13 #ifdef CONFIG_VALGRIND
14 #include <valgrind/memcheck.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/completion.h>
21 #include <linux/kthread.h>
23 #include "tools-util.h"
27 void (*cleanup)(void);
28 void (*read)(struct bio *bio, struct iovec * iov, unsigned i);
29 void (*write)(struct bio *bio, struct iovec * iov, unsigned i);
32 static struct fops *fops;
33 static io_context_t aio_ctx;
34 static atomic_t running_requests;
36 void generic_make_request(struct bio *bio)
39 struct bvec_iter iter;
44 if (bio->bi_opf & REQ_PREFLUSH) {
45 ret = fdatasync(bio->bi_bdev->bd_fd);
47 fprintf(stderr, "fsync error: %m\n");
48 bio->bi_status = BLK_STS_IOERR;
55 bio_for_each_segment(bv, bio, iter)
58 iov = alloca(sizeof(*iov) * i);
61 bio_for_each_segment(bv, bio, iter) {
62 void *start = page_address(bv.bv_page) + bv.bv_offset;
63 size_t len = bv.bv_len;
65 iov[i++] = (struct iovec) {
70 #ifdef CONFIG_VALGRIND
71 /* To be pedantic it should only be on IO completion. */
72 if (bio_op(bio) == REQ_OP_READ)
73 VALGRIND_MAKE_MEM_DEFINED(start, len);
77 switch (bio_op(bio)) {
79 fops->read(bio, iov, i);
82 fops->write(bio, iov, i);
85 ret = fsync(bio->bi_bdev->bd_fd);
87 die("fsync error: %m");
95 static void submit_bio_wait_endio(struct bio *bio)
97 complete(bio->bi_private);
100 int submit_bio_wait(struct bio *bio)
102 struct completion done;
104 init_completion(&done);
105 bio->bi_private = &done;
106 bio->bi_end_io = submit_bio_wait_endio;
107 bio->bi_opf |= REQ_SYNC;
109 wait_for_completion(&done);
111 return blk_status_to_errno(bio->bi_status);
114 int blkdev_issue_discard(struct block_device *bdev,
115 sector_t sector, sector_t nr_sects,
121 int blkdev_issue_zeroout(struct block_device *bdev,
122 sector_t sector, sector_t nr_sects,
123 gfp_t gfp_mask, unsigned flags)
125 /* Not yet implemented: */
129 unsigned bdev_logical_block_size(struct block_device *bdev)
135 ret = fstat(bdev->bd_fd, &statbuf);
138 if (!S_ISBLK(statbuf.st_mode))
139 return statbuf.st_blksize;
141 xioctl(bdev->bd_fd, BLKPBSZGET, &blksize);
145 sector_t get_capacity(struct gendisk *disk)
147 struct block_device *bdev =
148 container_of(disk, struct block_device, __bd_disk);
153 ret = fstat(bdev->bd_fd, &statbuf);
156 if (!S_ISBLK(statbuf.st_mode))
157 return statbuf.st_size >> 9;
159 ret = ioctl(bdev->bd_fd, BLKGETSIZE64, &bytes);
165 void blkdev_put(struct block_device *bdev, fmode_t mode)
167 fdatasync(bdev->bd_fd);
168 close(bdev->bd_sync_fd);
173 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
176 struct block_device *bdev;
177 int fd, sync_fd, buffered_fd, flags = 0;
179 if ((mode & (FMODE_READ|FMODE_WRITE)) == (FMODE_READ|FMODE_WRITE))
181 else if (mode & FMODE_READ)
183 else if (mode & FMODE_WRITE)
187 /* using O_EXCL doesn't work with opening twice for an O_SYNC fd: */
188 if (mode & FMODE_EXCL)
191 buffered_fd = open(path, flags);
193 return ERR_PTR(-errno);
195 fd = open(path, flags|O_DIRECT);
197 fd = dup(buffered_fd);
200 return ERR_PTR(-errno);
203 sync_fd = open(path, flags|O_DIRECT|O_SYNC);
205 sync_fd = open(path, flags|O_SYNC);
209 return ERR_PTR(-errno);
212 bdev = malloc(sizeof(*bdev));
213 memset(bdev, 0, sizeof(*bdev));
215 strncpy(bdev->name, path, sizeof(bdev->name));
216 bdev->name[sizeof(bdev->name) - 1] = '\0';
218 bdev->bd_dev = xfstat(fd).st_rdev;
220 bdev->bd_sync_fd = sync_fd;
221 bdev->bd_buffered_fd = buffered_fd;
222 bdev->bd_holder = holder;
