procfs: Add a procfs

Userland can now interface with processes by writing/reading from
`/proc/<pid>/<entry>`

It can send signals, for example `echo 15 > /proc/1/signal`

7 files changed, 212 insertions, 1 deletions

diff --git a/kernel/Makefile b/kernel/Makefile
index 34b9b30..3b25dbc 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -1,6 +1,6 @@
 CC="i686-sb-gcc"
 AS="i686-sb-as"
-OBJ = arch/i386/boot.o init/kernel.o cpu/gdt.o cpu/reload_gdt.o cpu/idt.o cpu/io.o libc/stdio/print.o drivers/keyboard.o log.o drivers/pit.o libc/string/memcpy.o libc/string/strlen.o libc/string/memcmp.o drivers/ata.o libc/string/memset.o cpu/syscall.o  read_eip.o libc/exit/assert.o process.o libc/string/strcpy.o arch/i386/mmu.o kmalloc.o fs/ext2.o fs/vfs.o fs/devfs.o cpu/spinlock.o random.o libc/string/strcmp.o crypto/ChaCha20/chacha20.o crypto/SHA1/sha1.o fs/tmpfs.o libc/string/isequal.o drivers/pst.o kubsan.o drivers/serial.o socket.o poll.o fs/fifo.o hashmap/hashmap.o fs/shm.o elf.o sched/scheduler.o libc/string/copy.o drivers/mouse.o libc/string/strlcpy.o libc/string/strcat.o drivers/vbe.o drivers/pci.o drivers/rtl8139.o network/ethernet.o network/arp.o network/bytes.o network/ipv4.o network/udp.o signal.o network/tcp.o drivers/ahci.o crypto/xoshiro256plusplus/xoshiro256plusplus.o arch/i386/interrupts.o cpu/isr.o lib/stack.o lib/buffered_write.o lib/list.o cpu/arch_inst.o cpu/int_syscall.o lib/ringbuffer.o lib/relist.o arch/i386/tsc.o arch/i386/asm_tsc.o drivers/cmos.o timer.o queue.o fonts.o drivers/ac97.o audio.o libc/ctype/isdigit.o libc/stdlib/parsenum.o libc/ctype/tolower.o lib/sv.o lib/sb.o libc/string/memmove.o libc/ctype/isspace.o
+OBJ = arch/i386/boot.o init/kernel.o cpu/gdt.o cpu/reload_gdt.o cpu/idt.o cpu/io.o libc/stdio/print.o drivers/keyboard.o log.o drivers/pit.o libc/string/memcpy.o libc/string/strlen.o libc/string/memcmp.o drivers/ata.o libc/string/memset.o cpu/syscall.o  read_eip.o libc/exit/assert.o process.o libc/string/strcpy.o arch/i386/mmu.o kmalloc.o fs/ext2.o fs/vfs.o fs/devfs.o cpu/spinlock.o random.o libc/string/strcmp.o crypto/ChaCha20/chacha20.o crypto/SHA1/sha1.o fs/tmpfs.o libc/string/isequal.o drivers/pst.o kubsan.o drivers/serial.o socket.o poll.o fs/fifo.o hashmap/hashmap.o fs/shm.o elf.o sched/scheduler.o libc/string/copy.o drivers/mouse.o libc/string/strlcpy.o libc/string/strcat.o drivers/vbe.o drivers/pci.o drivers/rtl8139.o network/ethernet.o network/arp.o network/bytes.o network/ipv4.o network/udp.o signal.o network/tcp.o drivers/ahci.o crypto/xoshiro256plusplus/xoshiro256plusplus.o arch/i386/interrupts.o cpu/isr.o lib/stack.o lib/buffered_write.o lib/list.o cpu/arch_inst.o cpu/int_syscall.o lib/ringbuffer.o lib/relist.o arch/i386/tsc.o arch/i386/asm_tsc.o drivers/cmos.o timer.o queue.o fonts.o drivers/ac97.o audio.o libc/ctype/isdigit.o libc/stdlib/parsenum.o libc/ctype/tolower.o lib/sv.o lib/sb.o libc/string/memmove.o libc/ctype/isspace.o fs/procfs.o
 CFLAGS = -std=c99 -O0 -fsanitize=vla-bound,shift-exponent,pointer-overflow,shift,signed-integer-overflow,bounds -ggdb -ffreestanding -Wall -Wextra -Wno-int-conversion -Wno-unused-parameter -Werror -mgeneral-regs-only -Wimplicit-fallthrough -I./libc/include/ -I. -Wno-pointer-sign -DKERNEL
 LDFLAGS=
 #CFLAGS = -std=c99 -O3 -flto -ggdb -ffreestanding -Wall -Wextra -Wno-int-conversion -Wno-unused-parameter -Werror -mgeneral-regs-only -Wimplicit-fallthrough -I./libc/include/ -I. -Wno-pointer-sign -DKERNEL
diff --git a/kernel/cpu/syscall.c b/kernel/cpu/syscall.c
index a78e63e..31f1fc5 100644
--- a/kernel/cpu/syscall.c
+++ b/kernel/cpu/syscall.c
@@ -427,6 +427,8 @@ void syscall_wait(int *status) {
     if (status) {
       *status = current_task->child_rc;
     }
+    process_remove_reference(current_task->child);
+    current_task->child = NULL;
     return;
   }
   do {
@@ -444,6 +446,10 @@ void syscall_wait(int *status) {
   if (status) {
     *status = current_task->child_rc;
   }
+  if (current_task->child->dead) {
+    process_remove_reference(current_task->child);
+    current_task->child = NULL;
+  }
 }
 
