 direct-io: inline... |
338 |
static inline void |
eb28be2b4 direct-io: separa... |
339 340 341 |
dio_bio_alloc(struct dio *dio, struct dio_submit *sdio, struct block_device *bdev, sector_t first_sector, int nr_vecs) |
1da177e4c Linux-2.6.12-rc2 |
342 343 |
{ struct bio *bio; |
20d9600cb fs/direct-io.c: d... |
344 345 346 347 |
/* * bio_alloc() is guaranteed to return a bio when called with * __GFP_WAIT and we request a valid number of vectors. */ |
1da177e4c Linux-2.6.12-rc2 |
348 |
bio = bio_alloc(GFP_KERNEL, nr_vecs); |
1da177e4c Linux-2.6.12-rc2 |
349 350 |
bio->bi_bdev = bdev; |
4f024f379 block: Abstract o... |
351 |
bio->bi_iter.bi_sector = first_sector; |
1da177e4c Linux-2.6.12-rc2 |
352 353 354 355 |
if (dio->is_async) bio->bi_end_io = dio_bio_end_aio; else bio->bi_end_io = dio_bio_end_io; |
eb28be2b4 direct-io: separa... |
356 357 |
sdio->bio = bio; sdio->logical_offset_in_bio = sdio->cur_page_fs_offset; |
1da177e4c Linux-2.6.12-rc2 |
358 359 360 361 362 363 |
} /* * 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 [PATCH] dio: form... |
364 365 |
* * bios hold a dio reference between submit_bio and ->end_io. |
1da177e4c Linux-2.6.12-rc2 |
366 |
*/ |
ba253fbf6 direct-io: inline... |
367 |
static inline void dio_bio_submit(struct dio *dio, struct dio_submit *sdio) |
1da177e4c Linux-2.6.12-rc2 |
368 |
{ |
eb28be2b4 direct-io: separa... |
369 |
struct bio *bio = sdio->bio; |
5eb6c7a2a [PATCH] dio: lock... |
370 |
unsigned long flags; |
1da177e4c Linux-2.6.12-rc2 |
371 372 |
bio->bi_private = dio; |
5eb6c7a2a [PATCH] dio: lock... |
373 374 375 376 |
spin_lock_irqsave(&dio->bio_lock, flags); dio->refcount++; spin_unlock_irqrestore(&dio->bio_lock, flags); |
1da177e4c Linux-2.6.12-rc2 |
377 378 |
if (dio->is_async && dio->rw == READ) bio_set_pages_dirty(bio); |
5eb6c7a2a [PATCH] dio: lock... |
379 |
|
eb28be2b4 direct-io: separa... |
380 381 382 |
if (sdio->submit_io) sdio->submit_io(dio->rw, bio, dio->inode, sdio->logical_offset_in_bio); |
facd07b07 direct-io: add a ... |
383 384 |
else submit_bio(dio->rw, bio); |
1da177e4c Linux-2.6.12-rc2 |
385 |
|
eb28be2b4 direct-io: separa... |
386 387 388 |
sdio->bio = NULL; sdio->boundary = 0; sdio->logical_offset_in_bio = 0; |
1da177e4c Linux-2.6.12-rc2 |
389 390 391 392 393 |
} /* * Release any resources in case of a failure */ |
ba253fbf6 direct-io: inline... |
394 |
static inline void dio_cleanup(struct dio *dio, struct dio_submit *sdio) |
1da177e4c Linux-2.6.12-rc2 |
395 |
{ |
7b2c99d15 new helper: iov_i... |
396 397 |
while (sdio->head < sdio->tail) page_cache_release(dio->pages[sdio->head++]); |
1da177e4c Linux-2.6.12-rc2 |
398 399 400 |
} /* |
0273201e6 [PATCH] dio: form... |
401 402 403 404 |
* 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 Linux-2.6.12-rc2 |
405 406 407 408 |
*/ static struct bio *dio_await_one(struct dio *dio) { unsigned long flags; |
0273201e6 [PATCH] dio: form... |
409 |
struct bio *bio = NULL; |
1da177e4c Linux-2.6.12-rc2 |
410 411 |
spin_lock_irqsave(&dio->bio_lock, flags); |
5eb6c7a2a [PATCH] dio: lock... |
412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 |
/* * 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); io_schedule(); /* wake up sets us TASK_RUNNING */ spin_lock_irqsave(&dio->bio_lock, flags); dio->waiter = NULL; |
1da177e4c Linux-2.6.12-rc2 |
427 |
} |
0273201e6 [PATCH] dio: form... |
428 429 430 431 |
if (dio->bio_list) { bio = dio->bio_list; dio->bio_list = bio->bi_private; } |
1da177e4c Linux-2.6.12-rc2 |
432 433 434 435 436 437 438 439 440 |
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 block: Convert so... |
441 442 |
struct bio_vec *bvec; unsigned i; |
9b81c8423 block: don't acce... |
443 |
int err; |
1da177e4c Linux-2.6.12-rc2 |
444 |
|
4246a0b63 block: add a bi_e... |
445 |
if (bio->bi_error) |
174e27c60 [PATCH] direct-io... |
446 |
dio->io_error = -EIO; |
1da177e4c Linux-2.6.12-rc2 |
447 448 449 |
if (dio->is_async && dio->rw == READ) { bio_check_pages_dirty(bio); /* transfers ownership */ |
