#include "userprog/process.h" #include #include #include #include #include #include #include #include #include "userprog/gdt.h" #include "userprog/pagedir.h" #include "userprog/syscall.h" #include "userprog/tss.h" #include "filesys/directory.h" #include "filesys/file.h" #include "filesys/filesys.h" #include "threads/flags.h" #include "threads/init.h" #include "threads/interrupt.h" #include "threads/synch.h" #include "threads/palloc.h" #include "threads/malloc.h" #include "threads/thread.h" #include "threads/vaddr.h" #include "threads/synch.h" #include "devices/timer.h" #include "vm/page.h" #include "vm/mmap.h" #ifdef VM #include "vm/frame.h" #endif /* Defines the native number of bytes processed by the processor (for the purposes of alignment). */ #define WORD_SIZE 4 /* Defines non-negative integer division wherein the result is always rounded up. */ #define DIV_CEIL(x, y) ((x + (y - 1)) / y) /* Keeps track of the position of pointers to user program arguments within a linked list. */ struct arg_elem { char* arg; struct list_elem elem; }; /* Holds the data required to be passed from a kernel thread to a thread that executes process_start for the purpose of starting a user process. */ struct process_start_data { char *cmd_saveptr; /* Value pointed to by 'saveptr' argument used by successive calls to strtok_r to split 'cmd' into tokens while maintaining state. */ char file_name[FNAME_MAX_LEN + 1]; /* Name of the file of the process to be started. */ bool success; /* Indicates whether the process was successfully loaded. */ struct semaphore loaded; /* Semaphore used to signal that the process has finished attempting to load. */ }; static thread_func start_process NO_RETURN; static void destruct_process_result (struct hash_elem *e, void *aux UNUSED); static bool load (const char *cmdline, void (**eip) (void), void **esp); /* Starts a new thread running a user program executed via CMD. The new thread may be scheduled (and may even exit) before process_execute() returns. Returns the new process's thread id, or TID_ERROR if the thread cannot be created. */ tid_t process_execute (const char *cmd) { char *cmd_copy; tid_t tid; struct process_start_data data; /* Make a copy of command. Otherwise there's a race between the caller and load(). */ cmd_copy = palloc_get_page (0); if (cmd_copy == NULL) return TID_ERROR; /* Imposing implicit limit that the command line arguments including the user program name fit within a single page. */ strlcpy (cmd_copy, cmd, PGSIZE); /* Retrieve first argument of command, which is the file name of the process. */ char *file_name = strtok_r (cmd_copy, " ", &data.cmd_saveptr); /* Validates that the current file to be executed can be opened/exists. */ lock_acquire (&filesys_lock); struct file *file = filesys_open (file_name); lock_release (&filesys_lock); if (file == NULL) return TID_ERROR; /* Create a new thread to execute the command, by initializing it running the function 'start_process' with the appropriate arguments. For details of arguments, see 'start_process'. */ strlcpy (data.file_name, file_name, FNAME_MAX_LEN + 1); sema_init (&data.loaded, 0); data.success = false; tid = thread_create (file_name, PRI_DEFAULT, start_process, &data); /* Wait until process file has finished attempting to load via the child thread before reporting success of starting execution. */ if (tid != TID_ERROR) { sema_down (&data.loaded); if (!data.success) tid = TID_ERROR; } palloc_free_page (cmd_copy); return tid; } static void *get_usr_kpage (enum palloc_flags flags, void *upage); static void free_usr_kpage (void *kpage); bool install_page (void *upage, void *kpage, bool writable); static bool process_init_stack (char *cmd_saveptr, void **esp, char *file_name); static void *push_to_stack (void **esp, void *data, size_t data_size); #define push_var_to_stack(esp, var) (push_to_stack (esp, &var, sizeof (var))) /* Make the current thread execute 'cmd', passing in a copy of the command string used for processing, the saveptr used by strtok_r (in order to further tokenize the same command and retrieve its arguments), the