Operating SystemsThread Concepts

Thread Fundamentals

LevelIntermediate

Duration60 mins

TopicThread Concepts

3 / 5

Shared Resources

The Shared Everything Model

When threads execute within the same process, they share a remarkable amount of state. This sharing is the source of threading's power—and its peril. Understanding exactly what is shared, why it's shared, and how to manage that sharing safely is fundamental to writing correct concurrent programs.

The resources threads share fall into several categories:

Memory resources: Code segment, data segment, heap
I/O resources: File descriptors, sockets, pipes
Process-level state: Signal handlers, working directory, environment
System resources: User/group IDs, resource limits, credentials

Each shared resource introduces specific considerations for concurrent access. A misunderstanding of any one can lead to subtle, difficult-to-debug problems.

What You Will Learn

By the end of this page, you will understand every category of shared resource in detail: why it's shared, what problems can arise from concurrent access, and what strategies exist for safe sharing. You'll gain the knowledge needed to reason about thread safety for any shared resource.

The Code Segment (Text Section)

The code segment (also called the text section) contains the compiled machine instructions of the program. This is the actual executable code—the functions, loops, and logic that threads execute.

Why It's Shared:

There's only one copy of the program code in memory, and all threads read from it. This is both efficient and logical:

Memory efficiency: N threads don't need N copies of the program
Cache efficiency: Instruction cache hits benefit all threads
Logical necessity: Threads execute the same program

Is It Safe?

The code segment is read-only (marked as executable but not writable in the memory map). Threads cannot modify it at runtime. This means:

No synchronization needed for code access
Multiple threads can execute the same function simultaneously
Each thread has its own program counter, so they can be at different points

code_segment_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
#include <pthread.h>
#include <stdio.h>
 
/* This function exists once in the code segment */
void shared_function(int thread_id, int value) {
    /* The function's code is shared, but: */
    /* - 'thread_id' and 'value' are on each thread's stack (private) */
    /* - The CPU registers holding them are per-thread (private) */
    
    int local_result = value * 2;  /* Stack variable - private to this call */
    
    printf("Thread %d computed: %d\n", thread_id, local_result);
    /* printf's code is also shared, but each call is independent */
}
 
void *worker(void *arg) {
    int id = *(int*)arg;
    
    /* Multiple threads call the same function simultaneously */
    /* Each has its own stack frame, so they don't interfere */
    for (int i = 0; i < 3; i++) {
        shared_function(id, i + id * 100);
    }
    
    return NULL;
}
 
int main() {
    pthread_t threads[3];
    int ids[] = {1, 2, 3};
    
    for (int i = 0; i < 3; i++) {
        pthread_create(&threads[i], NULL, worker, &ids[i]);
    }
    
    for (int i = 0; i < 3; i++) {
        pthread_join(threads[i], NULL);
    }
    
    return 0;
}

Functions Are Reentrant by Design

Because code is read-only and each thread has its own stack, multiple threads can execute the same function simultaneously without interference—as long as the function only uses local variables and parameters. Functions that rely solely on local state are called 'reentrant' and are inherently thread-safe.

Position-Independent Code (PIC):

Shared libraries use position-independent code, allowing the same library code to be mapped at different virtual addresses in different processes while being physically shared in RAM. Threads within the same process see the library at the same address, further reinforcing the shared nature of code.

The Data Segment (Global and Static Data)

The data segment contains global and static variables. This includes:

Initialized data section (.data): Global/static variables with explicit initial values
Uninitialized data section (.bss): Global/static variables initialized to zero

Unlike the code segment, the data segment is read-write, and all threads can both read and modify it.

Why It's Shared:

Global and static variables are meant to have a single instance visible throughout the program. In a single-threaded world, this is straightforward. With threads, it becomes a source of complexity.

The Danger:

Concurrent access to shared mutable data without synchronization causes data races—one of the most common and pernicious bugs in concurrent programming.

data_race_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
#include <pthread.h>
#include <stdio.h>
 
/* Global variable - shared between all threads */
int global_counter = 0;  /* Stored in .bss (initialized to zero) */
 
/* Static variable - also shared! */
static int static_counter = 100;  /* Stored in .data */
 
void *unsafe_increment(void *arg) {
    for (int i = 0; i < 1000000; i++) {
        /* DATA RACE: Two threads may read the same value, */
        /* compute the same result, and write it back, */
        /* losing one increment. */
        global_counter++;
        
        /* This looks like one operation but is actually: */
        /* 1. Read global_counter from memory into register */
        /* 2. Add 1 to register */
        /* 3. Write register back to global_counter */
        /* A context switch between any steps causes lost updates */
    }
    return NULL;
}
 
int main() {
    pthread_t t1, t2;
    
    printf("Initial counter: %d\n", global_counter);
    
    pthread_create(&t1, NULL, unsafe_increment, NULL);
    pthread_create(&t2, NULL, unsafe_increment, NULL);
    
    pthread_join(t1, NULL);
    pthread_join(t2, NULL);
    
    printf("Final counter: %d\n", global_counter);
    printf("Expected: 2000000\n");
    printf("Difference (lost updates): %d\n", 2000000 - global_counter);
    
    return 0;
}
/* Typical output:
 * Initial counter: 0
 * Final counter: 1234567  (some value less than 2000000)
 * Expected: 2000000
 * Difference (lost updates): 765433
 */

Types of Problematic Global/Static Data

•Simple counters and flags — As shown above, even counter++ is not atomic and can lose updates.
•State variables — Global state machines where threads read and update state can have race conditions.
•Configuration structures — If one thread modifies config while another reads, partial updates can be seen.
•Buffers and arrays — Multiple threads writing to the same buffer without coordination corrupts data.
•Function-static variables — A static variable inside a function persists across calls and is shared by all threads calling that function.