9b81c8423 block: don't acce... |
450 |
err = bio->bi_error; |
1da177e4c Linux-2.6.12-rc2 |
451 |
} else { |
cb34e057a block: Convert so... |
452 453 |
bio_for_each_segment_all(bvec, bio, i) { struct page *page = bvec->bv_page; |
1da177e4c Linux-2.6.12-rc2 |
454 455 456 457 458 |
if (dio->rw == READ && !PageCompound(page)) set_page_dirty_lock(page); page_cache_release(page); } |
9b81c8423 block: don't acce... |
459 |
err = bio->bi_error; |
1da177e4c Linux-2.6.12-rc2 |
460 461 |
bio_put(bio); } |
9b81c8423 block: don't acce... |
462 |
return err; |
1da177e4c Linux-2.6.12-rc2 |
463 464 465 |
} /* |
0273201e6 [PATCH] dio: form... |
466 467 468 |
* 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 Fix misspellings ... |
469 |
* errors are propagated through dio->io_error and should be propagated via |
0273201e6 [PATCH] dio: form... |
470 |
* dio_complete(). |
1da177e4c Linux-2.6.12-rc2 |
471 |
*/ |
6d544bb4d [PATCH] dio: cent... |
472 |
static void dio_await_completion(struct dio *dio) |
1da177e4c Linux-2.6.12-rc2 |
473 |
{ |
0273201e6 [PATCH] dio: form... |
474 475 476 477 478 479 |
struct bio *bio; do { bio = dio_await_one(dio); if (bio) dio_bio_complete(dio, bio); } while (bio); |
1da177e4c Linux-2.6.12-rc2 |
480 481 482 483 484 485 486 487 488 |
} /* * 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 direct-io: inline... |
489 |
static inline int dio_bio_reap(struct dio *dio, struct dio_submit *sdio) |
1da177e4c Linux-2.6.12-rc2 |
490 491 |
{ int ret = 0; |
eb28be2b4 direct-io: separa... |
492 |
if (sdio->reap_counter++ >= 64) { |
1da177e4c Linux-2.6.12-rc2 |
493 494 495 496 497 498 499 500 501 502 503 504 505 |
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 direct-io: separa... |
506 |
sdio->reap_counter = 0; |
1da177e4c Linux-2.6.12-rc2 |
507 508 509 510 511 |
} return ret; } /* |
7b7a8665e direct-io: Implem... |
512 513 514 515 516 517 518 |
* 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 direct-io: Use re... |
519 |
struct workqueue_struct *old; |
7b7a8665e direct-io: Implem... |
520 521 522 523 524 525 526 527 |
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 direct-io: Use re... |
528 |
old = cmpxchg(&sb->s_dio_done_wq, NULL, wq); |
7b7a8665e direct-io: Implem... |
529 |
/* Someone created workqueue before us? Free ours... */ |
45150c43b direct-io: Use re... |
530 |
if (old) |
7b7a8665e direct-io: Implem... |
531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 |
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 Linux-2.6.12-rc2 |
548 |
* Call into the fs to map some more disk blocks. We record the current number |
eb28be2b4 direct-io: separa... |
549 |
* of available blocks at sdio->blocks_available. These are in units of the |
1da177e4c Linux-2.6.12-rc2 |
550 551 552 553 554 |
* 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 [PATCH] remove ->... |
555 |
* get_block() is passed the number of i_blkbits-sized blocks which direct_io |
1da177e4c Linux-2.6.12-rc2 |
556 557 558 559 560 561 562 563 564 565 566 567 |
* 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 [PATCH] remove ->... |
568 |
* block at a time - it will repeatedly call get_block() as it walks the hole. |
1da177e4c Linux-2.6.12-rc2 |
569 |
*/ |
18772641d direct-io: separa... |
570 571 |
static int get_more_blocks(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) |
1da177e4c Linux-2.6.12-rc2 |
572 573 |
{ int ret; |
1da177e4c Linux-2.6.12-rc2 |
574 |
sector_t fs_startblk; /* Into file, in filesystem-sized blocks */ |
ae55e1aaa fs/direct-io.c: c... |
575 |
sector_t fs_endblk; /* Into file, in filesystem-sized blocks */ |
1da177e4c Linux-2.6.12-rc2 |
576 |
unsigned long fs_count; /* Number of filesystem-sized blocks */ |
1da177e4c Linux-2.6.12-rc2 |
577 |
int create; |
ab73857e3 direct-io: don't ... |
578 |
unsigned int i_blkbits = sdio->blkbits + sdio->blkfactor; |
1da177e4c Linux-2.6.12-rc2 |
579 580 581 582 583 584 585 |
/* * 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 direct-io: separa... |