name of the file being executed, and a semaphore that calls sema_up to indicate that the 'success' variable passed to it has been updated to indicate whether the process file loading succeeded. This involves loading the specified file and calling its main () function with the specified command arguments. */ static void start_process (void *proc_start_data) { struct intr_frame if_; struct process_start_data *data = proc_start_data; /* Initialize interrupt frame and load executable. */ memset (&if_, 0, sizeof if_); if_.gs = if_.fs = if_.es = if_.ds = if_.ss = SEL_UDSEG; if_.cs = SEL_UCSEG; if_.eflags = FLAG_IF | FLAG_MBS; /* Acquire the file system lock to prevent race conditions. */ lock_acquire (&filesys_lock); struct file *exec_file = filesys_open (data->file_name); if (exec_file == NULL) { /* If the executable file cannot be opened, free resources and quit. */ lock_release (&filesys_lock); sema_up (&data->loaded); thread_exit (); } /* Deny write to the executable file to prevent writing to it and release the file system lock. */ file_deny_write (exec_file); lock_release (&filesys_lock); thread_current ()->exec_file = exec_file; /* Load the ELF executable file, and store the success of the operation in the 'success' variable in data. */ data->success = load (data->file_name, &if_.eip, &if_.esp); /* If load was sucessful, initialize user process stack and free page used to store the command that executed the process. */ if (data->success) { data->success = process_init_stack (data->cmd_saveptr, &if_.esp, data->file_name); } /* Signal that the process has finished attempting to load. */ bool success = data->success; sema_up (&data->loaded); /* If the load was unsuccessful or if it was but the stack initialization failed, exit the thread. */ if (!success) thread_exit (); /* Start the user process by simulating a return from an interrupt, implemented by intr_exit (in threads/intr-stubs.S). Because intr_exit takes all of its arguments on the stack in the form of a `struct intr_frame', we just point the stack pointer (%esp) to our stack frame and jump to it. */ asm volatile ("movl %0, %%esp; jmp intr_exit" : : "g" (&if_) : "memory"); NOT_REACHED (); } /* Helper function that initializes the stack of a newly created user process. Returns true if successful, false otherwise. */ static bool process_init_stack (char *cmd_saveptr, void **esp, char *file_name) { ASSERT (cmd_saveptr != NULL); ASSERT (esp != NULL); ASSERT (file_name != NULL); /* Load command line argument *data* to user process stack. This can't cause overflow due to enforcing that the size of command line input must fit in a page. Also keep track of pointers to the argument data within a linked list. */ struct list arg_list; list_init (&arg_list); char *arg = file_name; int arg_count = 0; while (arg != NULL) { /* filename has already been validated to be a safe-to-access string, so we can safely use strlen here. Filename has already been split from the command line arguments. */ push_to_stack (esp, arg, (strlen (arg) + 1) * sizeof (char)); /* Try to allocate memory for the argument pointer. */ struct arg_elem *arg_elem = malloc (sizeof (struct arg_elem)); if (arg_elem == NULL) { printf("ERROR: Couldn't allocate argument pointer memory for %s!\n", thread_current ()->name); return false; } /* Store the argument pointer in the linked list. */ arg_elem->arg = *esp; list_push_front (&arg_list, &arg_elem->elem); /* Increment the argument count and get the next argument. */ arg_count++; arg = strtok_r (NULL, " ", &cmd_saveptr); } /* Calculate the remaining number of bytes that need to be written to the user process stack in order to check for possible overflow. */ size_t align_size = ((unsigned int) *esp % WORD_SIZE) * sizeof (uint8_t); size_t argv_data_size = (arg_count + 1) * sizeof (char *); size_t argv_size = sizeof (char **); size_t argc_size = sizeof (int); size_t return_addr_size = sizeof (void *); size_t remaining_size = align_size + argv_data_size + argv_size + argc_size + return_addr_size; /* If pushing the rest of the data required for the stack