Static Variables in Functions

A particular trap: static variables declared inside functions. These persist across calls and are shared by all threads. The classic example is strtok(), which uses internal static state and is notoriously thread-unsafe. The thread-safe alternative is strtok_r(), which takes an explicit state pointer.

static_function_danger.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
/* Classic thread-unsafe pattern: static buffer in function */
char *get_timestamp(void) {
    static char buffer[64];  /* SHARED between all callers! */
    
    time_t now = time(NULL);
    strftime(buffer, sizeof(buffer), "%Y-%m-%d %H:%M:%S", localtime(&now));
    
    return buffer;
    /* If two threads call this simultaneously, they overwrite */
    /* each other's buffer contents! */
}
 
/* Thread-safe alternative: caller provides buffer */
void get_timestamp_safe(char *buffer, size_t size) {
    time_t now = time(NULL);
    struct tm tm_buf;
    localtime_r(&now, &tm_buf);  /* Thread-safe localtime */
    strftime(buffer, size, "%Y-%m-%d %H:%M:%S", &tm_buf);
}

The Heap (Dynamic Memory)

The heap is the region of memory used for dynamic allocation (malloc, new, etc.). All threads share the same heap, meaning memory allocated by one thread is accessible by all threads (if they have the pointer).

Key Characteristics:

All threads can allocate from the same heap
Memory allocated by one thread can be freed by another
Memory allocated by one thread can be used by another
The heap allocator itself must be thread-safe

Heap Allocator Thread Safety:

Modern malloc implementations (glibc, jemalloc, tcmalloc) are thread-safe. They use internal locking and/or per-thread arenas to prevent corruption from concurrent allocations. However, this has costs:

Lock contention when many threads allocate/free simultaneously
Per-thread arenas use more memory but reduce contention

Common Heap Allocators and Their Threading Strategies
Allocator	Strategy	Characteristics
glibc malloc	Arenas with locks	Good general-purpose performance; arenas reduce contention
jemalloc (FreeBSD, Rust)	Thread-local caches + arenas	Excellent multi-threaded performance; used by Firefox, Redis
tcmalloc (Google)	Thread-caching	Very fast for allocation-heavy workloads
mimalloc (Microsoft)	Free-list sharding	Excellent performance with minimal memory overhead

heap_sharing_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
#include <pthread.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
 
/* Producer allocates on heap and passes pointer to consumer */
typedef struct {
    int id;
    char message[256];
} Task;
 
Task *task_queue[100];
int queue_head = 0, queue_tail = 0;
pthread_mutex_t queue_lock = PTHREAD_MUTEX_INITIALIZER;
pthread_cond_t queue_not_empty = PTHREAD_COND_INITIALIZER;
 
void *producer(void *arg) {
    for (int i = 0; i < 10; i++) {
        /* Allocate on heap - this memory is accessible to all threads */
        Task *task = malloc(sizeof(Task));
        task->id = i;
        snprintf(task->message, sizeof(task->message), 
                 "Task %d from producer", i);
        
        pthread_mutex_lock(&queue_lock);
        task_queue[queue_tail++] = task;  /* Hand pointer to consumer */
        pthread_cond_signal(&queue_not_empty);
        pthread_mutex_unlock(&queue_lock);
    }
    return NULL;
}
 
void *consumer(void *arg) {
    for (int i = 0; i < 10; i++) {
        pthread_mutex_lock(&queue_lock);
        while (queue_head == queue_tail) {
            pthread_cond_wait(&queue_not_empty, &queue_lock);
        }
        Task *task = task_queue[queue_head++];
        pthread_mutex_unlock(&queue_lock);
        
        /* Use the heap-allocated task from producer */
        printf("Consumer received: %s\n", task->message);
        
        /* Free memory allocated by another thread - perfectly valid */
        free(task);
    }
    return NULL;
}

Common Heap-Related Threading Errors

•Double free — Thread A frees memory, thread B doesn't know and also frees it. Heap corruption ensues.
•Use after free — Thread A frees memory, thread B continues using the pointer. Undefined behavior.
•Memory leaks — Thread A allocates, but the pointer is lost when the thread exits before anyone frees it.
•Data races on heap objects — Just because the object is on the heap doesn't make access to its contents thread-safe.
•False sharing — Two threads modify different objects that happen to share a cache line, causing performance degradation.

Ownership Transfer Pattern

A clean pattern for heap sharing: define clear ownership transfer semantics. The producer 'owns' the memory until it's placed in the queue. The consumer then 'owns' it and is responsible for freeing. At any point, exactly one thread 'owns' each object. This eliminates double-free and use-after-free by design.

