From c787006ee3742ee8384210c4355f73aef7c8d0b8 Mon Sep 17 00:00:00 2001 From: Luke Shumaker Date: Tue, 15 May 2018 17:36:35 -0400 Subject: initial commit --- .gitignore | 2 ++ Makefile | 13 +++++++ README.md | 4 +++ lib/dedupe-range.c | 86 ++++++++++++++++++++++++++++++++++++++++++++ lib/dedupe-range.h | 10 ++++++ src/cow-dedupe-range.c | 97 ++++++++++++++++++++++++++++++++++++++++++++++++++ 6 files changed, 212 insertions(+) create mode 100644 .gitignore create mode 100644 Makefile create mode 100644 README.md create mode 100644 lib/dedupe-range.c create mode 100644 lib/dedupe-range.h create mode 100644 src/cow-dedupe-range.c diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..59a4ba5 --- /dev/null +++ b/.gitignore @@ -0,0 +1,2 @@ +/cow-* +*.o diff --git a/Makefile b/Makefile new file mode 100644 index 0000000..10ebe01 --- /dev/null +++ b/Makefile @@ -0,0 +1,13 @@ +CFLAGS += -std=c11 +CPPFLAGS += -I$(CURDIR)/lib +CFLAGS += -Wall -Werror -Wextra +CPPFLAGS += -O2 -D_FORTIFY_SOURCE=2 + +all: cow-dedupe-range +.PHONY: all + +cow-dedupe-range: src/cow-dedupe-range.o lib/dedupe-range.o + $(CC) $(LDFLAGS) -o $@ $^ + +.SECONDARY: +.DELETE_ON_ERROR: diff --git a/README.md b/README.md new file mode 100644 index 0000000..bd7afd9 --- /dev/null +++ b/README.md @@ -0,0 +1,4 @@ +- ioctl + * FICLONE + * FICLONERANGE + * FIDEDUPERANGE diff --git a/lib/dedupe-range.c b/lib/dedupe-range.c new file mode 100644 index 0000000..dea70c7 --- /dev/null +++ b/lib/dedupe-range.c @@ -0,0 +1,86 @@ +#include /* for assert(3p) */ +#include /* for errno */ +#include /* for error(3gnu) */ +#include /* for open(2) */ +#include /* for uint64_t, PRIu64 */ +#include /* for FIDEDUPRANGE and related */ +#include /* for bool, true, false */ +#include /* for exit(3p), EXIT_SUCCESS, EXIT_FAILURE */ +#include /* for ioctl(2) */ +#include /* for sysconf(3p), _SC_PAGESIZE */ + +#include "dedupe-range.h" + +#define MIN(a, b) ((a) < (b) ? (a) : (b)) + +void dedupe_range(struct range src, struct range *dsts) { + size_t dst_count; + for (dst_count = 0; dsts[dst_count].filename; dst_count++); + + const size_t max_dst_count = (sysconf(_SC_PAGESIZE) - sizeof(struct file_dedupe_range)) + / sizeof(struct file_dedupe_range_info); + + int src_fd = open(src.filename, src.flags); + if (src_fd < 0) + error(EXIT_FAILURE, errno, "open src: %s", src.filename); + + struct file_dedupe_range_info *range_info = + calloc(dst_count, sizeof(struct file_dedupe_range_info)); + for (size_t i = 0; i < dst_count; i++) { + int dst_fd = open(dsts[i].filename, dsts[i].flags); + if (dst_fd < 0) + error(EXIT_FAILURE, errno, "open dst: %s", dsts[i].filename); + range_info[i].dest_fd = dst_fd; + range_info[i].dest_offset = dsts[i].offset; + } + + for (size_t files_deduped = 0; files_deduped < dst_count; ) { + uint16_t dest_count = MIN(dst_count - files_deduped, max_dst_count); + struct file_dedupe_range *range = malloc(sizeof(struct file_dedupe_range) + dest_count * sizeof(struct file_dedupe_range_info)); + *range = (struct file_dedupe_range){ + .src_offset = src.offset, + .src_length = src.length, + .dest_count = dest_count, + .reserved1 = 0, + .reserved2 = 0, + }; + for (size_t i = 0; i < dest_count; i++) + range->info[i] = range_info[files_deduped+i]; + + bool erred = false; + while (range->src_length > 0) { + if (ioctl(src_fd, FIDEDUPERANGE, &range) < 0) + error(EXIT_FAILURE, errno, "ioctl (FIDEDUPERANGE)"); + uint64_t bytes_deduped = range->info[0].bytes_deduped; + assert(bytes_deduped <= range->src_length); + for (size_t i = 0; i < range->dest_count; i ++) { + if (range->info[i].bytes_deduped != bytes_deduped) { + error(0, errno, "dedupe: %"PRIu64" != %"PRIu64": %s", + bytes_deduped, + /* on platforms where both "long" and "long long" + * are 64 bits, linux __u64 might disagree with + * glibc uint64_t and thus PRIu64 */ + (uint64_t)range->info[i].bytes_deduped, + dsts[files_deduped+i].filename); + erred = true; + } + switch (range->info[i].status) { + case