586 587 |
BUG_ON(sdio->block_in_file >= sdio->final_block_in_request); fs_startblk = sdio->block_in_file >> sdio->blkfactor; |
ae55e1aaa fs/direct-io.c: c... |
588 589 590 |
fs_endblk = (sdio->final_block_in_request - 1) >> sdio->blkfactor; fs_count = fs_endblk - fs_startblk + 1; |
1da177e4c Linux-2.6.12-rc2 |
591 |
|
3c674e742 Fixes a regressio... |
592 |
map_bh->b_state = 0; |
ab73857e3 direct-io: don't ... |
593 |
map_bh->b_size = fs_count << i_blkbits; |
3c674e742 Fixes a regressio... |
594 |
|
5fe878ae7 direct-io: cleanu... |
595 596 597 598 599 600 601 602 603 604 605 |
/* * 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 [PATCH] Kill PF_S... |
606 |
create = dio->rw & WRITE; |
5fe878ae7 direct-io: cleanu... |
607 |
if (dio->flags & DIO_SKIP_HOLES) { |
eb28be2b4 direct-io: separa... |
608 609 |
if (sdio->block_in_file < (i_size_read(dio->inode) >> sdio->blkbits)) |
1da177e4c Linux-2.6.12-rc2 |
610 |
create = 0; |
1da177e4c Linux-2.6.12-rc2 |
611 |
} |
3c674e742 Fixes a regressio... |
612 |
|
eb28be2b4 direct-io: separa... |
613 |
ret = (*sdio->get_block)(dio->inode, fs_startblk, |
1da177e4c Linux-2.6.12-rc2 |
614 |
map_bh, create); |
18772641d direct-io: separa... |
615 616 617 |
/* Store for completion */ dio->private = map_bh->b_private; |
7b7a8665e direct-io: Implem... |
618 619 620 |
if (ret == 0 && buffer_defer_completion(map_bh)) ret = dio_set_defer_completion(dio); |
1da177e4c Linux-2.6.12-rc2 |
621 622 623 624 625 626 627 |
} return ret; } /* * There is no bio. Make one now. */ |
ba253fbf6 direct-io: inline... |
628 629 |
static inline int dio_new_bio(struct dio *dio, struct dio_submit *sdio, sector_t start_sector, struct buffer_head *map_bh) |
1da177e4c Linux-2.6.12-rc2 |
630 631 632 |
{ sector_t sector; int ret, nr_pages; |
eb28be2b4 direct-io: separa... |
633 |
ret = dio_bio_reap(dio, sdio); |
1da177e4c Linux-2.6.12-rc2 |
634 635 |
if (ret) goto out; |
eb28be2b4 direct-io: separa... |
636 |
sector = start_sector << (sdio->blkbits - 9); |
b54ffb73c block: remove bio... |
637 |
nr_pages = min(sdio->pages_in_io, BIO_MAX_PAGES); |
1da177e4c Linux-2.6.12-rc2 |
638 |
BUG_ON(nr_pages <= 0); |
18772641d direct-io: separa... |
639 |
dio_bio_alloc(dio, sdio, map_bh->b_bdev, sector, nr_pages); |
eb28be2b4 direct-io: separa... |
640 |
sdio->boundary = 0; |
1da177e4c Linux-2.6.12-rc2 |
641 642 643 644 645 646 647 648 649 650 651 |
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 direct-io: inline... |
652 |
static inline int dio_bio_add_page(struct dio_submit *sdio) |
1da177e4c Linux-2.6.12-rc2 |
653 654 |
{ int ret; |
eb28be2b4 direct-io: separa... |
655 656 657 |
ret = bio_add_page(sdio->bio, sdio->cur_page, sdio->cur_page_len, sdio->cur_page_offset); if (ret == sdio->cur_page_len) { |
1da177e4c Linux-2.6.12-rc2 |
658 659 660 |
/* * Decrement count only, if we are done with this page */ |
eb28be2b4 direct-io: separa... |
661 662 663 664 665 |
if ((sdio->cur_page_len + sdio->cur_page_offset) == PAGE_SIZE) sdio->pages_in_io--; page_cache_get(sdio->cur_page); sdio->final_block_in_bio = sdio->cur_page_block + (sdio->cur_page_len >> sdio->blkbits); |
1da177e4c Linux-2.6.12-rc2 |
666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 |
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 direct-io: inline... |
683 684 |
static inline int dio_send_cur_page(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) |
1da177e4c Linux-2.6.12-rc2 |
685 686 |
{ int ret = 0; |
eb28be2b4 direct-io: separa... |
687 688 689 |
if (sdio->bio) { loff_t cur_offset = sdio->cur_page_fs_offset; loff_t bio_next_offset = sdio->logical_offset_in_bio + |
4f024f379 block: Abstract o... |
690 |
sdio->bio->bi_iter.bi_size; |
c2c6ca417 direct-io: do not... |
691 |
|
1da177e4c Linux-2.6.12-rc2 |
692 |
/* |
c2c6ca417 direct-io: do not... |
693 694 |
* See whether this new request is contiguous with the old. * |
f0940cee2 dio: fix typos in... |
695 696 |
* Btrfs cannot handle having logically non-contiguous requests * submitted. For example if you have |