File Descriptors and I/O Resources

All threads in a process share the file descriptor table. This means:

A file opened by any thread is accessible by all threads
The file offset (position for reading/writing) is shared
A file closed by any thread is closed for all threads
Sockets, pipes, and other I/O resources are similarly shared

The Shared File Offset Problem:

When multiple threads read from or write to the same file descriptor, they share the file offset. This leads to unpredictable behavior:

shared_fd_problem.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
#include <pthread.h>
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
 
int shared_fd;  /* File descriptor shared by all threads */
 
void *writer(void *arg) {
    int id = *(int*)arg;
    char buffer[100];
    
    for (int i = 0; i < 5; i++) {
        snprintf(buffer, sizeof(buffer), "Thread %d: Message %d\n", id, i);
        
        /* PROBLEM: These operations are NOT atomic together */
        /* Thread could be preempted between lseek and write */
        /* or between two write() calls */
        
        write(shared_fd, buffer, strlen(buffer));
        /* After each write, file offset advances */
        /* But another thread might write in between! */
    }
    return NULL;
}
 
/* Result: Garbled, interleaved output in the file */
 
/* Solution 1: Use mutex to serialize access */
pthread_mutex_t fd_lock = PTHREAD_MUTEX_INITIALIZER;
 
void *safe_writer(void *arg) {
    int id = *(int*)arg;
    char buffer[100];
    
    for (int i = 0; i < 5; i++) {
        snprintf(buffer, sizeof(buffer), "Thread %d: Message %d\n", id, i);
        
        pthread_mutex_lock(&fd_lock);
        write(shared_fd, buffer, strlen(buffer));
        pthread_mutex_unlock(&fd_lock);
    }
    return NULL;
}
 
/* Solution 2: Use pwrite() for atomic positioned write */
void *atomic_writer(void *arg) {
    int id = *(int*)arg;
    static _Atomic off_t global_offset = 0;  /* Atomic offset counter */
    char buffer[100];
    
    for (int i = 0; i < 5; i++) {
        int len = snprintf(buffer, sizeof(buffer), 
                           "Thread %d: Message %d\n", id, i);
        
        off_t my_offset = atomic_fetch_add(&global_offset, len);
        pwrite(shared_fd, buffer, len, my_offset);  /* Atomic at offset */
    }
    return NULL;
}

Shared I/O Resource Implications
Resource Type	Sharing Behavior	Thread Safety Concern
Regular files	Shared offset, shared content	Interleaved reads/writes, race on offset
Pipes	Shared read/write ends	Interleaved data, partial reads
Sockets	Shared connection state	Interleaved sends, protocol corruption
stdin/stdout/stderr	fd 0, 1, 2 shared	Interleaved console output
Directory handles	Shared position in readdir	Missing/duplicate entries in listing

Thread-Safe I/O Strategies

Options for safe concurrent I/O: (1) Serialize access with mutexes, (2) Use pread()/pwrite() for positioned I/O that doesn't affect shared offset, (3) Each thread opens its own file descriptor (separate offset per thread), (4) Design so only one thread handles each file descriptor.

Close-While-Using Hazard:

Another danger: if one thread closes a file descriptor while another is using it, the behavior is undefined. Worse, the kernel may reuse the file descriptor number for a new file, so the thread that didn't know about the close might now be reading/writing an entirely different file:

Thread A: read(fd, buffer, size);  // Starts read on file X
Thread B: close(fd);                // Closes file X
Thread B: open("other", O_RDONLY);  // Kernel reuses 'fd' for file Y
Thread A: // Completes read from file Y! Silently wrong.

Solution: Coordinate the lifecycle of file descriptors. Use reference counting or clear ownership conventions.

Signal Handlers and Disposition

Signal handling in multi-threaded programs is notoriously complex. The key facts:

Signal handlers are process-wide: When you call signal() or sigaction(), you're setting the handler for all threads.
Signal masks are per-thread: Each thread has its own signal mask controlling which signals are blocked.
Signal delivery varies: Process-directed signals (e.g., SIGINT from Ctrl+C) go to any thread that hasn't masked the signal. Thread-directed signals (via pthread_kill) go to a specific thread.

The Threading Challenges:

Signal Handling Complexities

•Non-deterministic delivery — When SIGINT arrives, which thread handles it? Any thread that hasn't masked it. This is effectively random from the program's perspective.
•Async-signal-safety — Signal handlers can be invoked at any point, even while the thread holds locks. Only a limited set of functions are safe to call from signal handlers.
•Interrupted system calls — A signal can interrupt a blocking system call (like read), causing it to return EINTR. All threads must handle this.
•Handler reentrancy — A signal handler might be invoked while another (or the same) handler is running. Global state in handlers is dangerous.

signal_handling_threads.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
#include <pthread.h>
#include <signal.h>
#include <stdio.h>
#include <unistd.h>
 
volatile sig_atomic_t got_signal = 0;
 
void signal_handler(int sig) {
    /* Only async-signal-safe operations here! */
    /* Can't use printf, malloc, pthread_mutex_lock, etc. */
    got_signal = 1;  /* sig_atomic_t write is safe */
}
 
/* RECOMMENDED PATTERN: Dedicated signal-handling thread */
 
sigset_t signal_mask;
pthread_t signal_thread;
 
void *signal_handler_thread(void *arg) {
    int sig;
    
    while (1) {
        /* sigwait() blocks until a signal from the set is pending */
        sigwait(&signal_mask, &sig);
        
        /* Now we're in a normal thread context - can use any function */
        printf("Received signal %d\n", sig);
        
        if (sig == SIGINT) {
            printf("Initiating graceful shutdown...\n");
            /* Can safely coordinate with other threads here */
            break;
        }
    }
    return NULL;
}
 
int main() {
    /* Step 1: Block signals in main thread BEFORE creating other threads */
    sigemptyset(&signal_mask);
    sigaddset(&signal_mask, SIGINT);
    sigaddset(&signal_mask, SIGTERM);
    pthread_sigmask(SIG_BLOCK, &signal_mask, NULL);
    
    /* Step 2: Create a dedicated thread to handle signals */
    pthread_create(&signal_thread, NULL, signal_handler_thread, NULL);
    
    /* Step 3: Create worker threads - they inherit blocked signals */
    /* Workers don't need to worry about signal handling at all */
    
    pthread_join(signal_thread, NULL);
    return 0;
}

The Dedicated Signal Thread Pattern

The clean solution: block all signals in main() before creating threads. Then create a dedicated signal-handling thread that uses sigwait() to synchronously receive signals. This thread is in a normal context and can safely use any functions. Other threads never receive asynchronous signals.

