Merge branch 'rnou/fixcuda' into 'master' (4d66cd27) · Commits · hpc / gekkofs

CHANGELOG.md

+1 −0

Original line number	Diff line number	Diff line
		@@ -44,6 +44,7 @@ to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
		- SYS_lstat does not exists on some architectures, change to newfstatat ([!269](https://storage.bsc.es/gitlab/hpc/gekkofs/-/merge_requests/269))
		- We cannot use lstat directly as may cause a recursion call on libc interception.
		- Un/Packing order of directory entries in compressed format was incorrect ([!281](https://storage.bsc.es/gitlab/hpc/gekkofs/-/merge_requests/281))
		- Fix pytorch mmap ([!291](https://storage.bsc.es/gitlab/hpc/gekkofs/-/merge_requests/291))


		## [0.9.5] - 2025-08

include/client/gkfs_functions.hpp

+2 −0

Original line number	Diff line number	Diff line
		@@ -145,6 +145,7 @@ gkfs_read_ws(const gkfs::filemap::OpenFile& file, char* buf, size_t count,
		ssize_t
		gkfs_pread(int fd, void* buf, size_t count, off64_t offset);


		ssize_t
		gkfs_read(int fd, void* buf, size_t count);

		@@ -192,6 +193,7 @@ gkfs_munmap(void* addr, size_t length);
		int
		gkfs_msync(void* addr, size_t length, int flags);


		} // namespace gkfs::syscall

		// gkfs_getsingleserverdir is using extern "C" to demangle it for C usage

include/client/hooks.hpp

+9 −0

Original line number	Diff line number	Diff line
		@@ -135,6 +135,15 @@ ssize_t
		hook_pwritev(unsigned long fd, const struct iovec* iov, unsigned long iovcnt,
		unsigned long pos_l, unsigned long pos_h);

		ssize_t
		hook_sendfile(int out_fd, int in_fd, off_t* offset, size_t count);

		#ifdef SYS_copy_file_range
		ssize_t
		hook_copy_file_range(int fd_in, loff_t* off_in, int fd_out, loff_t* off_out,
		size_t len, unsigned int flags);
		#endif

		int
		hook_unlinkat(int dirfd, const char* cpath, int flags);

include/client/open_file_map.hpp

+1 −0

Original line number	Diff line number	Diff line
		@@ -63,6 +63,7 @@ enum class OpenFile_flags {
		cloexec,
		created, // indicates if the file was created during open
		creation_pending, // indicates if the file creation is delayed
		trunc_pending, // indicates O_TRUNC has been deferred
		flag_count // this is purely used as a size variable of this enum class
		};

src/client/gkfs_data.cpp

+68 −38

Original line number	Diff line number	Diff line
		@@ -133,7 +133,6 @@ gkfs_do_write(gkfs::filemap::OpenFile& file, const char* buf, size_t count,
		}
		file.set_flag(gkfs::filemap::OpenFile_flags::creation_pending, false);
		}

		// clear inline data cache as it is stale
		if(!file.inline_data().empty())
		file.inline_data("");
		@@ -331,14 +330,19 @@ gkfs_do_write(gkfs::filemap::OpenFile& file, const char* buf, size_t count,
		ssize_t
		gkfs_write_ws(gkfs::filemap::OpenFile& file, const char* buf, size_t count,
		off64_t offset, bool update_pos) {
		const auto pos_before = file.pos();
		#ifdef GKFS_ENABLE_CLIENT_METRICS
		auto start_t = std::chrono::high_resolution_clock::now();
		auto written = gkfs_do_write(file, buf, count, offset, update_pos);
		CTX->write_metrics()->add_event(written, start_t);
		return written;
		#else
		return gkfs_do_write(file, buf, count, offset, update_pos);
		auto written = gkfs_do_write(file, buf, count, offset, update_pos);
		#endif
		LOG(DEBUG,
		"{}() path '{}' count {} offset {} update_pos {} written {} pos_before {} pos_after {}",
		__func__, file.path(), count, offset, update_pos, written, pos_before,
		file.pos());
		return written;
		}

		/**
		@@ -353,8 +357,10 @@ gkfs_write_ws(gkfs::filemap::OpenFile& file, const char* buf, size_t count,
		ssize_t
		gkfs_pwrite(int fd, const void* buf, size_t count, off64_t offset) {
		auto file = CTX->file_map()->get(fd);
		if(!file)
		return 0;
		if(!file) {
		errno = EBADF;
		return -1;
		}
		return gkfs_write_ws(file, reinterpret_cast<const char>(buf), count,
		offset);
		}
		@@ -370,8 +376,10 @@ gkfs_pwrite(int fd, const void* buf, size_t count, off64_t offset) {
		ssize_t
		gkfs_write(int fd, const void* buf, size_t count) {
		auto gkfs_fd = CTX->file_map()->get(fd);
		if(!gkfs_fd)
		return 0;
		if(!gkfs_fd) {
		errno = EBADF;
		return -1;
		}
		// call pwrite and update pos
		auto ret = gkfs_write_ws(gkfs_fd, reinterpret_cast<const char>(buf),
		count, gkfs_fd->pos(), true);
		@@ -391,8 +399,10 @@ ssize_t
		gkfs_pwritev(int fd, const struct iovec* iov, int iovcnt, off_t offset) {

