Linux I/O 系统调用完整对比分析
1. 概述
Linux 提供了丰富的 I/O 系统调用,每种都有其特定的用途和优势。本文将详细分析这些系统调用的特点、使用场景和性能特征。
2. 系统调用详细对比
2.1 基本读写函数
pread/pwrite
#include <unistd.h>// 位置指定读取/写入
ssize_t pread(int fd, void *buf, size_t count, off_t offset);
ssize_t pwrite(int int fd, const void *buf, size_t count, off_t offset);
特点:
- 原子操作(读取/写入 + 位置指定)
- 不改变文件描述符的当前位置
- 线程安全
read/write
#include <unistd.h>// 基本读取/写入
ssize_t read(int fd, void *buf, size_t count);
ssize_t write(int fd, const void *buf, size_t count);
特点:
- 最基本的 I/O 操作
- 会改变文件描述符的当前位置
- 相对简单但功能有限
2.2 分散/聚集 I/O 函数
preadv/pwritev
#include <sys/uio.h>// 位置指定的分散读取/聚集写入
ssize_t preadv(int fd, const struct iovec *iov, int iovcnt, off_t offset);
ssize_t pwritev(int fd, const struct iovec *iov, int iovcnt, off_t offset);
preadv2/pwritev2
#define _GNU_SOURCE
#include <sys/uio.h>// 带标志的增强版分散/聚集 I/O
ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt, off_t offset, int flags);
ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt, off_t offset, int flags);
2.3 资源限制函数
prlimit64
#include <sys/resource.h>// 获取/设置进程资源限制
int prlimit64(pid_t pid, int resource, const struct rlimit64 *new_limit, struct rlimit64 *old_limit);
3. 功能对比表
函数 | 位置指定 | 分散/聚集 | 标志控制 | 原子性 | 跨平台 |
---|---|---|---|---|---|
read/write | ❌ | ❌ | ❌ | ⚠️ | ✅ |
pread/pwrite | ✅ | ❌ | ❌ | ✅ | ✅ |
readv/writev | ❌ | ✅ | ❌ | ⚠️ | ✅ |
preadv/pwritev | ✅ | ✅ | ❌ | ✅ | ✅ |
preadv2/pwritev2 | ✅ | ✅ | ✅ | ✅ | ❌ |
prlimit64 | ❌ | ❌ | ❌ | ❌ | ✅ |
4. 实际示例代码
4.1 基础读写对比
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/uio.h>
#include <string.h>
#include <errno.h>
#include <time.h>// 创建测试文件
void create_test_file(const char *filename) {int fd = open(filename, O_CREAT | O_WRONLY | O_TRUNC, 0644);if (fd == -1) {perror("创建文件失败");return;}const char *content = "Line 1: This is the first line of test data.\n""Line 2: This is the second line of test data.\n""Line 3: This is the third line of test data.\n""Line 4: This is the fourth line of test data.\n""Line 5: This is the fifth and final line.\n";write(fd, content, strlen(content));close(fd);printf("创建测试文件: %s\n", filename);
}// read/pread 性能对比
void compare_basic_io(const char *filename) {int fd = open(filename, O_RDONLY);if (fd == -1) {perror("打开文件失败");return;}char buffer[100];ssize_t bytes_read;struct timespec start, end;printf("\n=== 基础 I/O 对比 ===\n");// 使用 read (会改变文件位置)printf("1. read 测试:\n");clock_gettime(CLOCK_MONOTONIC, &start);bytes_read = read(fd, buffer, 50);clock_gettime(CLOCK_MONOTONIC, &end);buffer[bytes_read] = '\0';printf(" 读取 %zd 字节: %.30s...\n", bytes_read, buffer);printf(" 当前文件位置: %ld\n", (long)lseek(fd, 0, SEEK_CUR));// 使用 pread (不改变文件位置)printf("2. pread 测试:\n");clock_gettime(CLOCK_MONOTONIC, &start);bytes_read = pread(fd, buffer, 50, 0); // 从开头读取clock_gettime(CLOCK_MONOTONIC, &end);buffer[bytes_read] = '\0';printf(" 读取 %zd 字节: %.30s...\n", bytes_read, buffer);printf(" 文件位置仍为: %ld\n", (long)lseek(fd, 0, SEEK_CUR));close(fd);