Environment and Process Attributes

Several additional process-level attributes are shared among all threads:

Current Working Directory:

All threads share the process's current working directory
If one thread calls chdir("/tmp"), all threads now see /tmp as their working directory
Relative path operations in all threads are affected

Environment Variables:

getenv() and setenv() operate on the same environment
Modifying environment in one thread affects all threads
setenv() is not guaranteed to be thread-safe on all systems

User and Group IDs:

All threads share the same UID, GID, effective IDs, and supplementary groups
setuid() and similar calls affect all threads
This is essential for security: a process runs as one user

Resource Limits:

Limits set via setrlimit() (file size, CPU time, stack size, etc.) are process-wide
All threads compete for the same resources

shared_process_attributes.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
#include <pthread.h>
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <limits.h>
 
void *worker1(void *arg) {
    /* Print current working directory */
    char cwd[PATH_MAX];
    getcwd(cwd, sizeof(cwd));
    printf("Worker 1: CWD = %s\n", cwd);
    
    /* Change directory */
    chdir("/tmp");
    printf("Worker 1: Changed CWD to /tmp\n");
    
    return NULL;
}
 
void *worker2(void *arg) {
    sleep(1);  /* Wait for worker1 to change directory */
    
    /* This thread sees the new working directory! */
    char cwd[PATH_MAX];
    getcwd(cwd, sizeof(cwd));
    printf("Worker 2: CWD = %s\n", cwd);  /* Will print /tmp */
    
    /* Similarly, environment changes are visible */
    printf("Worker 2: MY_VAR = %s\n", getenv("MY_VAR") ?: "(null)");
    
    return NULL;
}
 
int main() {
    /* Set an environment variable */
    setenv("MY_VAR", "shared_value", 1);
    
    pthread_t t1, t2;
    pthread_create(&t1, NULL, worker1, NULL);
    pthread_create(&t2, NULL, worker2, NULL);
    
    pthread_join(t1, NULL);
    pthread_join(t2, NULL);
    
    return 0;
}

The chdir Problem

If any thread calls chdir(), all threads are affected. This is a common source of bugs in multi-threaded file processing. The solution: use absolute paths exclusively, or use the *at() system calls (openat, faccessat, etc.) that work relative to a directory file descriptor rather than the current directory.

Process-Wide Attributes Summary
Attribute	Shared?	Thread-Safe Modification?	Notes
Current working directory	Yes	No (use *at() functions)	chdir() affects all threads
Environment variables	Yes	Varies by platform	Prefer not to modify after startup
UID/GID	Yes	N/A (rarely changed)	Security requirement
Resource limits	Yes	Thread-safe (usually)	Affects all threads' usage
umask	Yes	Not thread-safe	Affects file creation permissions
Session/process group	Yes	N/A (set once)	Rarely modified in multi-threaded code

The Synchronization Imperative

Given the extent of resource sharing, the natural question is: how do we safely access shared resources? The answer is synchronization—ensuring that concurrent accesses don't conflict.

When Is Synchronization Needed?

When multiple threads access the same data
When at least one access is a write
When the accesses are not atomic

If all three conditions hold, you have a potential data race and must synchronize.

Synchronization Mechanisms Overview

•Mutexes (Mutual Exclusion Locks) — Only one thread can hold the lock. Others block until released. Use for protecting critical sections.
•Read-Write Locks — Multiple readers simultaneously, or one writer exclusively. Use when reads vastly outnumber writes.
•Condition Variables — Wait for a condition to become true. Use for producer-consumer, state changes.
•Semaphores — Generalized counters. Use for resource counting, limiting concurrency.
•Atomic Operations — Hardware-guaranteed atomicity for simple operations. Use for counters, flags, lock-free algorithms.
•Barriers — All threads wait until everyone reaches the barrier. Use for phased computation.

synchronization_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
#include <pthread.h>
#include <stdatomic.h>
#include <stdio.h>
 
/* Example 1: Mutex for complex data structure */
struct shared_data {
    int count;
    char buffer[1024];
    pthread_mutex_t lock;
};
 
void update_data(struct shared_data *data, const char *msg) {
    pthread_mutex_lock(&data->lock);
    /* Critical section - only one thread here at a time */
    data->count++;
    snprintf(data->buffer, sizeof(data->buffer), "%s (count=%d)", 
             msg, data->count);
    pthread_mutex_unlock(&data->lock);
}
 
/* Example 2: Atomic for simple counter */
atomic_int atomic_counter = 0;
 
void increment_atomic(void) {
    /* No lock needed - hardware guarantees atomicity */
    atomic_fetch_add(&atomic_counter, 1);
}
 
/* Example 3: Read-write lock for cache */
struct cache {
    int data[1000];
    pthread_rwlock_t rwlock;
};
 
int read_cache(struct cache *c, int index) {
    pthread_rwlock_rdlock(&c->rwlock);  /* Multiple readers OK */
    int value = c->data[index];
    pthread_rwlock_unlock(&c->rwlock);
    return value;
}
 
void write_cache(struct cache *c, int index, int value) {
    pthread_rwlock_wrlock(&c->rwlock);  /* Exclusive access */
    c->data[index] = value;
    pthread_rwlock_unlock(&c->rwlock);
}

Choosing the Right Synchronization

Match the synchronization mechanism to the access pattern. Mutexes for general protection. Atomics for simple counters (faster, no kernel involvement). Read-write locks when reads dominate. Condition variables when threads must wait for state changes. Over-synchronizing hurts performance; under-synchronizing causes bugs.