		auto file = CTX->file_map()->get(fd);
		if(!file)
		return 0;
		if(!file) {
		errno = EBADF;
		return -1;
		}
		auto pos = offset; // keep track of current position
		ssize_t written = 0;
		ssize_t ret;
		@@ -432,8 +442,10 @@ ssize_t
		gkfs_writev(int fd, const struct iovec* iov, int iovcnt) {

		auto gkfs_fd = CTX->file_map()->get(fd);
		if(!gkfs_fd)
		return 0;
		if(!gkfs_fd) {
		errno = EBADF;
		return -1;
		}
		auto pos = gkfs_fd->pos(); // retrieve the current offset
		auto ret = gkfs_pwritev(fd, iov, iovcnt, pos);
		assert(ret != 0);
		@@ -529,10 +541,12 @@ gkfs_do_read(const gkfs::filemap::OpenFile& file, char* buf, size_t count,
		__func__);
		}

		auto do_chunk_read = [&]() -> std::pair<int, long> {
		pair<int, long> ret;
		if(gkfs::config::proxy::fwd_io && CTX->use_proxy() &&
		count > gkfs::config::proxy::fwd_io_count_threshold) {
		ret = gkfs::rpc::forward_read_proxy(file.path(), buf, offset, count);
		ret = gkfs::rpc::forward_read_proxy(file.path(), buf, offset,
		count);
		} else {
		std::set<int8_t> failed; // set with failed targets.
		if(CTX->get_replicas() != 0) {
		@@ -540,17 +554,23 @@ gkfs_do_read(const gkfs::filemap::OpenFile& file, char* buf, size_t count,
		ret = gkfs::rpc::forward_read(file.path(), buf, offset, count,
		CTX->get_replicas(), failed);
		while(ret.first == EIO) {
		ret = gkfs::rpc::forward_read(file.path(), buf, offset, count,
		CTX->get_replicas(), failed);
		LOG(WARNING, "gkfs::rpc::forward_read() failed with ret '{}'",
		ret = gkfs::rpc::forward_read(file.path(), buf, offset,
		count, CTX->get_replicas(),
		failed);
		LOG(WARNING,
		"gkfs::rpc::forward_read() failed with ret '{}'",
		ret.first);
		}

		} else {
		ret = gkfs::rpc::forward_read(file.path(), buf, offset, count, 0,
		failed);
		ret = gkfs::rpc::forward_read(file.path(), buf, offset, count,
		0, failed);
		}
		}
		return ret;
		};

		auto ret = do_chunk_read();
		auto err = ret.first;
		if(err) {
		LOG(WARNING, "gkfs::rpc::forward_read() failed with ret '{}'", err);
		@@ -594,8 +614,10 @@ gkfs_read_ws(const gkfs::filemap::OpenFile& file, char* buf, size_t count,
		ssize_t
		gkfs_pread(int fd, void* buf, size_t count, off64_t offset) {
		auto gkfs_fd = CTX->file_map()->get(fd);
		if(!gkfs_fd)
		return 0;
		if(!gkfs_fd) {
		errno = EBADF;
		return -1;
		}
		return gkfs_read_ws(gkfs_fd, reinterpret_cast<char>(buf), count, offset);
		}

		@@ -610,8 +632,12 @@ gkfs_pread(int fd, void* buf, size_t count, off64_t offset) {
		ssize_t
		gkfs_read(int fd, void* buf, size_t count) {
		auto gkfs_fd = CTX->file_map()->get(fd);
		if(!gkfs_fd)
		return 0;
		if(!gkfs_fd) {
		errno = EBADF;
		return -1;
		}
		LOG(DEBUG, "{}() reading path '{}' count {} pos {}", __func__,
		gkfs_fd->path(), count, gkfs_fd->pos());
		auto pos = gkfs_fd->pos(); // retrieve the current offset
		auto ret = gkfs_read_ws(gkfs_fd, reinterpret_cast<char>(buf), count, pos);
		// Update offset in file descriptor in the file map
		@@ -634,8 +660,10 @@ ssize_t
		gkfs_preadv(int fd, const struct iovec* iov, int iovcnt, off_t offset) {

		auto file = CTX->file_map()->get(fd);
		if(!file)
		return 0;
		if(!file) {
		errno = EBADF;
		return -1;
		}
		auto pos = offset; // keep track of current position
		ssize_t read = 0;
		ssize_t ret;
		@@ -675,8 +703,10 @@ ssize_t
		gkfs_readv(int fd, const struct iovec* iov, int iovcnt) {

		auto gkfs_fd = CTX->file_map()->get(fd);
		if(!gkfs_fd)
		return 0;
		if(!gkfs_fd) {
		errno = EBADF;
		return -1;
		}
		auto pos = gkfs_fd->pos(); // retrieve the current offset
		auto ret = gkfs_preadv(fd, iov, iovcnt, pos);
		assert(ret != 0);
		@@ -691,8 +721,8 @@ int
		gkfs_fsync(unsigned int fd) {
		auto file = CTX->file_map()->get(fd);
		if(!file) {
		errno = 0;
		return 0;
		errno = EBADF;
		return -1;
		}
		// flush write size cache to be server consistent
		if(CTX->use_write_size_cache()) {