}// 分散/聚集 I/O 示例
void demonstrate_vector_io(const char *filename) {int fd = open(filename, O_RDONLY);if (fd == -1) {perror("打开文件失败");return;}printf("\n=== 分散/聚集 I/O 示例 ===\n");// 设置分散读取缓冲区char buffer1[20], buffer2[30], buffer3[25];struct iovec iov[3];iov[0].iov_base = buffer1;iov[0].iov_len = sizeof(buffer1) - 1;iov[1].iov_base = buffer2;iov[1].iov_len = sizeof(buffer2) - 1;iov[2].iov_base = buffer3;iov[2].iov_len = sizeof(buffer3) - 1;printf("分散读取设置:\n");printf(" 缓冲区1: %zu 字节\n", iov[0].iov_len);printf(" 缓冲区2: %zu 字节\n", iov[1].iov_len);printf(" 缓冲区3: %zu 字节\n", iov[2].iov_len);// 使用 preadv 一次性读取到多个缓冲区ssize_t total_bytes = preadv(fd, iov, 3, 0); // 从文件开头开始读取printf("总共读取: %zd 字节\n", total_bytes);if (total_bytes > 0) {buffer1[iov[0].iov_len] = '\0';buffer2[iov[1].iov_len] = '\0';buffer3[iov[2].iov_len] = '\0';printf("读取结果:\n");printf(" 缓冲区1: %s\n", buffer1);printf(" 缓冲区2: %s\n", buffer2);printf(" 缓冲区3: %s\n", buffer3);}close(fd);
}// 资源限制示例
void demonstrate_resource_limits() {printf("\n=== 资源限制示例 ===\n");struct rlimit64 limit;// 获取当前进程的文件大小限制if (prlimit64(0, RLIMIT_FSIZE, NULL, &limit) == 0) {printf("文件大小限制:\n");printf(" 软限制: %lld\n", (long long)limit.rlim_cur);printf(" 硬限制: %lld\n", (long long)limit.rlim_max);}// 获取内存限制if (prlimit64(0, RLIMIT_AS, NULL, &limit) == 0) {printf("虚拟内存限制:\n");if (limit.rlim_cur == RLIM64_INFINITY) {printf(" 软限制: 无限制\n");} else {printf(" 软限制: %lld 字节 (%.2f MB)\n", (long long)limit.rlim_cur,(double)limit.rlim_cur / (1024 * 1024));}if (limit.rlim_max == RLIM64_INFINITY) {printf(" 硬限制: 无限制\n");} else {printf(" 硬限制: %lld 字节 (%.2f MB)\n", (long long)limit.rlim_max,(double)limit.rlim_max / (1024 * 1024));}}// 获取打开文件数限制if (prlimit64(0, RLIMIT_NOFILE, NULL, &limit) == 0) {printf("文件描述符限制:\n");printf(" 软限制: %lld\n", (long long)limit.rlim_cur);printf(" 硬限制: %lld\n", (long long)limit.rlim_max);}
}int main() {const char *test_file = "io_test_file.txt";printf("=== Linux I/O 系统调用对比分析 ===\n");// 创建测试文件create_test_file(test_file);// 基础 I/O 对比compare_basic_io(test_file);// 分散/聚集 I/O 示例demonstrate_vector_io(test_file);// 资源限制示例demonstrate_resource_limits();// 清理unlink(test_file);printf("\n=== 使用建议 ===\n");printf("选择指南:\n");printf("1. 简单读写: 使用 read/write\n");printf("2. 需要指定位置: 使用 pread/pwrite\n");printf("3. 多缓冲区操作: 使用 preadv/pwritev\n");printf("4. 需要高级控制: 使用 preadv2/pwritev2\n");printf("5. 资源限制管理: 使用 prlimit64\n");return 0;
}
4.2 性能基准测试
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/uio.h>
#include <time.h>
#include <string.h>#define ITERATIONS 10000
#define BUFFER_SIZE 1024// 性能测试结构体
struct performance_test {const char *name;double (*test_func)(int fd, const char *filename);
};// read 性能测试