Summary: Shared Resources

We've thoroughly examined the resources that threads share within a process. Let's consolidate the key insights:

Key Takeaways

•Code segment is shared and read-only — Multiple threads safely execute the same functions; each has its own stack frame for local state.
•Data segment (globals/statics) is shared and writable — Concurrent access without synchronization causes data races. This includes function-static variables.
•Heap is shared — Memory allocated by one thread is accessible by all. Ownership conventions prevent use-after-free and double-free bugs.
•File descriptors are shared — All threads see the same files, with shared file offsets. Coordinate I/O with locks or use positioned I/O (pread/pwrite).
•Signal handlers are process-wide — Use the dedicated signal thread pattern for safe, clean signal handling in multi-threaded programs.
•Environment, CWD, and credentials are shared — Modifications affect all threads. Prefer absolute paths and avoid environment modifications at runtime.
•Synchronization is mandatory — Shared mutable data requires explicit synchronization. Choose the appropriate mechanism for your access pattern.

What's Next:

Having explored what threads share, we'll now examine what threads keep private. The next page covers Thread-Specific Resources—the stack, registers, thread-local storage, and other per-thread state that enables threads to work independently despite sharing so much else.

Shared Resources Mastered

You now have deep knowledge of every category of shared resource in a multi-threaded program—and the synchronization challenges each presents. This understanding is essential for writing correct, efficient concurrent code.

3 / 5

Loading learning content...

Operating SystemsThread Concepts

Thread Fundamentals

LevelIntermediate

Duration60 mins

TopicThread Concepts

3 / 5

Shared Resources

The Shared Everything Model

The resources threads share fall into several categories:

Memory resources: Code segment, data segment, heap
I/O resources: File descriptors, sockets, pipes
Process-level state: Signal handlers, working directory, environment
System resources: User/group IDs, resource limits, credentials

Each shared resource introduces specific considerations for concurrent access. A misunderstanding of any one can lead to subtle, difficult-to-debug problems.

What You Will Learn

The Code Segment (Text Section)

Why It's Shared:

There's only one copy of the program code in memory, and all threads read from it. This is both efficient and logical:

Memory efficiency: N threads don't need N copies of the program
Cache efficiency: Instruction cache hits benefit all threads
Logical necessity: Threads execute the same program

Is It Safe?

The code segment is read-only (marked as executable but not writable in the memory map). Threads cannot modify it at runtime. This means:

No synchronization needed for code access
Multiple threads can execute the same function simultaneously
Each thread has its own program counter, so they can be at different points

code_segment_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
#include <pthread.h>
#include <stdio.h>
 
/* This function exists once in the code segment */
void shared_function(int thread_id, int value) {
    /* The function's code is shared, but: */
    /* - 'thread_id' and 'value' are on each thread's stack (private) */
    /* - The CPU registers holding them are per-thread (private) */
    
    int local_result = value * 2;  /* Stack variable - private to this call */
    
    printf("Thread %d computed: %d\n", thread_id, local_result);
    /* printf's code is also shared, but each call is independent */
}
 
void *worker(void *arg) {
    int id = *(int*)arg;
    
    /* Multiple threads call the same function simultaneously */
    /* Each has its own stack frame, so they don't interfere */
    for (int i = 0; i < 3; i++) {
        shared_function(id, i + id * 100);
    }
    
    return NULL;
}
 
int main() {
    pthread_t threads[3];
    int ids[] = {1, 2, 3};
    
    for (int i = 0; i < 3; i++) {
        pthread_create(&threads[i], NULL, worker, &ids[i]);
    }
    
    for (int i = 0; i < 3; i++) {
        pthread_join(threads[i], NULL);
    }
    
    return 0;
}

Functions Are Reentrant by Design

Position-Independent Code (PIC):

The Data Segment (Global and Static Data)

The data segment contains global and static variables. This includes:

Initialized data section (.data): Global/static variables with explicit initial values
Uninitialized data section (.bss): Global/static variables initialized to zero

Unlike the code segment, the data segment is read-write, and all threads can both read and modify it.

Why It's Shared:

Global and static variables are meant to have a single instance visible throughout the program. In a single-threaded world, this is straightforward. With threads, it becomes a source of complexity.

The Danger:

Concurrent access to shared mutable data without synchronization causes data races—one of the most common and pernicious bugs in concurrent programming.

data_race_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
#include <pthread.h>
#include <stdio.h>
 
/* Global variable - shared between all threads */
int global_counter = 0;  /* Stored in .bss (initialized to zero) */
 
/* Static variable - also shared! */
static int static_counter = 100;  /* Stored in .data */
 
void *unsafe_increment(void *arg) {
    for (int i = 0; i < 1000000; i++) {
        /* DATA RACE: Two threads may read the same value, */
        /* compute the same result, and write it back, */
        /* losing one increment. */
        global_counter++;
        