 int syscall_fork(void) {
diff --git a/kernel/fs/procfs.c b/kernel/fs/procfs.c
new file mode 100644
index 0000000..f6a4fef
--- /dev/null
+++ b/kernel/fs/procfs.c
@@ -0,0 +1,185 @@
+#include <errno.h>
+#include <fs/procfs.h>
+#include <lib/sb.h>
+#include <lib/sv.h>
+#include <stdio.h>
+
+#define ARRAY_LENGTH(a) (sizeof(a) / sizeof((a)[0]))
+
+vfs_inode_t *procfs_open(const char *p);
+
+void procfs_close(vfs_fd_t *fd) {
+  return;
+}
+
+void process_close(vfs_fd_t *fd) {
+  process_t *p = fd->inode->internal_object;
+  if (!p) {
+    return;
+  }
+  process_remove_reference(p);
+}
+
+int procfs_read(u8 *buffer, u64 offset, u64 len, vfs_fd_t *fd) {
+  struct sb builder;
+  sb_init_buffer(&builder, buffer, len);
+
+  size_t read_amount = 0;
+
+  process_t *p = ready_queue;
+  for (; p; p = p->next) {
+    struct dirent entry;
+    entry.d_ino = p->pid;
+    ksnprintf(entry.d_name, sizeof(entry.d_name), "%u", p->pid);
+
+    if (read_amount >= offset) {
+      if (0 == sb_append_buffer(&builder, (u8 *)&entry, sizeof(entry))) {
+        break;
+      }
+    }
+    read_amount += sizeof(struct dirent);
+  }
+  return sv_length(SB_TO_SV(builder));
+}
+
+#define PROCESS_ROOT 0
+#define PROCESS_SIGNAL 1
+#define PROCESS_NAME 2
+
+struct dirent process_entries[] = {
+    {
+        .d_ino = PROCESS_ROOT,
+        .d_name = "",
+    },
+    {
+        .d_ino = PROCESS_ROOT,
+        .d_name = ".",
+    },
+    {
+        .d_ino = PROCESS_SIGNAL,
+        .d_name = "signal",
+    },
+    {
+        .d_ino = PROCESS_NAME,
+        .d_name = "name",
+    },
+};
+
+int process_read(u8 *buffer, u64 offset, u64 len, vfs_fd_t *fd) {
+  process_t *p = fd->inode->internal_object;
+
+  struct sb builder;
+  sb_init_buffer(&builder, buffer, len);
+
+  int id = fd->inode->inode_num;
+  if (PROCESS_ROOT == id) {
+    size_t read_amount = 0;
+    for (size_t i = 0; i < ARRAY_LENGTH(process_entries); i++) {
+      if (0 == strlen(process_entries[i].d_name)) {
+        continue;
+      }
+      if (read_amount >= offset) {
+        if (0 == sb_append_buffer(&builder, (u8 *)&process_entries[i],
+                                  sizeof(struct dirent))) {
+          break;
+        }
+      }
+      read_amount += sizeof(struct dirent);
+    }
+    return sv_length(SB_TO_SV(builder));
+  }
+  if (PROCESS_NAME == id) {
+    struct sv program_name = C_TO_SV(p->program_name);
+    sv_take(program_name, &program_name, offset);
+    sb_append_sv(&builder, program_name);
+    return sv_length(SB_TO_SV(builder));
+  }
+  return -EBADF;
+}
+
+int process_write(u8 *buffer, u64 offset, u64 len, vfs_fd_t *fd) {
+  (void)offset;
+  process_t *p = fd->inode->internal_object;
+
+  int id = fd->inode->inode_num;
+  if (PROCESS_SIGNAL == id) {
+    struct sv t = sv_init(buffer, len);
+    int got_num;
+    u64 signal = sv_parse_unsigned_number(t, NULL, &got_num);
+    if (got_num) {
+      signal_process(p, signal);
+    }
+    // TODO: Add the ability to write "SIGTERM", "SIGKILL" etc without
+    // writing the number.
+    return 0;