c2c6ca417 direct-io: do not... |
697 698 |
* * Logical: [0-4095][HOLE][8192-12287] |
f0940cee2 dio: fix typos in... |
699 |
* Physical: [0-4095] [4096-8191] |
c2c6ca417 direct-io: do not... |
700 701 702 703 704 |
* * 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 Linux-2.6.12-rc2 |
705 |
*/ |
eb28be2b4 direct-io: separa... |
706 |
if (sdio->final_block_in_bio != sdio->cur_page_block || |
c2c6ca417 direct-io: do not... |
707 |
cur_offset != bio_next_offset) |
eb28be2b4 direct-io: separa... |
708 |
dio_bio_submit(dio, sdio); |
1da177e4c Linux-2.6.12-rc2 |
709 |
} |
eb28be2b4 direct-io: separa... |
710 |
if (sdio->bio == NULL) { |
18772641d direct-io: separa... |
711 |
ret = dio_new_bio(dio, sdio, sdio->cur_page_block, map_bh); |
1da177e4c Linux-2.6.12-rc2 |
712 713 714 |
if (ret) goto out; } |
eb28be2b4 direct-io: separa... |
715 716 |
if (dio_bio_add_page(sdio) != 0) { dio_bio_submit(dio, sdio); |
18772641d direct-io: separa... |
717 |
ret = dio_new_bio(dio, sdio, sdio->cur_page_block, map_bh); |
1da177e4c Linux-2.6.12-rc2 |
718 |
if (ret == 0) { |
eb28be2b4 direct-io: separa... |
719 |
ret = dio_bio_add_page(sdio); |
1da177e4c Linux-2.6.12-rc2 |
720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 |
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 direct-io: inline... |
744 |
static inline int |
eb28be2b4 direct-io: separa... |
745 |
submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page, |
18772641d direct-io: separa... |
746 747 |
unsigned offset, unsigned len, sector_t blocknr, struct buffer_head *map_bh) |
1da177e4c Linux-2.6.12-rc2 |
748 749 |
{ int ret = 0; |
98c4d57de [PATCH] io-accoun... |
750 751 752 753 754 755 |
if (dio->rw & WRITE) { /* * Read accounting is performed in submit_bio() */ task_io_account_write(len); } |
1da177e4c Linux-2.6.12-rc2 |
756 757 758 |
/* * Can we just grow the current page's presence in the dio? */ |
eb28be2b4 direct-io: separa... |
759 760 761 762 763 |
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 Linux-2.6.12-rc2 |
764 765 766 767 768 769 |
goto out; } /* * If there's a deferred page already there then send it. */ |
eb28be2b4 direct-io: separa... |
770 |
if (sdio->cur_page) { |
18772641d direct-io: separa... |
771 |
ret = dio_send_cur_page(dio, sdio, map_bh); |
eb28be2b4 direct-io: separa... |
772 773 |
page_cache_release(sdio->cur_page); sdio->cur_page = NULL; |
1da177e4c Linux-2.6.12-rc2 |
774 |
if (ret) |
b1058b981 direct-io: submit... |
775 |
return ret; |
1da177e4c Linux-2.6.12-rc2 |
776 777 778 |
} page_cache_get(page); /* It is in dio */ |
eb28be2b4 direct-io: separa... |
779 780 781 782 783 |
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 Linux-2.6.12-rc2 |
784 |
out: |
b1058b981 direct-io: submit... |
785 786 787 788 789 790 791 792 793 794 |
/* * 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); page_cache_release(sdio->cur_page); sdio->cur_page = NULL; } |
1da177e4c Linux-2.6.12-rc2 |
795 796 797 798 799 800 801 802 |
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 direct-io: separa... |
803 |
static void clean_blockdev_aliases(struct dio *dio, struct buffer_head *map_bh) |
1da177e4c Linux-2.6.12-rc2 |
804 805 806 |
{ unsigned i; unsigned nblocks; |
18772641d direct-io: separa... |
807 |
nblocks = map_bh->b_size >> dio->inode->i_blkbits; |
1da177e4c Linux-2.6.12-rc2 |
808 809 |
for (i = 0; i < nblocks; i++) { |
18772641d direct-io: separa... |
810 811 |
unmap_underlying_metadata(map_bh->b_bdev, map_bh->b_blocknr + i); |
1da177e4c Linux-2.6.12-rc2 |
812 813 814 815 816 817 818 819 820 821 822 823 |
} } /* * 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 direct-io: inline... |
824 825 |
static inline void dio_zero_block(struct dio *dio, struct dio_submit *sdio, int end, struct buffer_head *map_bh) |
1da177e4c Linux-2.6.12-rc2 |
826 827 828 829 830 |
{ unsigned dio_blocks_per_fs_block; unsigned this_chunk_blocks; /* In dio_blocks */ unsigned this_chunk_bytes; struct page *page; |