        /* This looks like one operation but is actually: */
        /* 1. Read global_counter from memory into register */
        /* 2. Add 1 to register */
        /* 3. Write register back to global_counter */
        /* A context switch between any steps causes lost updates */
    }
    return NULL;
}
 
int main() {
    pthread_t t1, t2;
    
    printf("Initial counter: %d\n", global_counter);
    
    pthread_create(&t1, NULL, unsafe_increment, NULL);
    pthread_create(&t2, NULL, unsafe_increment, NULL);
    
    pthread_join(t1, NULL);
    pthread_join(t2, NULL);
    
    printf("Final counter: %d\n", global_counter);
    printf("Expected: 2000000\n");
    printf("Difference (lost updates): %d\n", 2000000 - global_counter);
    
    return 0;
}
/* Typical output:
 * Initial counter: 0
 * Final counter: 1234567  (some value less than 2000000)
 * Expected: 2000000
 * Difference (lost updates): 765433
 */

Types of Problematic Global/Static Data

•Simple counters and flags — As shown above, even counter++ is not atomic and can lose updates.
•State variables — Global state machines where threads read and update state can have race conditions.
•Configuration structures — If one thread modifies config while another reads, partial updates can be seen.
•Buffers and arrays — Multiple threads writing to the same buffer without coordination corrupts data.
•Function-static variables — A static variable inside a function persists across calls and is shared by all threads calling that function.

Static Variables in Functions

static_function_danger.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
/* Classic thread-unsafe pattern: static buffer in function */
char *get_timestamp(void) {
    static char buffer[64];  /* SHARED between all callers! */
    
    time_t now = time(NULL);
    strftime(buffer, sizeof(buffer), "%Y-%m-%d %H:%M:%S", localtime(&now));
    
    return buffer;
    /* If two threads call this simultaneously, they overwrite */
    /* each other's buffer contents! */
}
 
/* Thread-safe alternative: caller provides buffer */
void get_timestamp_safe(char *buffer, size_t size) {
    time_t now = time(NULL);
    struct tm tm_buf;
    localtime_r(&now, &tm_buf);  /* Thread-safe localtime */
    strftime(buffer, size, "%Y-%m-%d %H:%M:%S", &tm_buf);
}

The Heap (Dynamic Memory)

Key Characteristics:

All threads can allocate from the same heap
Memory allocated by one thread can be freed by another
Memory allocated by one thread can be used by another
The heap allocator itself must be thread-safe

Heap Allocator Thread Safety:

Lock contention when many threads allocate/free simultaneously
Per-thread arenas use more memory but reduce contention

Common Heap Allocators and Their Threading Strategies
Allocator	Strategy	Characteristics
glibc malloc	Arenas with locks	Good general-purpose performance; arenas reduce contention
jemalloc (FreeBSD, Rust)	Thread-local caches + arenas	Excellent multi-threaded performance; used by Firefox, Redis
tcmalloc (Google)	Thread-caching	Very fast for allocation-heavy workloads
mimalloc (Microsoft)	Free-list sharding	Excellent performance with minimal memory overhead

heap_sharing_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
#include <pthread.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
 
/* Producer allocates on heap and passes pointer to consumer */
typedef struct {
    int id;
    char message[256];
} Task;
 
Task *task_queue[100];
int queue_head = 0, queue_tail = 0;
pthread_mutex_t queue_lock = PTHREAD_MUTEX_INITIALIZER;
pthread_cond_t queue_not_empty = PTHREAD_COND_INITIALIZER;
 
void *producer(void *arg) {
    for (int i = 0; i < 10; i++) {
        /* Allocate on heap - this memory is accessible to all threads */
        Task *task = malloc(sizeof(Task));
        task->id = i;
        snprintf(task->message, sizeof(task->message), 
                 "Task %d from producer", i);
        
        pthread_mutex_lock(&queue_lock);
        task_queue[queue_tail++] = task;  /* Hand pointer to consumer */
        pthread_cond_signal(&queue_not_empty);
        pthread_mutex_unlock(&queue_lock);
    }
    return NULL;
}
 
void *consumer(void *arg) {
    for (int i = 0; i < 10; i++) {
        pthread_mutex_lock(&queue_lock);
        while (queue_head == queue_tail) {
            pthread_cond_wait(&queue_not_empty, &queue_lock);
        }
        Task *task = task_queue[queue_head++];
        pthread_mutex_unlock(&queue_lock);
        
        /* Use the heap-allocated task from producer */
        printf("Consumer received: %s\n", task->message);
        
        /* Free memory allocated by another thread - perfectly valid */
        free(task);
    }
    return NULL;
}

Common Heap-Related Threading Errors

•Double free — Thread A frees memory, thread B doesn't know and also frees it. Heap corruption ensues.
•Use after free — Thread A frees memory, thread B continues using the pointer. Undefined behavior.
•Memory leaks — Thread A allocates, but the pointer is lost when the thread exits before anyone frees it.
•Data races on heap objects — Just because the object is on the heap doesn't make access to its contents thread-safe.
•False sharing — Two threads modify different objects that happen to share a cache line, causing performance degradation.

Ownership Transfer Pattern

File Descriptors and I/O Resources

All threads in a process share the file descriptor table. This means:

A file opened by any thread is accessible by all threads
The file offset (position for reading/writing) is shared
A file closed by any thread is closed for all threads
Sockets, pipes, and other I/O resources are similarly shared

The Shared File Offset Problem:

When multiple threads read from or write to the same file descriptor, they share the file offset. This leads to unpredictable behavior:

shared_fd_problem.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
#include <pthread.h>
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
 
int shared_fd;  /* File descriptor shared by all threads */
 
void *writer(void *arg) {
    int id = *(int*)arg;
    char buffer[100];
    
    for (int i = 0; i < 5; i++) {
        snprintf(buffer, sizeof(buffer), "Thread %d: Message %d\n", id, i);
        