+  }
+  return -EBADF;
+}
+
+vfs_inode_t *open_process(u64 pid, int id) {
+  process_t *process = NULL;
+  process_t *t = ready_queue;
+  for (; t; t = t->next) {
+    if ((u64)t->pid == pid) {
+      process = t;
+      break;
+    }
+  }
+  if (!process) {
+    return NULL;
+  }
+
+  vfs_inode_t *inode = vfs_create_inode(
+      id /*inode_num*/, 0 /*type*/, 0 /*has_data*/, 0 /*can_write*/,
+      0 /*is_open*/, 0, process /*internal_object*/, 0 /*file_size*/,
+      procfs_open, NULL /*create_file*/, process_read, process_write,
+      process_close, NULL /*create_directory*/, NULL /*get_vm_object*/,
+      NULL /*truncate*/, NULL /*stat*/, NULL /*send_signal*/, NULL /*connect*/);
+  if (!inode) {
+    return NULL;
+  }
+  process->reference_count++;
+  return inode;
+}
+
+vfs_inode_t *procfs_open(const char *p) {
+  struct sv path = C_TO_SV(p);
+
+  if ((sv_try_eat(path, &path, C_TO_SV("/")) && 0 == sv_length(path)) ||
+      0 == sv_length(path)) {
+    return vfs_create_inode(
+        0 /*inode_num*/, 0 /*type*/, 0 /*has_data*/, 0 /*can_write*/,
+        0 /*is_open*/, 0, NULL /*internal_object*/, 0 /*file_size*/,
+        procfs_open, NULL /*create_file*/, procfs_read, NULL /* write */,
+        procfs_close, NULL /*create_directory*/, NULL /*get_vm_object*/,
+        NULL /*truncate*/, NULL /*stat*/, NULL /*send_signal*/,
+        NULL /*connect*/);
+  }
+
+  int got_num;
+  u64 pid = sv_parse_unsigned_number(path, &path, &got_num);
+  if (!got_num) {
+    return NULL;
+  }
+
+  // NOTE: There should only be one '/' since the previous vfs_open will
+  // call vfs_clean_path first. This is a bit hacky and that code might
+  // change later and as a result introduce bugs here.
+  sv_try_eat(path, &path, C_TO_SV("/"));
+  for (size_t i = 0; i < ARRAY_LENGTH(process_entries); i++) {
+    if (sv_eq(path, C_TO_SV(process_entries[i].d_name))) {
+      return open_process(pid, process_entries[i].d_ino);
+    }
+  }
+  return NULL;
+}
+
+vfs_inode_t *procfs_mount(void) {
+  vfs_inode_t *inode = vfs_create_inode(
+      0 /*inode_num*/, 0 /*type*/, 0 /*has_data*/, 0 /*can_write*/,
+      0 /*is_open*/, 0, NULL /*internal_object*/, 0 /*file_size*/, procfs_open,
+      NULL /*create_file*/, procfs_read, NULL /* write */, procfs_close,
+      NULL /*create_directory*/, NULL /*get_vm_object*/, NULL /*truncate*/,
+      NULL /*stat*/, NULL /*send_signal*/, NULL /*connect*/);
+  return inode;
+}
diff --git a/kernel/fs/procfs.h b/kernel/fs/procfs.h
new file mode 100644
index 0000000..8715789
--- /dev/null
+++ b/kernel/fs/procfs.h
@@ -0,0 +1,3 @@
+#include <fs/vfs.h>
+
+vfs_inode_t *procfs_mount(void);
diff --git a/kernel/init/kernel.c b/kernel/init/kernel.c
index 79c6366..5bf959b 100644
--- a/kernel/init/kernel.c
+++ b/kernel/init/kernel.c
@@ -17,6 +17,7 @@
 #include <fcntl.h>
 #include <fs/devfs.h>
 #include <fs/ext2.h>
+#include <fs/procfs.h>
 #include <fs/shm.h>
 #include <fs/vfs.h>
 #include <interrupts.h>
@@ -95,6 +96,7 @@ void kernel_main(u32 kernel_end, unsigned long magic, unsigned long addr,
 