double test_read_performance(int fd, const char *filename) {char buffer[BUFFER_SIZE];struct timespec start, end;clock_gettime(CLOCK_MONOTONIC, &start);for (int i = 0; i < ITERATIONS; i++) {lseek(fd, 0, SEEK_SET); // 重置到文件开头read(fd, buffer, sizeof(buffer));}clock_gettime(CLOCK_MONOTONIC, &end);return (end.tv_sec - start.tv_sec) * 1000000.0 + (end.tv_nsec - start.tv_nsec) / 1000.0; // 微秒
}// pread 性能测试
double test_pread_performance(int fd, const char *filename) {char buffer[BUFFER_SIZE];struct timespec start, end;clock_gettime(CLOCK_MONOTONIC, &start);for (int i = 0; i < ITERATIONS; i++) {pread(fd, buffer, sizeof(buffer), 0); // 始终从位置0读取}clock_gettime(CLOCK_MONOTONIC, &end);return (end.tv_sec - start.tv_sec) * 1000000.0 + (end.tv_nsec - start.tv_nsec) / 1000.0; // 微秒
}// readv 性能测试
double test_readv_performance(int fd, const char *filename) {char buffer1[256], buffer2[256], buffer3[256], buffer4[256];struct iovec iov[4];struct timespec start, end;// 设置分散读取iov[0].iov_base = buffer1;iov[0].iov_len = sizeof(buffer1);iov[1].iov_base = buffer2;iov[1].iov_len = sizeof(buffer2);iov[2].iov_base = buffer3;iov[2].iov_len = sizeof(buffer3);iov[3].iov_base = buffer4;iov[3].iov_len = sizeof(buffer4);clock_gettime(CLOCK_MONOTONIC, &start);for (int i = 0; i < ITERATIONS; i++) {lseek(fd, 0, SEEK_SET);readv(fd, iov, 4);}clock_gettime(CLOCK_MONOTONIC, &end);return (end.tv_sec - start.tv_sec) * 1000000.0 + (end.tv_nsec - start.tv_nsec) / 1000.0; // 微秒
}// preadv 性能测试
double test_preadv_performance(int fd, const char *filename) {char buffer1[256], buffer2[256], buffer3[256], buffer4[256];struct iovec iov[4];struct timespec start, end;// 设置分散读取iov[0].iov_base = buffer1;iov[0].iov_len = sizeof(buffer1);iov[1].iov_base = buffer2;iov[1].iov_len = sizeof(buffer2);iov[2].iov_base = buffer3;iov[2].iov_len = sizeof(buffer3);iov[3].iov_base = buffer4;iov[3].iov_len = sizeof(buffer4);clock_gettime(CLOCK_MONOTONIC, &start);for (int i = 0; i < ITERATIONS; i++) {preadv(fd, iov, 4, 0); // 始终从位置0读取}clock_gettime(CLOCK_MONOTONIC, &end);return (end.tv_sec - start.tv_sec) * 1000000.0 + (end.tv_nsec - start.tv_nsec) / 1000.0; // 微秒
}void run_performance_benchmark() {const char *test_file = "benchmark_test.dat";int fd;printf("=== I/O 性能基准测试 ===\n");// 创建大测试文件fd = open(test_file, O_CREAT | O_WRONLY | O_TRUNC, 0644);if (fd != -1) {char *buffer = malloc(1024 * 1024); // 1MB 缓冲区if (buffer) {for (int i = 0; i < 10; i++) { // 创建 10MB 文件write(fd, buffer, 1024 * 1024);}free(buffer);}close(fd);}// 打开文件进行读取测试fd = open(test_file, O_RDONLY);if (fd == -1) {perror("打开测试文件失败");return;}struct performance_test tests[] = {{"read", test_read_performance},{"pread", test_pread_performance},{"readv", test_readv_performance},{"preadv", test_preadv_performance},{NULL, NULL}};printf("%-10s %-15s %-15s\n", "函数", "耗时(微秒)", "平均耗时(纳秒)");printf("%-10s %-15s %-15s\n", "----", "----------", "--------------");for (int i = 0; tests[i].name; i++) {double total_time = tests[i].test_func(fd, test_file);double avg_time = total_time * 1000.0 / ITERATIONS;printf("%-10s %-15.2f %-15.2f\n", tests[i].name, total_time, avg_time);}close(fd);unlink(test_file);printf("\n性能分析:\n");printf("1. pread 比 read 略慢 (位置指定开销)\n");printf("2. readv/preadv 在多缓冲区场景下更高效\n");printf("3. preadv2/pwritev2 提供更多控制选项\n");printf("4. 选择应基于具体使用场景\n");