eb28be2b4 direct-io: separa... |
831 |
sdio->start_zero_done = 1; |
18772641d direct-io: separa... |
832 |
if (!sdio->blkfactor || !buffer_new(map_bh)) |
1da177e4c Linux-2.6.12-rc2 |
833 |
return; |
eb28be2b4 direct-io: separa... |
834 835 |
dio_blocks_per_fs_block = 1 << sdio->blkfactor; this_chunk_blocks = sdio->block_in_file & (dio_blocks_per_fs_block - 1); |
1da177e4c Linux-2.6.12-rc2 |
836 837 838 839 840 841 842 843 844 845 |
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 direct-io: separa... |
846 |
this_chunk_bytes = this_chunk_blocks << sdio->blkbits; |
1da177e4c Linux-2.6.12-rc2 |
847 |
|
557ed1fa2 remove ZERO_PAGE |
848 |
page = ZERO_PAGE(0); |
eb28be2b4 direct-io: separa... |
849 |
if (submit_page_section(dio, sdio, page, 0, this_chunk_bytes, |
18772641d direct-io: separa... |
850 |
sdio->next_block_for_io, map_bh)) |
1da177e4c Linux-2.6.12-rc2 |
851 |
return; |
eb28be2b4 direct-io: separa... |
852 |
sdio->next_block_for_io += this_chunk_blocks; |
1da177e4c Linux-2.6.12-rc2 |
853 854 855 856 857 858 859 860 861 862 863 |
} /* * 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 [PATCH] remove ->... |
864 |
* So what we do is to permit the ->get_block function to populate bh.b_size |
1da177e4c Linux-2.6.12-rc2 |
865 866 867 |
* 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 [PATCH] remove ->... |
868 |
* it should set b_size to PAGE_SIZE or more inside get_block(). This gives |
1da177e4c Linux-2.6.12-rc2 |
869 870 |
* fine alignment but still allows this function to work in PAGE_SIZE units. */ |
18772641d direct-io: separa... |
871 872 |
static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) |
1da177e4c Linux-2.6.12-rc2 |
873 |
{ |
eb28be2b4 direct-io: separa... |
874 |
const unsigned blkbits = sdio->blkbits; |
1da177e4c Linux-2.6.12-rc2 |
875 |
int ret = 0; |
eb28be2b4 direct-io: separa... |
876 |
while (sdio->block_in_file < sdio->final_block_in_request) { |
7b2c99d15 new helper: iov_i... |
877 878 |
struct page *page; size_t from, to; |
6fcc5420b direct-io: fix un... |
879 880 |
page = dio_get_page(dio, sdio); |
1da177e4c Linux-2.6.12-rc2 |
881 882 883 884 |
if (IS_ERR(page)) { ret = PTR_ERR(page); goto out; } |
6fcc5420b direct-io: fix un... |
885 886 887 |
from = sdio->head ? 0 : sdio->from; to = (sdio->head == sdio->tail - 1) ? sdio->to : PAGE_SIZE; sdio->head++; |
1da177e4c Linux-2.6.12-rc2 |
888 |
|
7b2c99d15 new helper: iov_i... |
889 |
while (from < to) { |
1da177e4c Linux-2.6.12-rc2 |
890 891 892 |
unsigned this_chunk_bytes; /* # of bytes mapped */ unsigned this_chunk_blocks; /* # of blocks */ unsigned u; |
eb28be2b4 direct-io: separa... |
893 |
if (sdio->blocks_available == 0) { |
1da177e4c Linux-2.6.12-rc2 |
894 895 896 897 898 |
/* * Need to go and map some more disk */ unsigned long blkmask; unsigned long dio_remainder; |
18772641d direct-io: separa... |
899 |
ret = get_more_blocks(dio, sdio, map_bh); |
1da177e4c Linux-2.6.12-rc2 |
900 901 902 903 904 905 |
if (ret) { page_cache_release(page); goto out; } if (!buffer_mapped(map_bh)) goto do_holes; |
eb28be2b4 direct-io: separa... |
906 907 908 909 |
sdio->blocks_available = map_bh->b_size >> sdio->blkbits; sdio->next_block_for_io = map_bh->b_blocknr << sdio->blkfactor; |
1da177e4c Linux-2.6.12-rc2 |
910 |
if (buffer_new(map_bh)) |
18772641d direct-io: separa... |
911 |
clean_blockdev_aliases(dio, map_bh); |
1da177e4c Linux-2.6.12-rc2 |
912 |
|
eb28be2b4 direct-io: separa... |
913 |
if (!sdio->blkfactor) |
1da177e4c Linux-2.6.12-rc2 |
914 |
goto do_holes; |
eb28be2b4 direct-io: separa... |
915 916 |
blkmask = (1 << sdio->blkfactor) - 1; dio_remainder = (sdio->block_in_file & blkmask); |
1da177e4c Linux-2.6.12-rc2 |
917 918 919 920 921 922 923 924 925 926 927 928 929 |
/* * 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 direct-io: separa... |
930 931 |
sdio->next_block_for_io += dio_remainder; sdio->blocks_available -= dio_remainder; |
1da177e4c Linux-2.6.12-rc2 |
932 933 934 935 |
} do_holes: /* Handle holes */ if (!buffer_mapped(map_bh)) { |