FILE_DEDUPE_RANGE_DIFFERS: + error(0, 0, "dedupe: range differs: %s", dsts[files_deduped+i].filename); + erred = true; + break; + case FILE_DEDUPE_RANGE_SAME: + range->info[i].dest_offset += range->info[i].bytes_deduped; + break; + default: + assert(false); + } + } + if (erred == true) + exit(EXIT_FAILURE); + range->src_offset += bytes_deduped; + range->src_length -= bytes_deduped; + } + files_deduped += range->dest_count; + } +} diff --git a/lib/dedupe-range.h b/lib/dedupe-range.h new file mode 100644 index 0000000..028fe08 --- /dev/null +++ b/lib/dedupe-range.h @@ -0,0 +1,10 @@ +#include + +struct range { + char *filename; + int flags; /* to pass to open(2) */ + uint64_t offset; + uint64_t length; +}; + +void dedupe_range(struct range src, struct range *dsts); diff --git a/src/cow-dedupe-range.c b/src/cow-dedupe-range.c new file mode 100644 index 0000000..7508f04 --- /dev/null +++ b/src/cow-dedupe-range.c @@ -0,0 +1,97 @@ +#define _GNU_SOURCE /* for program_invocation_name */ +#include /* for assert(3p) */ +#include /* for errno, program_invocation_name */ +#include /* for error(3gnu) */ +#include /* for O_RDONLY, O_RDWR */ +#include /* for getopt_long(3gnu), struct option, optind */ +#include /* for bool, true, false */ +#include /* for uint64_t */ +#include /* for printf(3gnu), fprintf(3p), stderr */ +#include /* for strtoll(3p), calloc(3p), free(3p) */ +#include /* for strlen(3p) */ +#include /* for dup2(3p) */ + +#include "dedupe-range.h" /* for dedupe_range, struct range */ + +bool atou64(const char *str, uint64_t *ret) { + assert(sizeof(uint64_t) <= sizeof(unsigned long long int)); + if (!('0' <= str[0] && str[0] <= '9')) + return false; + errno = 0; + char *end; + unsigned long long int n = strtoll(str, &end, 10); + if (end[0] != '\0' || errno != 0) + return false; + *ret = (uint64_t)n; + return true; +} + +void usage() { + printf("Usage: %2$*1$s [OPTIONS] SRC_FILENAME SRC_OFFSET SRC_LENGTH \\\n" + " %3$*1$s DST_FILENAME DST_OFFSET \\\n" + " %3$*1$s [DST_FILENAME DST_OFFSET]...\n" + "Submit a file deduplication request to the kernel.\n" + "If the file ranges are not duplicates, the kernel will ignore this request.\n" + "\n" + "OPTIONS:\n" + " -r, --readonly Open destination files read-only, rather than read/write.\n" + " Generically, the destination must be open for writing.\n" + " However, on Btrfs, this allows deduplication to be done\n" + " on read-only subvolumes when invoked as root.\n" + " -h, --help Display this help and exit.\n", + (int)strlen(program_invocation_name), program_invocation_name, ""); +} + +int main(int argc, char *argv[]) { + int ro = 0; + struct option long_options[] = { + {"readonly", no_argument, &ro, 1}, + {"help", no_argument, NULL, 'h'}, + {0} + }; + int flag; + while ((flag = getopt_long(argc, argv, "rh", long_options, NULL)) >= 0) { + switch (flag) { + case 0: + break; + case 'r': + ro = 1; + break; + case 'h': + usage(); + return 0; + case '?': + fprintf(stderr, "Try '%s --help' for more information.\n", program_invocation_name); + return 2; + default: + assert(false); + } + } + + if (argc - optind < 5) { + error(0, 0, "too few arguments"); + fprintf(stderr, "Try '%s --help' for more information.\n", program_invocation_name); + return 2; + } + + struct range src; + src.filename = argv[optind]; + src.flags = O_RDONLY; + if (!atou64(argv[optind+1], &src.offset)) + error(2, errno, "invalid offset '%s'", argv[optind+1]); + if (!atou64(argv[optind+2], &src.length)) + error(2, errno, "invalid length '%s'", argv[optind+2]); + + const size_t dst_count = (argc - optind - 3) / 2; + struct range *dsts = calloc(dst_count + 1, sizeof(struct range)); + for (size_t i = 0; i < dst_count; i++) { + dsts[i].filename = argv[optind+3+(i*2)]; + dsts[i].flags = ro ? O_RDONLY : O_RDWR; + if (!atou64(argv[optind+3+(i*2)+1], &dsts[i].offset)) + error(2, errno, "invalid offset '%s'", argv[optind+3+(i*2)+1]); + } + + dedupe_range(src, dsts); + free(dsts); + return 0; +} -- cgit v1.2.3