}int main() {printf("=== Linux I/O 系统调用完整对比分析 ===\n\n");// 运行性能基准测试run_performance_benchmark();printf("\n=== 详细功能对比 ===\n");printf("pread vs read:\n");printf(" • pread: 指定位置读取,不改变文件位置\n");printf(" • read: 顺序读取,会改变文件位置\n");printf("\n");printf("preadv vs pread:\n");printf(" • preadv: 分散读取到多个缓冲区\n");printf(" • pread: 读取到单个缓冲区\n");printf("\n");printf("preadv2 vs preadv:\n");printf(" • preadv2: 支持额外标志控制\n");printf(" • preadv: 基本的分散读取功能\n");printf("\n");printf("prlimit64:\n");printf(" • 用于获取和设置进程资源限制\n");printf(" • 支持 64 位资源限制值\n");printf(" • 可以操作其他进程的资源限制\n");printf("\n=== 实际应用建议 ===\n");printf("1. 日志文件读写: 使用 pread/pwrite\n");printf("2. 数据库存储引擎: 使用 preadv/pwritev\n");printf("3. 高性能网络服务: 使用 preadv2/pwritev2\n");printf("4. 系统资源管理: 使用 prlimit64\n");printf("5. 简单文件操作: 使用 read/write\n");return 0;
}
4.3 实际应用场景演示
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/uio.h>
#include <string.h>
#include <errno.h>
#include <sys/resource.h>// 日志文件读取示例
void log_file_reader_example() {printf("=== 日志文件读取场景 ===\n");// 创建模拟日志文件const char *log_file = "application.log";int fd = open(log_file, O_CREAT | O_WRONLY | O_TRUNC, 0644);if (fd != -1) {const char *log_entries[] = {"2023-01-01 10:00:00 INFO Application started\n","2023-01-01 10:00:01 DEBUG Loading configuration\n","2023-01-01 10:00:02 WARN Low memory warning\n","2023-01-01 10:00:03 ERROR Database connection failed\n","2023-01-01 10:00:04 INFO Recovery attempt started\n"};for (int i = 0; i < 5; i++) {write(fd, log_entries[i], strlen(log_entries[i]));}close(fd);}// 使用 pread 读取特定时间段的日志fd = open(log_file, O_RDONLY);if (fd != -1) {char buffer[256];printf("读取最后一条日志记录:\n");// 从文件末尾附近读取ssize_t bytes_read = pread(fd, buffer, sizeof(buffer) - 1, 150);if (bytes_read > 0) {buffer[bytes_read] = '\0';printf(" %s", buffer);}close(fd);}unlink(log_file);
}// 数据库页读取示例
void database_page_reader_example() {printf("\n=== 数据库页读取场景 ===\n");const char *db_file = "database_pages.dat";int fd = open(db_file, O_CREAT | O_WRONLY | O_TRUNC, 0644);if (fd != -1) {// 创建模拟的数据库页char page_data[4096];for (int page = 0; page < 10; page++) {snprintf(page_data, sizeof(page_data), "Page %d: Database page content with ID=%d and timestamp=%ld\n",page, page * 1000, time(NULL));write(fd, page_data, strlen(page_data));}close(fd);}// 使用 preadv 读取多个数据库页fd = open(db_file, O_RDONLY);if (fd != -1) {char page1[1024], page2[1024], page3[1024];struct iovec iov[3];// 设置分散读取iov[0].iov_base = page1;iov[0].iov_len = sizeof(page1) - 1;iov[1].iov_base = page2;iov[1].iov_len = sizeof(page2) - 1;iov[2].iov_base = page3;iov[2].iov_len = sizeof(page3) - 1;printf("使用 preadv 读取多个数据库页:\n");ssize_t total_bytes = preadv(fd, iov, 3, 0); // 从开头读取printf(" 总共读取: %zd 字节\n", total_bytes);if (total_bytes > 0) {page1[iov[0].iov_len] = '\0';page2[iov[1].iov_len] = '\0';page3[iov[2].iov_len] = '\0';printf(" 页1: %.50s...