223 bdev->bd_disk = &bdev->__bd_disk;
224 bdev->bd_disk->bdi = &bdev->bd_disk->__bdi;
225 bdev->queue.backing_dev_info = bdev->bd_disk->bdi;
230 void bdput(struct block_device *bdev)
235 int lookup_bdev(const char *path, dev_t *dev)
240 static void io_fallback(void)
243 if (fops->init == NULL)
244 die("no fallback possible, something is very wrong");
248 static void sync_check(struct bio *bio, int ret)
250 if (ret != bio->bi_iter.bi_size) {
251 die("IO error: %s\n", strerror(-ret));
254 if (bio->bi_opf & REQ_FUA) {
255 ret = fdatasync(bio->bi_bdev->bd_fd);
257 die("fsync error: %s\n", strerror(-ret));
262 static void sync_init(void) {}
264 static void sync_cleanup(void)
270 static void sync_read(struct bio *bio, struct iovec * iov, unsigned i)
273 int fd = bio->bi_opf & REQ_FUA
274 ? bio->bi_bdev->bd_sync_fd
275 : bio->bi_bdev->bd_fd;
276 ssize_t ret = preadv(fd, iov, i, bio->bi_iter.bi_sector << 9);
277 sync_check(bio, ret);
280 static void sync_write(struct bio *bio, struct iovec * iov, unsigned i)
282 int fd = bio->bi_opf & REQ_FUA
283 ? bio->bi_bdev->bd_sync_fd
284 : bio->bi_bdev->bd_fd;
285 ssize_t ret = pwritev(fd, iov, i, bio->bi_iter.bi_sector << 9);
286 sync_check(bio, ret);
289 static int aio_completion_thread(void *arg)
291 struct io_event events[8], *ev;
296 ret = io_getevents(aio_ctx, 1, ARRAY_SIZE(events),
299 if (ret < 0 && ret == -EINTR)
302 die("io_getevents() error: %s", strerror(-ret));
304 for (ev = events; ev < events + ret; ev++) {
305 struct bio *bio = (struct bio *) ev->data;
307 /* This should only happen during blkdev_cleanup() */
309 BUG_ON(atomic_read(&running_requests) != 0);
314 if (ev->res != bio->bi_iter.bi_size)
315 bio->bi_status = BLK_STS_IOERR;
318 atomic_dec(&running_requests);
325 static struct task_struct *aio_task = NULL;
327 static void aio_init(void)
329 struct task_struct *p;
330 long err = io_setup(256, &aio_ctx);
332 p = kthread_run(aio_completion_thread, NULL, "aio_completion");
337 } else if (err == -ENOSYS) {
340 die("io_setup() error: %s", strerror(err));
344 static void aio_cleanup(void)
346 struct task_struct *p = NULL;
350 /* I mean, really?! IO_CMD_NOOP is even defined, but not implemented. */
354 die("pipe err: %s", strerror(ret));
356 /* Wake up the completion thread with spurious work. */
359 .aio_lio_opcode = IO_CMD_PWRITE,
360 .data = NULL, /* Signal to stop */
361 .aio_fildes = fds[1],
365 ret = io_submit(aio_ctx, 1, &iocbp);
367 die("io_submit cleanup err: %s", strerror(-ret));
369 ret = kthread_stop(p);
378 static void aio_op(struct bio *bio, struct iovec *iov, unsigned i, int opcode)
383 .aio_fildes = bio->bi_opf & REQ_FUA
384 ? bio->bi_bdev->bd_sync_fd
385 : bio->bi_bdev->bd_fd,
386 .aio_lio_opcode = opcode,
389 .u.c.offset = bio->bi_iter.bi_sector << 9,
393 atomic_inc(&running_requests);
394 ret = io_submit(aio_ctx, 1, &iocbp);
396 die("io_submit err: %s", strerror(-ret));
399 static void aio_read(struct bio *bio, struct iovec *iov, unsigned i)
401 aio_op(bio, iov, i, IO_CMD_PREADV);
404 static void aio_write(struct bio *bio, struct iovec * iov, unsigned i)
406 aio_op(bio, iov, i, IO_CMD_PWRITEV);
410 /* not implemented */
411 static void uring_init(void) {
415 struct fops fops_list[] = {
420 .cleanup = aio_cleanup,
425 .cleanup = sync_cleanup,
433 __attribute__((constructor(102)))
434 static void blkdev_init(void)
440 __attribute__((destructor(102)))
441 static void blkdev_cleanup(void)