35dc8161d [PATCH] fix O_DIR... |
936 |
loff_t i_size_aligned; |
1da177e4c Linux-2.6.12-rc2 |
937 938 |
/* AKPM: eargh, -ENOTBLK is a hack */ |
b31dc66a5 [PATCH] Kill PF_S... |
939 |
if (dio->rw & WRITE) { |
1da177e4c Linux-2.6.12-rc2 |
940 941 942 |
page_cache_release(page); return -ENOTBLK; } |
35dc8161d [PATCH] fix O_DIR... |
943 944 945 946 947 948 |
/* * 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 direct-io: separa... |
949 |
if (sdio->block_in_file >= |
35dc8161d [PATCH] fix O_DIR... |
950 |
i_size_aligned >> blkbits) { |
1da177e4c Linux-2.6.12-rc2 |
951 952 953 954 |
/* We hit eof */ page_cache_release(page); goto out; } |
7b2c99d15 new helper: iov_i... |
955 |
zero_user(page, from, 1 << blkbits); |
eb28be2b4 direct-io: separa... |
956 |
sdio->block_in_file++; |
7b2c99d15 new helper: iov_i... |
957 |
from += 1 << blkbits; |
3320c60b3 dio: take updatin... |
958 |
dio->result += 1 << blkbits; |
1da177e4c Linux-2.6.12-rc2 |
959 960 961 962 963 964 965 966 |
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 direct-io: separa... |
967 |
if (unlikely(sdio->blkfactor && !sdio->start_zero_done)) |
18772641d direct-io: separa... |
968 |
dio_zero_block(dio, sdio, 0, map_bh); |
1da177e4c Linux-2.6.12-rc2 |
969 970 971 972 973 |
/* * Work out, in this_chunk_blocks, how much disk we * can add to this page */ |
eb28be2b4 direct-io: separa... |
974 |
this_chunk_blocks = sdio->blocks_available; |
7b2c99d15 new helper: iov_i... |
975 |
u = (to - from) >> blkbits; |
1da177e4c Linux-2.6.12-rc2 |
976 977 |
if (this_chunk_blocks > u) this_chunk_blocks = u; |
eb28be2b4 direct-io: separa... |
978 |
u = sdio->final_block_in_request - sdio->block_in_file; |
1da177e4c Linux-2.6.12-rc2 |
979 980 981 982 |
if (this_chunk_blocks > u) this_chunk_blocks = u; this_chunk_bytes = this_chunk_blocks << blkbits; BUG_ON(this_chunk_bytes == 0); |
092c8d46e direct-io: fix bo... |
983 984 |
if (this_chunk_blocks == sdio->blocks_available) sdio->boundary = buffer_boundary(map_bh); |
eb28be2b4 direct-io: separa... |
985 |
ret = submit_page_section(dio, sdio, page, |
7b2c99d15 new helper: iov_i... |
986 |
from, |
eb28be2b4 direct-io: separa... |
987 |
this_chunk_bytes, |
18772641d direct-io: separa... |
988 989 |
sdio->next_block_for_io, map_bh); |
1da177e4c Linux-2.6.12-rc2 |
990 991 992 993 |
if (ret) { page_cache_release(page); goto out; } |
eb28be2b4 direct-io: separa... |
994 |
sdio->next_block_for_io += this_chunk_blocks; |
1da177e4c Linux-2.6.12-rc2 |
995 |
|
eb28be2b4 direct-io: separa... |
996 |
sdio->block_in_file += this_chunk_blocks; |
7b2c99d15 new helper: iov_i... |
997 998 |
from += this_chunk_bytes; dio->result += this_chunk_bytes; |
eb28be2b4 direct-io: separa... |
999 |
sdio->blocks_available -= this_chunk_blocks; |
1da177e4c Linux-2.6.12-rc2 |
1000 |
next_block: |
eb28be2b4 direct-io: separa... |
1001 1002 |
BUG_ON(sdio->block_in_file > sdio->final_block_in_request); if (sdio->block_in_file == sdio->final_block_in_request) |
1da177e4c Linux-2.6.12-rc2 |
1003 1004 1005 1006 1007 |
break; } /* Drop the ref which was taken in get_user_pages() */ page_cache_release(page); |
1da177e4c Linux-2.6.12-rc2 |
1008 1009 1010 1011 |
} out: return ret; } |
847cc6371 direct-io: merge ... |
1012 |
static inline int drop_refcount(struct dio *dio) |
1da177e4c Linux-2.6.12-rc2 |
1013 |
{ |
847cc6371 direct-io: merge ... |
1014 |
int ret2; |
5eb6c7a2a [PATCH] dio: lock... |
1015 |
unsigned long flags; |
1da177e4c Linux-2.6.12-rc2 |
1016 |
|
8459d86af [PATCH] dio: only... |
1017 1018 |
/* * Sync will always be dropping the final ref and completing the |
5eb6c7a2a [PATCH] dio: lock... |
1019 1020 1021 |
* 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 fs: split generic... |
1022 |
* return code that the caller will hand to ->complete(). |
5eb6c7a2a [PATCH] dio: lock... |
1023 1024 1025 1026 |
* * 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 [PATCH] dio: only... |
1027 |
*/ |
5eb6c7a2a [PATCH] dio: lock... |