        /* PROBLEM: These operations are NOT atomic together */
        /* Thread could be preempted between lseek and write */
        /* or between two write() calls */
        
        write(shared_fd, buffer, strlen(buffer));
        /* After each write, file offset advances */
        /* But another thread might write in between! */
    }
    return NULL;
}
 
/* Result: Garbled, interleaved output in the file */
 
/* Solution 1: Use mutex to serialize access */
pthread_mutex_t fd_lock = PTHREAD_MUTEX_INITIALIZER;
 
void *safe_writer(void *arg) {
    int id = *(int*)arg;
    char buffer[100];
    
    for (int i = 0; i < 5; i++) {
        snprintf(buffer, sizeof(buffer), "Thread %d: Message %d\n", id, i);
        
        pthread_mutex_lock(&fd_lock);
        write(shared_fd, buffer, strlen(buffer));
        pthread_mutex_unlock(&fd_lock);
    }
    return NULL;
}
 
/* Solution 2: Use pwrite() for atomic positioned write */
void *atomic_writer(void *arg) {
    int id = *(int*)arg;
    static _Atomic off_t global_offset = 0;  /* Atomic offset counter */
    char buffer[100];
    
    for (int i = 0; i < 5; i++) {
        int len = snprintf(buffer, sizeof(buffer), 
                           "Thread %d: Message %d\n", id, i);
        
        off_t my_offset = atomic_fetch_add(&global_offset, len);
        pwrite(shared_fd, buffer, len, my_offset);  /* Atomic at offset */
    }
    return NULL;
}

Shared I/O Resource Implications
Resource Type	Sharing Behavior	Thread Safety Concern
Regular files	Shared offset, shared content	Interleaved reads/writes, race on offset
Pipes	Shared read/write ends	Interleaved data, partial reads
Sockets	Shared connection state	Interleaved sends, protocol corruption
stdin/stdout/stderr	fd 0, 1, 2 shared	Interleaved console output
Directory handles	Shared position in readdir	Missing/duplicate entries in listing

Thread-Safe I/O Strategies

Close-While-Using Hazard:

Thread A: read(fd, buffer, size);  // Starts read on file X
Thread B: close(fd);                // Closes file X
Thread B: open("other", O_RDONLY);  // Kernel reuses 'fd' for file Y
Thread A: // Completes read from file Y! Silently wrong.

Solution: Coordinate the lifecycle of file descriptors. Use reference counting or clear ownership conventions.

Signal Handlers and Disposition

Signal handling in multi-threaded programs is notoriously complex. The key facts:

Signal handlers are process-wide: When you call signal() or sigaction(), you're setting the handler for all threads.
Signal masks are per-thread: Each thread has its own signal mask controlling which signals are blocked.
Signal delivery varies: Process-directed signals (e.g., SIGINT from Ctrl+C) go to any thread that hasn't masked the signal. Thread-directed signals (via pthread_kill) go to a specific thread.

The Threading Challenges:

Signal Handling Complexities

•Non-deterministic delivery — When SIGINT arrives, which thread handles it? Any thread that hasn't masked it. This is effectively random from the program's perspective.
•Async-signal-safety — Signal handlers can be invoked at any point, even while the thread holds locks. Only a limited set of functions are safe to call from signal handlers.
•Interrupted system calls — A signal can interrupt a blocking system call (like read), causing it to return EINTR. All threads must handle this.
•Handler reentrancy — A signal handler might be invoked while another (or the same) handler is running. Global state in handlers is dangerous.

signal_handling_threads.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
#include <pthread.h>
#include <signal.h>
#include <stdio.h>
#include <unistd.h>
 
volatile sig_atomic_t got_signal = 0;
 
void signal_handler(int sig) {
    /* Only async-signal-safe operations here! */
    /* Can't use printf, malloc, pthread_mutex_lock, etc. */
    got_signal = 1;  /* sig_atomic_t write is safe */
}
 
/* RECOMMENDED PATTERN: Dedicated signal-handling thread */
 
sigset_t signal_mask;
pthread_t signal_thread;
 
void *signal_handler_thread(void *arg) {
    int sig;
    
    while (1) {
        /* sigwait() blocks until a signal from the set is pending */
        sigwait(&signal_mask, &sig);
        
        /* Now we're in a normal thread context - can use any function */
        printf("Received signal %d\n", sig);
        
        if (sig == SIGINT) {
            printf("Initiating graceful shutdown...\n");
            /* Can safely coordinate with other threads here */
            break;
        }
    }
    return NULL;
}
 
int main() {
    /* Step 1: Block signals in main thread BEFORE creating other threads */
    sigemptyset(&signal_mask);
    sigaddset(&signal_mask, SIGINT);
    sigaddset(&signal_mask, SIGTERM);
    pthread_sigmask(SIG_BLOCK, &signal_mask, NULL);
    
    /* Step 2: Create a dedicated thread to handle signals */
    pthread_create(&signal_thread, NULL, signal_handler_thread, NULL);
    
    /* Step 3: Create worker threads - they inherit blocked signals */
    /* Workers don't need to worry about signal handling at all */
    
    pthread_join(signal_thread, NULL);
    return 0;
}

The Dedicated Signal Thread Pattern

Environment and Process Attributes

Several additional process-level attributes are shared among all threads:

Current Working Directory:

All threads share the process's current working directory
If one thread calls chdir("/tmp"), all threads now see /tmp as their working directory
Relative path operations in all threads are affected

Environment Variables:

getenv() and setenv() operate on the same environment
Modifying environment in one thread affects all threads
setenv() is not guaranteed to be thread-safe on all systems

User and Group IDs:

All threads share the same UID, GID, effective IDs, and supplementary groups
setuid() and similar calls affect all threads
This is essential for security: a process runs as one user

Resource Limits:

Limits set via setrlimit() (file size, CPU time, stack size, etc.) are process-wide
All threads compete for the same resources

shared_process_attributes.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
#include <pthread.h>
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <limits.h>
 
void *worker1(void *arg) {
    /* Print current working directory */
    char cwd[PATH_MAX];
    getcwd(cwd, sizeof(cwd));
    printf("Worker 1: CWD = %s\n", cwd);
    
    /* Change directory */
    chdir("/tmp");
    printf("Worker 1: Changed CWD to /tmp\n");
    
    return NULL;
}
 
void *worker2(void *arg) {
    sleep(1);  /* Wait for worker1 to change directory */
    
    /* This thread sees the new working directory! */
    char cwd[PATH_MAX];
    getcwd(cwd, sizeof(cwd));
    printf("Worker 2: CWD = %s\n", cwd);  /* Will print /tmp */
    
    /* Similarly, environment changes are visible */
    printf("Worker 2: MY_VAR = %s\n", getenv("MY_VAR") ?: "(null)");
    
    return NULL;
}
 
int main() {
    /* Set an environment variable */
    setenv("MY_VAR", "shared_value", 1);
    
    pthread_t t1, t2;
    pthread_create(&t1, NULL, worker1, NULL);
    pthread_create(&t2, NULL, worker2, NULL);
    
    pthread_join(t1, NULL);
    pthread_join(t2, NULL);
    
    return 0;
}

The chdir Problem

Process-Wide Attributes Summary
Attribute	Shared?	Thread-Safe Modification?	Notes
Current working directory	Yes	No (use *at() functions)	chdir() affects all threads
Environment variables	Yes	Varies by platform	Prefer not to modify after startup
UID/GID	Yes	N/A (rarely changed)	Security requirement
Resource limits	Yes	Thread-safe (usually)	Affects all threads' usage
umask	Yes	Not thread-safe	Affects file creation permissions
Session/process group	Yes	N/A (set once)	Rarely modified in multi-threaded code

The Synchronization Imperative

Given the extent of resource sharing, the natural question is: how do we safely access shared resources? The answer is synchronization—ensuring that concurrent accesses don't conflict.

When Is Synchronization Needed?

When multiple threads access the same data
When at least one access is a write
When the accesses are not atomic

If all three conditions hold, you have a potential data race and must synchronize.

Synchronization Mechanisms Overview

•Mutexes (Mutual Exclusion Locks) — Only one thread can hold the lock. Others block until released. Use for protecting critical sections.
•Read-Write Locks — Multiple readers simultaneously, or one writer exclusively. Use when reads vastly outnumber writes.
•Condition Variables — Wait for a condition to become true. Use for producer-consumer, state changes.
•Semaphores — Generalized counters. Use for resource counting, limiting concurrency.
•Atomic Operations — Hardware-guaranteed atomicity for simple operations. Use for counters, flags, lock-free algorithms.
•Barriers — All threads wait until everyone reaches the barrier. Use for phased computation.

synchronization_example.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
#include <pthread.h>
#include <stdatomic.h>
#include <stdio.h>
 
/* Example 1: Mutex for complex data structure */
struct shared_data {
    int count;
    char buffer[1024];
    pthread_mutex_t lock;
};
 
void update_data(struct shared_data *data, const char *msg) {
    pthread_mutex_lock(&data->lock);
    /* Critical section - only one thread here at a time */
    data->count++;
    snprintf(data->buffer, sizeof(data->buffer), "%s (count=%d)", 
             msg, data->count);
    pthread_mutex_unlock(&data->lock);
}
 
/* Example 2: Atomic for simple counter */
atomic_int atomic_counter = 0;
 
void increment_atomic(void) {
    /* No lock needed - hardware guarantees atomicity */
    atomic_fetch_add(&atomic_counter, 1);
}
 
/* Example 3: Read-write lock for cache */
struct cache {
    int data[1000];
    pthread_rwlock_t rwlock;
};
 
int read_cache(struct cache *c, int index) {
    pthread_rwlock_rdlock(&c->rwlock);  /* Multiple readers OK */
    int value = c->data[index];
    pthread_rwlock_unlock(&c->rwlock);
    return value;
}
 
void write_cache(struct cache *c, int index, int value) {
    pthread_rwlock_wrlock(&c->rwlock);  /* Exclusive access */
    c->data[index] = value;
    pthread_rwlock_unlock(&c->rwlock);
}

Choosing the Right Synchronization

Summary: Shared Resources

We've thoroughly examined the resources that threads share within a process. Let's consolidate the key insights:

Key Takeaways

•Code segment is shared and read-only — Multiple threads safely execute the same functions; each has its own stack frame for local state.
•Data segment (globals/statics) is shared and writable — Concurrent access without synchronization causes data races. This includes function-static variables.
•Heap is shared — Memory allocated by one thread is accessible by all. Ownership conventions prevent use-after-free and double-free bugs.
•File descriptors are shared — All threads see the same files, with shared file offsets. Coordinate I/O with locks or use positioned I/O (pread/pwrite).
•Signal handlers are process-wide — Use the dedicated signal thread pattern for safe, clean signal handling in multi-threaded programs.
•Environment, CWD, and credentials are shared — Modifications affect all threads. Prefer absolute paths and avoid environment modifications at runtime.
•Synchronization is mandatory — Shared mutable data requires explicit synchronization. Choose the appropriate mechanism for your access pattern.

What's Next:

Shared Resources Mastered

3 / 5