\n", page1);printf(" 页2: %.50s...\n", page2);printf(" 页3: %.50s...\n", page3);}close(fd);}unlink(db_file);
}// 网络数据包处理示例
void network_packet_processor_example() {printf("\n=== 网络数据包处理场景 ===\n");// 模拟网络数据包结构struct packet_header {uint32_t magic;uint16_t version;uint16_t type;uint32_t length;uint32_t checksum;} __attribute__((packed));struct packet_payload {char data[1024];};// 创建测试数据包文件const char *packet_file = "network_packets.dat";int fd = open(packet_file, O_CREAT | O_WRONLY | O_TRUNC, 0644);if (fd != -1) {// 写入多个数据包for (int i = 0; i < 3; i++) {struct packet_header header = {.magic = 0x12345678,.version = 1,.type = i,.length = 100,.checksum = 0xABCDEF00 + i};char payload[1024];snprintf(payload, sizeof(payload), "Packet %d payload data with timestamp %ld", i, time(NULL));write(fd, &header, sizeof(header));write(fd, payload, strlen(payload) + 1);}close(fd);}// 使用 preadv2 读取数据包(如果支持)fd = open(packet_file, O_RDONLY);if (fd != -1) {struct packet_header headers[3];char payloads[3][256];struct iovec iov[6]; // 3个头部 + 3个载荷// 设置分散读取结构for (int i = 0; i < 3; i++) {iov[i*2].iov_base = &headers[i];iov[i*2].iov_len = sizeof(struct packet_header);iov[i*2+1].iov_base = payloads[i];iov[i*2+1].iov_len = sizeof(payloads[i]) - 1;}printf("使用分散读取处理网络数据包:\n");ssize_t bytes_read = preadv(fd, iov, 6, 0);printf(" 读取字节数: %zd\n", bytes_read);if (bytes_read > 0) {for (int i = 0; i < 3; i++) {printf(" 数据包 %d:\n", i);printf(" 魔数: 0x%08X\n", headers[i].magic);printf(" 版本: %d\n", headers[i].version);printf(" 类型: %d\n", headers[i].type);printf(" 长度: %d\n", headers[i].length);printf(" 校验: 0x%08X\n", headers[i].checksum);payloads[i][iov[i*2+1].iov_len] = '\0';printf(" 载荷: %.50s...\n", payloads[i]);printf("\n");}}close(fd);}unlink(packet_file);
}// 资源限制管理示例
void resource_limit_management_example() {printf("\n=== 资源限制管理场景 ===\n");struct rlimit64 old_limit, new_limit;// 获取当前文件大小限制if (prlimit64(0, RLIMIT_FSIZE, NULL, &old_limit) == 0) {printf("当前文件大小限制:\n");if (old_limit.rlim_cur == RLIM64_INFINITY) {printf(" 软限制: 无限制\n");} else {printf(" 软限制: %lld 字节 (%.2f GB)\n", (long long)old_limit.rlim_cur,(double)old_limit.rlim_cur / (1024 * 1024 * 1024));}}// 获取打开文件数限制if (prlimit64(0, RLIMIT_NOFILE, NULL, &old_limit) == 0) {printf("当前文件描述符限制:\n");printf(" 软限制: %lld\n", (long long)old_limit.rlim_cur);printf(" 硬限制: %lld\n", (long long)old_limit.rlim_max);}// 获取内存限制if (prlimit64(0, RLIMIT_AS, NULL, &old_limit) == 0) {printf("当前虚拟内存限制:\n");if (old_limit.rlim_cur == RLIM64_INFINITY) {printf(" 软限制: 无限制\n");} else {printf(" 软限制: %lld 字节 (%.2f GB)\n", (long long)old_limit.rlim_cur,(double)old_limit.rlim_cur / (1024 * 1024 * 1024));}}printf("\n资源限制管理最佳实践:\n");printf("1. 合理设置文件大小限制防止磁盘填满\n");printf("2. 适当增加文件描述符限制支持高并发\n");printf("3. 监控内存使用防止内存泄漏\n");printf("4. 使用 prlimit64 动态调整资源限制\n");