1028 1029 1030 |
spin_lock_irqsave(&dio->bio_lock, flags); ret2 = --dio->refcount; spin_unlock_irqrestore(&dio->bio_lock, flags); |
847cc6371 direct-io: merge ... |
1031 |
return ret2; |
1da177e4c Linux-2.6.12-rc2 |
1032 |
} |
eafdc7d19 sort out blockdev... |
1033 1034 1035 1036 1037 1038 1039 1040 1041 |
/* * 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 sort out blockdev... |
1042 1043 1044 |
* - 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 fs: always mainta... |
1045 1046 1047 1048 1049 1050 1051 |
* * 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 direct-io: inline... |
1052 1053 1054 1055 1056 |
* * 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 sort out blockdev... |
1057 |
*/ |
65dd2aa90 dio: optimize cac... |
1058 |
static inline ssize_t |
17f8c842d Remove rw from {,... |
1059 1060 1061 1062 |
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 Linux-2.6.12-rc2 |
1063 |
{ |
ab73857e3 direct-io: don't ... |
1064 1065 |
unsigned i_blkbits = ACCESS_ONCE(inode->i_blkbits); unsigned blkbits = i_blkbits; |
1da177e4c Linux-2.6.12-rc2 |
1066 1067 |
unsigned blocksize_mask = (1 << blkbits) - 1; ssize_t retval = -EINVAL; |
af4364727 direct-io: fix AI... |
1068 1069 |
size_t count = iov_iter_count(iter); loff_t end = offset + count; |
1da177e4c Linux-2.6.12-rc2 |
1070 |
struct dio *dio; |
eb28be2b4 direct-io: separa... |
1071 |
struct dio_submit sdio = { 0, }; |
847cc6371 direct-io: merge ... |
1072 |
struct buffer_head map_bh = { 0, }; |
647d1e4c5 block: move down ... |
1073 |
struct blk_plug plug; |
886a39115 new primitive: io... |
1074 |
unsigned long align = offset | iov_iter_alignment(iter); |
1da177e4c Linux-2.6.12-rc2 |
1075 |
|
65dd2aa90 dio: optimize cac... |
1076 1077 1078 1079 |
/* * Avoid references to bdev if not absolutely needed to give * the early prefetch in the caller enough time. */ |
1da177e4c Linux-2.6.12-rc2 |
1080 |
|
886a39115 new primitive: io... |
1081 |
if (align & blocksize_mask) { |
1da177e4c Linux-2.6.12-rc2 |
1082 |
if (bdev) |
65dd2aa90 dio: optimize cac... |
1083 |
blkbits = blksize_bits(bdev_logical_block_size(bdev)); |
1da177e4c Linux-2.6.12-rc2 |
1084 |
blocksize_mask = (1 << blkbits) - 1; |
886a39115 new primitive: io... |
1085 |
if (align & blocksize_mask) |
1da177e4c Linux-2.6.12-rc2 |
1086 1087 |
goto out; } |
f9b5570d7 fs: simplify hand... |
1088 |
/* watch out for a 0 len io from a tricksy fs */ |
17f8c842d Remove rw from {,... |
1089 |
if (iov_iter_rw(iter) == READ && !iov_iter_count(iter)) |
f9b5570d7 fs: simplify hand... |
1090 |
return 0; |
6e8267f53 direct-io: use a ... |
1091 |
dio = kmem_cache_alloc(dio_cache, GFP_KERNEL); |
1da177e4c Linux-2.6.12-rc2 |
1092 1093 1094 |
retval = -ENOMEM; if (!dio) goto out; |
23aee091d dio: don't zero o... |
1095 1096 1097 1098 1099 1100 |
/* * 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 Linux-2.6.12-rc2 |
1101 |
|
5fe878ae7 direct-io: cleanu... |
1102 1103 |
dio->flags = flags; if (dio->flags & DIO_LOCKING) { |
17f8c842d Remove rw from {,... |
1104 |
if (iov_iter_rw(iter) == READ) { |
5fe878ae7 direct-io: cleanu... |
1105 1106 |
struct address_space *mapping = iocb->ki_filp->f_mapping; |
1da177e4c Linux-2.6.12-rc2 |
1107 |
|
5fe878ae7 direct-io: cleanu... |
1108 1109 |
/* will be released by direct_io_worker */ mutex_lock(&inode->i_mutex); |
1da177e4c Linux-2.6.12-rc2 |
1110 1111 1112 1113 |
retval = filemap_write_and_wait_range(mapping, offset, end - 1); if (retval) { |
5fe878ae7 direct-io: cleanu... |
1114 |
mutex_unlock(&inode->i_mutex); |
6e8267f53 direct-io: use a ... |
1115 |
kmem_cache_free(dio_cache, dio); |
1da177e4c Linux-2.6.12-rc2 |
1116 1117 |
goto out; } |
1da177e4c Linux-2.6.12-rc2 |
1118 |
} |
1da177e4c Linux-2.6.12-rc2 |
1119 1120 1121 |
} /* |
603925737 direct-io: add fl... |
1122 1123 1124 1125 |
* 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 Linux-2.6.12-rc2 |
1126 |
*/ |
603925737 direct-io: add fl... |
1127 1128 1129 |
