diff options
| -rw-r--r-- | .gitignore | 2 | ||||
| -rw-r--r-- | src/bindfs.c | 75 | ||||
| -rw-r--r-- | tests/Makefile.am | 4 | ||||
| -rw-r--r-- | tests/odirect_read.c | 59 | ||||
| -rw-r--r-- | tests/odirect_write.c | 78 | ||||
| -rwxr-xr-x | tests/test_bindfs.rb | 22 |
6 files changed, 218 insertions, 22 deletions
@@ -35,6 +35,8 @@ Makefile src/bindfs tests/fcntl_locker +tests/odirect_read +tests/odirect_write tests/readdir_inode tests/test_dir_rewind tests/utimens_nofollow diff --git a/src/bindfs.c b/src/bindfs.c index fff3a16..3c430a1 100644 --- a/src/bindfs.c +++ b/src/bindfs.c @@ -72,11 +72,14 @@ #include <sys/xattr.h> #endif +#ifdef __linux__ #include <sys/mman.h> #include <fcntl.h> - -#ifdef __LINUX__ #include <linux/fs.h> // For BLKGETSIZE64 + +#ifndef O_DIRECT +#define O_DIRECT 00040000 /* direct disk access hint */ +#endif #endif #include <fuse.h> @@ -99,10 +102,6 @@ #define XATTR_APPLE_PREFIX "com.apple." #endif -#ifndef O_DIRECT -#define O_DIRECT 00040000 /* direct disk access hint */ -#endif - /* We pessimistically assume signed uid_t and gid_t in our overflow checks, mostly because supporting both cases would require a bunch more code. */ static const uid_t UID_T_MAX = ((1LL << (sizeof(uid_t)*8-1)) - 1); @@ -194,10 +193,14 @@ static struct Settings { int enable_lock_forwarding; int enable_ioctl; - + uid_t uid_offset; gid_t gid_offset; +#ifdef __linux__ + size_t linux_page_size; +#endif + } settings; @@ -227,6 +230,10 @@ static int apply_gid_offset(gid_t *gid); static int unapply_uid_offset(uid_t *uid); static int unapply_gid_offset(gid_t *gid); +#ifdef __linux__ +static size_t round_up_buffer_size_for_direct_io(size_t size); +#endif + /* FUSE callbacks */ static void *bindfs_init(); static void bindfs_destroy(void *private_data); @@ -388,15 +395,15 @@ static int getattr_common(const char *procpath, struct stat *stbuf) /* Block files as regular files. */ if (settings.block_devices_as_files && S_ISBLK(stbuf->st_mode)) { stbuf->st_mode ^= S_IFBLK | S_IFREG; // Flip both bits -#ifdef __LINUX__ + int fd = open(procpath, O_RDONLY); +#ifdef __linux__ uint64_t size; - ioctl(file, BLKGETSIZE64, &size); + ioctl(fd, BLKGETSIZE64, &size); stbuf->st_size = (off_t)size; if (stbuf->st_size < 0) { // Underflow return -EOVERFLOW; } #else - int fd = open(procpath, O_RDONLY); if (fd == -1) { return -errno; } @@ -592,6 +599,21 @@ static int unapply_gid_offset(gid_t *gid) { return 1; } +#ifdef __linux__ +static size_t round_up_buffer_size_for_direct_io(size_t size) +{ + // `man 2 open` says this should be block-size aligned and that there's no + // general way to determine this. Empirically the page size should be good enough. + // See also: Pull Request #74 + size_t page_size = settings.linux_page_size; + size_t rem = size % page_size; + if (rem == 0) { + return size; + } + return size - rem + page_size; +} +#endif + static void *bindfs_init() { @@ -1105,29 +1127,33 @@ static int bindfs_read(const char *path, char *buf, size_t size, off_t offset, int res; (void) path; - char * target_buf = buf; + char *target_buf = buf; if (settings.read_limiter) { rate_limiter_wait(settings.read_limiter, size); } - unsigned int page_size = sysconf(_SC_PAGESIZE); +#ifdef __linux__ + size_t mmap_size = 0; if (fi->flags & O_DIRECT) { - // allocate 512 bytes aligned buffer for direct io to work - target_buf = mmap(NULL, ((page_size - 1 + size) / page_size) * page_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0); + mmap_size = round_up_buffer_size_for_direct_io(size); + target_buf = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0); if (target_buf == MAP_FAILED) { return -ENOMEM; } } +#endif res = pread(fi->fh, target_buf, size, offset); if (res == -1) res = -errno; +#ifdef __linux__ if (target_buf != buf) { memcpy(buf, target_buf, size); - munmap(target_buf, ((page_size - 1 + size) / page_size) * page_size); + munmap(target_buf, mmap_size); } +#endif return res; } @@ -1137,29 +1163,33 @@ static int bindfs_write(const char *path, const char *buf, size_t size, { int res; (void) path; - char * source_buf = buf; + char *source_buf = (char*)buf; if (settings.write_limiter) { rate_limiter_wait(settings.write_limiter, size); } - unsigned int page_size = sysconf(_SC_PAGESIZE); +#ifdef __linux__ + size_t mmap_size = 0; if (fi->flags & O_DIRECT) { - // allocate 512 bytes aligned buffer for direct io to work - source_buf = mmap(NULL, ((page_size - 1 + size) / page_size) * page_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0); + mmap_size = round_up_buffer_size_for_direct_io(size); + source_buf = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0); if (source_buf == MAP_FAILED) { return -ENOMEM; } memcpy(source_buf, buf, size); } +#endif res = pwrite(fi->fh, source_buf, size, offset); if (res == -1) res = -errno; - if (source_buf != buf) { - munmap(source_buf, ((page_size - 1 + size) / page_size) * page_size); +#ifdef __linux__ + if (source_buf != buf) { + munmap(source_buf, mmap_size); } +#endif return res; } @@ -2076,6 +2106,9 @@ int main(int argc, char *argv[]) settings.enable_ioctl = 0; settings.uid_offset = 0; settings.gid_offset = 0; +#ifdef __linux__ + settings.linux_page_size = sysconf(_SC_PAGESIZE); +#endif atexit(&atexit_func); diff --git a/tests/Makefile.am b/tests/Makefile.am index 0a52108..742899b 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -4,10 +4,12 @@ UNAME_S := $(shell uname -s) AM_CPPFLAGS = ${my_CPPFLAGS} AM_CFLAGS = ${my_CFLAGS} -noinst_PROGRAMS = readdir_inode utimens_nofollow fcntl_locker test_dir_rewind +noinst_PROGRAMS = readdir_inode utimens_nofollow fcntl_locker odirect_read odirect_write test_dir_rewind readdir_inode_SOURCES = readdir_inode.c utimens_nofollow_SOURCES = utimens_nofollow.c fcntl_locker_SOURCES = fcntl_locker.c +odirect_read_SOURCES = odirect_read.c +odirect_write_SOURCES = odirect_write.c test_dir_rewind_SOURCES = test_dir_rewind.c TESTS = test_bindfs.rb diff --git a/tests/odirect_read.c b/tests/odirect_read.c new file mode 100644 index 0000000..8fc746b --- /dev/null +++ b/tests/odirect_read.c @@ -0,0 +1,59 @@ +#ifdef __linux__ + +#include <stdio.h> + +#include <sys/types.h> +#include <sys/stat.h> +#include <sys/mman.h> +#include <fcntl.h> +#include <unistd.h> + +#ifndef O_DIRECT +#define O_DIRECT 00040000 /* direct disk access hint */ +#endif + +int main(int argc, char** argv) { + if (argc != 2) { + fprintf(stderr, "Expected 1 argument: the file to read.\n"); + return 1; + } + + int fd = open(argv[1], O_RDONLY | O_DIRECT); + if (fd == -1) { + perror("failed to open file"); + return 1; + } + + const size_t buf_size = 4096; + unsigned char* buf = mmap(NULL, buf_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0); + if (buf == MAP_FAILED) { + perror("mmap failed"); + return 1; + } + + while (1) { + ssize_t amt_read = read(fd, buf, buf_size); + if (amt_read == 0) { + break; + } + if (amt_read == -1) { + perror("failed to read file"); + return 1; + } + fwrite(buf, 1, amt_read, stdout); + fflush(stdout); + } + + return 0; +} + +#else // __linux__ + +#include <stdio.h> + +int main() { + fprintf(stderr, "Not supported on this platform.\n"); + return 1; +} + +#endif // __linux__ diff --git a/tests/odirect_write.c b/tests/odirect_write.c new file mode 100644 index 0000000..1376a51 --- /dev/null +++ b/tests/odirect_write.c @@ -0,0 +1,78 @@ +#ifdef __linux__ + +#include <stdio.h> +#include <string.h> + +#include <sys/types.h> +#include <sys/stat.h> +#include <sys/mman.h> +#include <fcntl.h> +#include <unistd.h> + +#ifndef O_DIRECT +#define O_DIRECT 00040000 /* direct disk access hint */ +#endif + +int main(int argc, char** argv) { + if (argc != 2) { + fprintf(stderr, "Expected 1 argument: the file to read.\n"); + return 1; + } + + int fd = open(argv[1], O_WRONLY | O_CREAT | O_DIRECT, 0644); + if (fd == -1) { + perror("failed to open file"); + return 1; + } + + const size_t buf_size = 4096; + unsigned char* buf = mmap(NULL, buf_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0); + if (buf == MAP_FAILED) { + perror("mmap failed"); + return 1; + } + + size_t total_size = 0; + while (1) { + if (feof(stdin)) { + break; + } + + memset(buf, 0, buf_size); + size_t amt_read = fread(buf, 1, buf_size, stdin); + if (ferror(stdin)) { + perror("failed to read stdin"); + return 1; + } + if (amt_read == 0) { + continue; + } + + total_size += amt_read; + + ssize_t res = write(fd, buf, buf_size); + if (res == -1) { + perror("failed to write"); + return 1; + } + if (res != buf_size) { + // Too lazy to write a loop here unless it turns out to be necessary. + fprintf(stderr, "Failed to write exactly %ld bytes", amt_read); + } + } + + munmap(buf, buf_size); + + return 0; +} + +#else // __linux__ + +#include <stdio.h> + +int main() { + fprintf(stderr, "Not supported on this platform.\n"); + return 1; +} + +#endif // __linux__ diff --git a/tests/test_bindfs.rb b/tests/test_bindfs.rb index c5d41f6..9e5d381 100755 --- a/tests/test_bindfs.rb +++ b/tests/test_bindfs.rb @@ -775,6 +775,28 @@ if `uname`.strip == 'Linux' && end end +# Pull Request #74 +if `uname`.strip == 'Linux' + def odirect_data + ('abc' * 10000)[0...8192] + end + + testenv("", :title => "O_DIRECT reads") do + File.write("src/f", odirect_data) + read_data = `#{$tests_dir}/odirect_read mnt/f` + assert { $?.success? } + assert { read_data == odirect_data } + end + + testenv("", :title => "O_DIRECT writes") do + IO.popen("#{$tests_dir}/odirect_write mnt/f", "w") do |pipe| + pipe.write(odirect_data) + end + assert { $?.success? } + assert { File.read("src/f") == odirect_data } + end +end + if `uname`.strip != 'FreeBSD' # -o dev is not supported on FreeBSD root_testenv("-odev") do system("mknod mnt/zero c 1 5") |