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;
133 xioctl(bdev->bd_fd, BLKPBSZGET, &blksize);
137 sector_t get_capacity(struct gendisk *disk)
139 struct block_device *bdev =
140 container_of(disk, struct block_device, __bd_disk);
145 ret = fstat(bdev->bd_fd, &statbuf);
148 if (!S_ISBLK(statbuf.st_mode))
149 return statbuf.st_size >> 9;
151 ret = ioctl(bdev->bd_fd, BLKGETSIZE64, &bytes);
157 void blkdev_put(struct block_device *bdev, fmode_t mode)
159 fdatasync(bdev->bd_fd);
160 close(bdev->bd_sync_fd);
165 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
168 struct block_device *bdev;
169 int fd, sync_fd, buffered_fd, flags = 0;
171 if ((mode & (FMODE_READ|FMODE_WRITE)) == (FMODE_READ|FMODE_WRITE))
173 else if (mode & FMODE_READ)
175 else if (mode & FMODE_WRITE)
179 /* using O_EXCL doesn't work with opening twice for an O_SYNC fd: */
180 if (mode & FMODE_EXCL)
184 fd = open(path, flags|O_DIRECT);
186 return ERR_PTR(-errno);
188 sync_fd = xopen(path, flags|O_DIRECT|O_SYNC);
189 buffered_fd = xopen(path, flags);
191 bdev = malloc(sizeof(*bdev));
192 memset(bdev, 0, sizeof(*bdev));
194 strncpy(bdev->name, path, sizeof(bdev->name));
195 bdev->name[sizeof(bdev->name) - 1] = '\0';
197 bdev->bd_dev = xfstat(fd).st_rdev;
199 bdev->bd_sync_fd = sync_fd;
200 bdev->bd_buffered_fd = buffered_fd;
201 bdev->bd_holder = holder;
202 bdev->bd_disk = &bdev->__bd_disk;
203 bdev->bd_disk->bdi = &bdev->bd_disk->__bdi;
204 bdev->queue.backing_dev_info = bdev->bd_disk->bdi;
209 void bdput(struct block_device *bdev)
214 int lookup_bdev(const char *path, dev_t *dev)
219 static void io_fallback(void)
222 if (fops->init == NULL)
223 die("no fallback possible, something is very wrong");
227 static void sync_check(struct bio *bio, int ret)
229 if (ret != bio->bi_iter.bi_size) {
230 die("IO error: %s\n", strerror(-ret));
233 if (bio->bi_opf & REQ_FUA) {
234 ret = fdatasync(bio->bi_bdev->bd_fd);
236 die("fsync error: %s\n", strerror(-ret));
241 static void sync_init(void) {}
243 static void sync_cleanup(void)
249 static void sync_read(struct bio *bio, struct iovec * iov, unsigned i)
252 int fd = bio->bi_opf & REQ_FUA
253 ? bio->bi_bdev->bd_sync_fd
254 : bio->bi_bdev->bd_fd;
255 ssize_t ret = preadv(fd, iov, i, bio->bi_iter.bi_sector << 9);
256 sync_check(bio, ret);
259 static void sync_write(struct bio *bio, struct iovec * iov, unsigned i)
261 int fd = bio->bi_opf & REQ_FUA
262 ? bio->bi_bdev->bd_sync_fd
263 : bio->bi_bdev->bd_fd;
264 ssize_t ret = pwritev(fd, iov, i, bio->bi_iter.bi_sector << 9);
265 sync_check(bio, ret);
268 static int aio_completion_thread(void *arg)
270 struct io_event events[8], *ev;
275 ret = io_getevents(aio_ctx, 1, ARRAY_SIZE(events),
278 if (ret < 0 && ret == -EINTR)
281 die("io_getevents() error: %s", strerror(-ret));
283 for (ev = events; ev < events + ret; ev++) {
284 struct bio *bio = (struct bio *) ev->data;
286 /* This should only happen during blkdev_cleanup() */
288 BUG_ON(atomic_read(&running_requests) != 0);
293 if (ev->res != bio->bi_iter.bi_size)
294 bio->bi_status = BLK_STS_IOERR;
297 atomic_dec(&running_requests);
304 static struct task_struct *aio_task = NULL;
306 static void aio_init(void)
308 struct task_struct *p;
309 long err = io_setup(256, &aio_ctx);
311 p = kthread_run(aio_completion_thread, NULL, "aio_completion");
316 } else if (err == -ENOSYS) {
319 die("io_setup() error: %s", strerror(err));
323 static void aio_cleanup(void)
325 struct task_struct *p = NULL;
329 /* I mean, really?! IO_CMD_NOOP is even defined, but not implemented. */
333 die("pipe err: %s", strerror(ret));
335 /* Wake up the completion thread with spurious work. */
338 .aio_lio_opcode = IO_CMD_PWRITE,
339 .data = NULL, /* Signal to stop */
340 .aio_fildes = fds[1],
344 ret = io_submit(aio_ctx, 1, &iocbp);
346 die("io_submit cleanup err: %s", strerror(-ret));
348 ret = kthread_stop(p);
357 static void aio_op(struct bio *bio, struct iovec *iov, unsigned i, int opcode)
362 .aio_fildes = bio->bi_opf & REQ_FUA
363 ? bio->bi_bdev->bd_sync_fd
364 : bio->bi_bdev->bd_fd,
365 .aio_lio_opcode = opcode,
368 .u.c.offset = bio->bi_iter.bi_sector << 9,
372 atomic_inc(&running_requests);
373 ret = io_submit(aio_ctx, 1, &iocbp);
375 die("io_submit err: %s", strerror(-ret));
378 static void aio_read(struct bio *bio, struct iovec *iov, unsigned i)
380 aio_op(bio, iov, i, IO_CMD_PREADV);
383 static void aio_write(struct bio *bio, struct iovec * iov, unsigned i)
385 aio_op(bio, iov, i, IO_CMD_PWRITEV);
389 /* not implemented */
390 static void uring_init(void) {
394 struct fops fops_list[] = {
399 .cleanup = aio_cleanup,
404 .cleanup = sync_cleanup,
412 __attribute__((constructor(102)))
413 static void blkdev_init(void)
419 __attribute__((destructor(102)))
420 static void blkdev_cleanup(void)