if (is_sync_kiocb(iocb)) dio->is_async = false; else if (!(dio->flags & DIO_ASYNC_EXTEND) && |
17f8c842d Remove rw from {,... |
1130 |
iov_iter_rw(iter) == WRITE && end > i_size_read(inode)) |
603925737 direct-io: add fl... |
1131 1132 1133 |
dio->is_async = false; else dio->is_async = true; |
847cc6371 direct-io: merge ... |
1134 |
dio->inode = inode; |
17f8c842d Remove rw from {,... |
1135 |
dio->rw = iov_iter_rw(iter) == WRITE ? WRITE_ODIRECT : READ; |
02afc27fa direct-io: Handle... |
1136 1137 1138 1139 1140 |
/* * For AIO O_(D)SYNC writes we need to defer completions to a workqueue * so that we can call ->fsync. */ |
17f8c842d Remove rw from {,... |
1141 |
if (dio->is_async && iov_iter_rw(iter) == WRITE && |
02afc27fa direct-io: Handle... |
1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 |
((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 direct-io: only i... |
1158 1159 |
if (!(dio->flags & DIO_SKIP_DIO_COUNT)) inode_dio_begin(inode); |
02afc27fa direct-io: Handle... |
1160 1161 |
retval = 0; |
847cc6371 direct-io: merge ... |
1162 |
sdio.blkbits = blkbits; |
ab73857e3 direct-io: don't ... |
1163 |
sdio.blkfactor = i_blkbits - blkbits; |
847cc6371 direct-io: merge ... |
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 |
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; dio->i_size = i_size_read(inode); spin_lock_init(&dio->bio_lock); dio->refcount = 1; |
7b2c99d15 new helper: iov_i... |
1177 1178 1179 |
sdio.iter = iter; sdio.final_block_in_request = (offset + iov_iter_count(iter)) >> blkbits; |
847cc6371 direct-io: merge ... |
1180 1181 1182 1183 1184 1185 |
/* * 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 new helper: iov_i... |
1186 |
sdio.pages_in_io += iov_iter_npages(iter, INT_MAX); |
847cc6371 direct-io: merge ... |
1187 |
|
647d1e4c5 block: move down ... |
1188 |
blk_start_plug(&plug); |
7b2c99d15 new helper: iov_i... |
1189 1190 1191 |
retval = do_direct_IO(dio, &sdio, &map_bh); if (retval) dio_cleanup(dio, &sdio); |
847cc6371 direct-io: merge ... |
1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 |
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; page_cache_release(sdio.cur_page); sdio.cur_page = NULL; } if (sdio.bio) dio_bio_submit(dio, &sdio); |
647d1e4c5 block: move down ... |
1217 |
blk_finish_plug(&plug); |
847cc6371 direct-io: merge ... |
1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 |
/* * 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 Remove rw from {,... |
1229 |
if (iov_iter_rw(iter) == READ && (dio->flags & DIO_LOCKING)) |
847cc6371 direct-io: merge ... |
1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 |
mutex_unlock(&dio->inode->i_mutex); /* * 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 Remove rw from {,... |
1241 |
(iov_iter_rw(iter) == READ || dio->result == count)) |
847cc6371 direct-io: merge ... |
1242 |
retval = -EIOCBQUEUED; |
af4364727 direct-io: fix AI... |
1243 |
else |
847cc6371 direct-io: merge ... |
1244 1245 1246 1247 |
dio_await_completion(dio); if (drop_refcount(dio) == 0) { retval = dio_complete(dio, offset, retval, false); |
847cc6371 direct-io: merge ... |
1248 1249 |
} else BUG_ON(retval != -EIOCBQUEUED); |
1da177e4c Linux-2.6.12-rc2 |
1250 |
|
7bb46a673 fs: introduce new... |
1251 1252 1253 |
out: return retval; } |
65dd2aa90 dio: optimize cac... |
1254 |
|
17f8c842d Remove rw from {,... |
1255 1256 1257 1258 1259 |
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 dio: optimize cac... |
1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 |
{ /* * 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 Remove rw from {,... |
1272 1273 |
return do_blockdev_direct_IO(iocb, inode, bdev, iter, offset, get_block, end_io, submit_io, flags); |
65dd2aa90 dio: optimize cac... |
1274 |
} |
1da177e4c Linux-2.6.12-rc2 |
1275 |
EXPORT_SYMBOL(__blockdev_direct_IO); |
6e8267f53 direct-io: use a ... |
1276 1277 1278 1279 1280 1281 1282 |
static __init int dio_init(void) { dio_cache = KMEM_CACHE(dio, SLAB_PANIC); return 0; } module_init(dio_init) |