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,
116 gfp_t gfp_mask, unsigned long flags)
121 unsigned bdev_logical_block_size(struct block_device *bdev)
127 ret = fstat(bdev->bd_fd, &statbuf);
130 if (!S_ISBLK(statbuf.st_mode))
131 return statbuf.st_blksize >> 9;
133 ret = ioctl(bdev->bd_fd, BLKPBSZGET, &blksize);
139 sector_t get_capacity(struct gendisk *disk)
141 struct block_device *bdev =
142 container_of(disk, struct block_device, __bd_disk);
147 ret = fstat(bdev->bd_fd, &statbuf);
150 if (!S_ISBLK(statbuf.st_mode))
151 return statbuf.st_size >> 9;
153 ret = ioctl(bdev->bd_fd, BLKGETSIZE64, &bytes);
159 void blkdev_put(struct block_device *bdev, fmode_t mode)
161 fdatasync(bdev->bd_fd);
162 close(bdev->bd_sync_fd);
167 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
170 struct block_device *bdev;
171 int fd, sync_fd, flags = O_DIRECT;
173 if ((mode & (FMODE_READ|FMODE_WRITE)) == (FMODE_READ|FMODE_WRITE))
175 else if (mode & FMODE_READ)
177 else if (mode & FMODE_WRITE)
181 /* using O_EXCL doesn't work with opening twice for an O_SYNC fd: */
182 if (mode & FMODE_EXCL)
186 fd = open(path, flags);
188 return ERR_PTR(-errno);
190 sync_fd = open(path, flags|O_SYNC);
194 return ERR_PTR(-errno);
197 bdev = malloc(sizeof(*bdev));
198 memset(bdev, 0, sizeof(*bdev));
200 strncpy(bdev->name, path, sizeof(bdev->name));
201 bdev->name[sizeof(bdev->name) - 1] = '\0';
203 bdev->bd_dev = xfstat(fd).st_rdev;
205 bdev->bd_sync_fd = sync_fd;
206 bdev->bd_holder = holder;
207 bdev->bd_disk = &bdev->__bd_disk;
208 bdev->bd_disk->bdi = &bdev->bd_disk->__bdi;
209 bdev->queue.backing_dev_info = bdev->bd_disk->bdi;
214 void bdput(struct block_device *bdev)
219 int lookup_bdev(const char *path, dev_t *dev)
224 static void io_fallback(void)
227 if (fops->init == NULL)
228 die("no fallback possible, something is very wrong");
232 static void sync_check(struct bio *bio, int ret)
234 if (ret != bio->bi_iter.bi_size) {
235 die("IO error: %s\n", strerror(-ret));
238 if (bio->bi_opf & REQ_FUA) {
239 ret = fdatasync(bio->bi_bdev->bd_fd);
241 die("fsync error: %s\n", strerror(-ret));
246 static void sync_init(void) {}
248 static void sync_cleanup(void)
254 static void sync_read(struct bio *bio, struct iovec * iov, unsigned i)
257 int fd = bio->bi_opf & REQ_FUA
258 ? bio->bi_bdev->bd_sync_fd
259 : bio->bi_bdev->bd_fd;
260 ssize_t ret = preadv(fd, iov, i, bio->bi_iter.bi_sector << 9);
261 sync_check(bio, ret);
264 static void sync_write(struct bio *bio, struct iovec * iov, unsigned i)
266 int fd = bio->bi_opf & REQ_FUA
267 ? bio->bi_bdev->bd_sync_fd
268 : bio->bi_bdev->bd_fd;
269 ssize_t ret = pwritev(fd, iov, i, bio->bi_iter.bi_sector << 9);
270 sync_check(bio, ret);
273 static int aio_completion_thread(void *arg)
275 struct io_event events[8], *ev;
280 ret = io_getevents(aio_ctx, 1, ARRAY_SIZE(events),
283 if (ret < 0 && ret == -EINTR)
286 die("io_getevents() error: %s", strerror(-ret));
288 for (ev = events; ev < events + ret; ev++) {
289 struct bio *bio = (struct bio *) ev->data;
291 /* This should only happen during blkdev_cleanup() */
293 BUG_ON(atomic_read(&running_requests) != 0);
298 if (ev->res != bio->bi_iter.bi_size)
299 bio->bi_status = BLK_STS_IOERR;
302 atomic_dec(&running_requests);
309 static struct task_struct *aio_task = NULL;
311 static void aio_init(void)
313 struct task_struct *p;
314 long err = io_setup(256, &aio_ctx);
316 p = kthread_run(aio_completion_thread, NULL, "aio_completion");
321 } else if (err == -ENOSYS) {
324 die("io_setup() error: %s", strerror(err));
328 static void aio_cleanup(void)
330 struct task_struct *p = NULL;
334 /* I mean, really?! IO_CMD_NOOP is even defined, but not implemented. */
338 die("pipe err: %s", strerror(ret));
340 /* Wake up the completion thread with spurious work. */
343 .aio_lio_opcode = IO_CMD_PWRITE,
344 .data = NULL, /* Signal to stop */
345 .aio_fildes = fds[1],
349 ret = io_submit(aio_ctx, 1, &iocbp);
351 die("io_submit cleanup err: %s", strerror(-ret));
353 ret = kthread_stop(p);
362 static void aio_op(struct bio *bio, struct iovec *iov, unsigned i, int opcode)
367 .aio_fildes = bio->bi_opf & REQ_FUA
368 ? bio->bi_bdev->bd_sync_fd
369 : bio->bi_bdev->bd_fd,
370 .aio_lio_opcode = opcode,
373 .u.c.offset = bio->bi_iter.bi_sector << 9,
377 atomic_inc(&running_requests);
378 ret = io_submit(aio_ctx, 1, &iocbp);
380 die("io_submit err: %s", strerror(-ret));
383 static void aio_read(struct bio *bio, struct iovec *iov, unsigned i)
385 aio_op(bio, iov, i, IO_CMD_PREADV);
388 static void aio_write(struct bio *bio, struct iovec * iov, unsigned i)
390 aio_op(bio, iov, i, IO_CMD_PWRITEV);
394 /* not implemented */
395 static void uring_init(void) {
399 struct fops fops_list[] = {
404 .cleanup = aio_cleanup,
409 .cleanup = sync_cleanup,
417 __attribute__((constructor(102)))
418 static void blkdev_init(void)
424 __attribute__((destructor(102)))
425 static void blkdev_cleanup(void)