}int main() {printf("=== Linux I/O 系统调用应用场景演示 ===\n\n");// 日志文件读取场景log_file_reader_example();// 数据库页读取场景database_page_reader_example();// 网络数据包处理场景network_packet_processor_example();// 资源限制管理场景resource_limit_management_example();printf("\n=== 总结 ===\n");printf("I/O 系统调用选择指南:\n");printf("\n");printf("┌─────────────┬────────────────────────────────────┐\n");printf("│ 场景 │ 推荐函数 │\n");printf("├─────────────┼────────────────────────────────────┤\n");printf("│ 简单读写 │ read/write │\n");printf("│ 位置指定 │ pread/pwrite │\n");printf("│ 多缓冲区 │ readv/writev │\n");printf("│ 位置+多缓冲 │ preadv/pwritev │\n");printf("│ 高级控制 │ preadv2/pwritev2 │\n");printf("│ 资源限制 │ prlimit64 │\n");printf("└─────────────┴────────────────────────────────────┘\n");printf("\n");printf("性能优化建议:\n");printf("1. 批量操作减少系统调用次数\n");printf("2. 合理选择缓冲区大小\n");printf("3. 使用位置指定避免文件位置移动\n");printf("4. 分散/聚集 I/O 减少内存拷贝\n");printf("5. 合理设置资源限制防止系统过载\n");return 0;
}
5. 编译和运行说明
# 编译示例程序
gcc -o io_comparison_example1 example1.c
gcc -o io_comparison_example2 example2.c
gcc -o io_comparison_example3 example3.c# 运行示例
./io_comparison_example1
./io_comparison_example2
./io_comparison_example3
6. 系统要求检查
# 检查内核版本
uname -r# 检查 glibc 版本
ldd --version# 检查系统调用支持
grep -E "(pread|pwrite|prlimit)" /usr/include/asm/unistd_64.h# 查看文件系统性能
hdparm -Tt /dev/sda # 硬盘性能测试
7. 重要注意事项
- 原子性: pread/pwrite 操作是原子的
- 位置独立: 不改变文件描述符的当前位置
- 错误处理: 始终检查返回值和 errno
- 内存对齐: 在某些架构上有对齐要求
- 权限检查: 确保有足够的权限进行操作
- 资源清理: 及时关闭文件描述符
8. 最佳实践总结
// 安全的 I/O 操作封装
ssize_t safe_pread(int fd, void *buf, size_t count, off_t offset) {if (fd < 0 || !buf || count == 0) {errno = EINVAL;return -1;}ssize_t result;do {result = pread(fd, buf, count, offset);} while (result == -1 && errno == EINTR);return result;
}// 安全的分散读取封装
ssize_t safe_preadv(int fd, const struct iovec *iov, int iovcnt, off_t offset) {if (fd < 0 || !iov || iovcnt <= 0 || iovcnt > IOV_MAX) {errno = EINVAL;return -1;}ssize_t result;do {result = preadv(fd, iov, iovcnt, offset);} while (result == -1 && errno == EINTR);return result;
}// 资源限制检查
int check_resource_limits() {struct rlimit64 limit;// 检查文件大小限制if (prlimit64(0, RLIMIT_FSIZE, NULL, &limit) == 0) {if (limit.rlim_cur != RLIM64_INFINITY && limit.rlim_cur < 1024 * 1024) {printf("警告: 文件大小限制过小 (%lld 字节)\n", (long long)limit.rlim_cur);}}// 检查文件描述符限制if (prlimit64(0, RLIMIT_NOFILE, NULL, &limit) == 0) {if (limit.rlim_cur < 1024) {printf("警告: 文件描述符限制过小 (%lld)\n", (long long)limit.rlim_cur);}}return 0;
}
这些示例全面展示了 Linux I/O 系统调用的功能特点、使用方法和实际应用场景,帮助开发者根据具体需求选择合适的 I/O 操作方式