   global_socket_init();
 
+  vfs_mount("/proc", procfs_mount());
   vfs_mount("/dev", devfs_mount());
   assert(ahci_init());
   vfs_inode_t *ext2_mount_point = ext2_mount();
diff --git a/kernel/sched/scheduler.c b/kernel/sched/scheduler.c
index d0452fd..1b36e8a 100644
--- a/kernel/sched/scheduler.c
+++ b/kernel/sched/scheduler.c
@@ -98,6 +98,7 @@ process_t *create_process(process_t *p, u32 esp, u32 eip) {
   if (!r) {
     return NULL;
   }
+  r->reference_count = 1;
   r->tcb = kcalloc(1, sizeof(struct TCB));
   if (!r->tcb) {
     kfree(r);
@@ -147,6 +148,9 @@ process_t *create_process(process_t *p, u32 esp, u32 eip) {
     r->cr3 = get_active_pagedirectory();
   }
   r->parent = p;
+  if (p) {
+    p->reference_count++;
+  }
 
   r->tcb->CR3 = r->cr3->physical_address;
 
@@ -205,6 +209,14 @@ void free_process(process_t *p) {
   mmu_free_pagedirectory(p->cr3);
 }
 
+void process_remove_reference(process_t *p) {
+  assert(0 != p->reference_count);
+  p->reference_count--;
+  if (0 == p->reference_count) {
+    kfree(p);
+  }
+}
+
 void exit_process(process_t *p, int status) {
   assert(p->pid != 1);
   disable_interrupts();
diff --git a/kernel/sched/scheduler.h b/kernel/sched/scheduler.h
index 9d548a3..d03277e 100644
--- a/kernel/sched/scheduler.h
+++ b/kernel/sched/scheduler.h
@@ -69,6 +69,8 @@ struct Process {
   struct list write_list;
   struct list disconnect_list;
 
+  int reference_count;
+
   struct list event_queue;
 
   struct stack restore_context_stack;
@@ -98,4 +100,5 @@ int get_task_from_pid(pid_t pid, process_t **out);
 void free_process(process_t *p);
 void *get_free_virtual_memory(size_t length);
 void signal_process(process_t *p, int sig);
+void process_remove_reference(process_t *p);
 #endif