would cause overflow, allocate as many extra pages as needed to the user process contiguously in the virtual address space below the initial page. */ int overflow_bytes = (PHYS_BASE - *esp) + remaining_size - PGSIZE; if (overflow_bytes > 0) { /* Calculate the number of pages needed to allocate. */ int pages_needed = DIV_CEIL (overflow_bytes, PGSIZE); /* Allocate the pages and map them to the user process. */ void *upage; uint8_t *kpage; for (int i = 1; i < pages_needed + 1; i++) { upage = ((uint8_t *) PHYS_BASE) - PGSIZE * (i + 1); kpage = get_usr_kpage (PAL_ZERO, upage); if (!install_page (upage, kpage, true)) return false; } } /* Align stack pointer to word size before pushing argv elements for performance. */ *esp -= align_size; /* Push a null pointer sentinel inside argv. */ char *null_sentinel = NULL; push_var_to_stack (esp, null_sentinel); /* Push pointers to process arguments from argument linked list */ struct list_elem *e = list_begin (&arg_list); struct list_elem *tail = list_tail (&arg_list); while (e != tail) { struct arg_elem *arg_elem = list_entry (e, struct arg_elem, elem); push_var_to_stack(esp, arg_elem->arg); e = list_next (e); free (arg_elem); } /* Push pointer to the start of argv array. */ char **argv = *esp; push_var_to_stack(esp, argv); /* Push the number of arguments to the stack. */ push_var_to_stack(esp, arg_count); /* Push fake return address (null pointer). */ push_var_to_stack (esp, null_sentinel); return true; } /* Helper function that pushes the first 'data_size' bytes stored in the address '*data' into the stack given a pointer to the stack pointer '**esp'. */ static void * push_to_stack (void **esp, void *data, size_t data_size) { *esp -= data_size; memcpy (*esp, data, data_size); return *esp; } /* Waits for thread TID to die and returns its exit status. * If it was terminated by the kernel (i.e. killed due to an exception), * returns -1. * If TID is invalid or if it was not a child of the calling process, or if * process_wait() has already been successfully called for the given TID, * returns -1 immediately, without waiting. * * This function will be implemented in task 2. * For now, it does nothing. */ int process_wait (tid_t child_tid) { struct thread *t = thread_current (); struct process_result fake_result; fake_result.tid = child_tid; struct hash_elem *e = hash_find (&t->child_results, &fake_result.elem); if (e == NULL) return -1; struct process_result *child_result = hash_entry (e, struct process_result, elem); /* Wait for child to die. */ sema_down (&child_result->sema); /* We need lock release in process_exit, so we need to acquire (and possibly wait) for it here to ensure we don't free the lock memory before it is released in process_exit. */ lock_acquire (&child_result->lock); /* To prevent waiting for child twice, remove it from the table. No need to use lock since this is the only thread with access to the struct process_result now. */ hash_delete (&t->child_results, &child_result->elem); /* Get the exit status of the child */ int exit_status = child_result->exit_status; /* Release the lock */ lock_release (&child_result->lock); /* Result no-longer used by parent, nor child. Deallocate it. */ free (child_result); return exit_status; } /* Free the current process's resources. */ void process_exit (void) { struct thread *cur = thread_current (); uint32_t *pd; /* Unmap all memory mapped files */ mmap_destroy (); /* Clean up all open files */ hash_destroy (&cur->open_files, fd_cleanup); /* Clean up the thread's supplemental page table. */ hash_destroy (&cur->pages, page_cleanup); /* Close the executable file, implicitly allowing it to be written to. */ if (cur->exec_file != NULL) { /* Acquire the file system lock to prevent race conditions. */ lock_acquire (&filesys_lock); file_close (cur->exec_file); lock_release (&filesys_lock); } if (cur->result != NULL) { printf ("%s: exit(%d)\n", cur->name, cur->result->exit_status); /* Update own process result. */ destruct_process_result (&cur->result->elem, cur); } /* Free child process results or signal parent's death. */ hash_destroy (&cur->child_results, destruct_process_result); /* Destroy the current process's page directory and switch back to the kernel-only page directory. */ pd = cur->pagedir; if (pd != NULL) { /* Correct ordering here is crucial. We must set cur->pagedir to NULL before switching page directories, so that a timer interrupt can't switch back to the process page directory. We must activate the base page directory before destroying the process's page directory, or our active page directory will be one that's been freed (and cleared). */ cur->pagedir = NULL; pagedir_activate (NULL); pagedir_destroy (pd); } } /* Destruct a process_result, with multi-thread awareness. If the other thread is running, simply signals death. Otherwise frees the result. */ static void destruct_process_result (struct hash_elem *e, void *aux UNUSED) { struct process_result *result = hash_entry (e, struct process_result, elem); lock_acquire (&result->lock); /* Other thread has died (and was not waited for). Free the result. */ if (sema_try_down (&result->sema)) { lock_release (&result->lock); free (result); } /* Other thread is still alive, signal via sema that parent has died. */ else { sema_up (&result->sema); lock_release (&result->lock); } } /* Sets up the CPU for running user code in the current thread. This function is called on every context switch. */ void process_activate (void) { struct thread *t = thread_current (); /* Activate thread's page tables. */ pagedir_activate (t->pagedir); /* Set thread's kernel stack for use in processing interrupts. */ tss_update (); } /* We load ELF binaries. The following definitions are taken from the ELF specification, [ELF1], more-or-less verbatim. */ /* ELF types. See [ELF1] 1-2. */ typedef uint32_t Elf32_Word, Elf32_Addr, Elf32_Off; typedef uint16_t Elf32_Half; /* For use with ELF types in printf(). */ #define PE32Wx PRIx32 /* Print Elf32_Word in hexadecimal. */ #define PE32Ax PRIx32 /* Print Elf32_Addr in hexadecimal. */ #define PE32Ox PRIx32 /* Print Elf32_Off in hexadecimal. */ #define PE32Hx PRIx16 /* Print Elf32_Half in hexadecimal. */ /* Executable header. See [ELF1] 1-4 to 1-8. This appears at the very beginning of an ELF binary. */ struct Elf32_Ehdr { unsigned char e_ident[16]; Elf32_Half e_type; Elf32_Half e_machine; Elf32_Word e_version; Elf32_Addr e_entry; Elf32_Off e_phoff; Elf32_Off e_shoff; Elf32_Word e_flags; Elf32_Half e_ehsize; Elf32_Half e_phentsize; Elf32_Half e_phnum; Elf32_Half e_shentsize; Elf32_Half e_shnum; Elf32_Half e_shstrndx; }; /* Program header. See [ELF1] 2-2 to 2-4. There are e_phnum of these, starting at file offset e_phoff (see [ELF1] 1-6). */ struct Elf32_Phdr { Elf32_Word p_type; Elf32_Off p_offset; Elf32_Addr p_vaddr; Elf32_Addr p_paddr; Elf32_Word p_filesz; Elf32_Word p_memsz; Elf32_Word p_flags; Elf32_Word p_align; }; /* Values for p_type. See [ELF1] 2-3. */ #define PT_NULL 0 /* Ignore. */ #define PT_LOAD 1 /* Loadable segment. */ #define PT_DYNAMIC 2 /* Dynamic linking info. */ #define PT_INTERP 3 /* Name of dynamic loader. */ #define PT_NOTE 4 /* Auxiliary info. */ #define PT_SHLIB 5 /* Reserved. */ #define PT_PHDR 6 /* Program header table. */ #define PT_STACK 0x6474e551 /* Stack segment. */ /* Flags for p_flags. See [ELF3] 2-3 and 2-4. */ #define PF_X 1 /* Executable. */ #define PF_W 2 /* Writable. */ #define PF_R 4 /* Readable. */ static bool setup_stack (void **esp); static bool validate_segment (const struct Elf32_Phdr *, struct file *); static bool load_segment (struct file *file, off_t ofs, uint8_t *upage, uint32_t read_bytes, uint32_t zero_bytes, bool writable); /* Loads an ELF executable from FILE_NAME into the current thread. Stores the executable's entry point into *EIP and its initial stack pointer into *ESP. Returns true if successful, false otherwise. */ bool load (const char *file_name, void (**eip) (void), void **esp) { struct thread *t = thread_current (); struct Elf32_Ehdr ehdr; struct file *file = NULL; off_t file_ofs; bool success = false; int i; lock_acquire (&filesys_lock); /* Allocate and activate page directory. */ t->pagedir = pagedir_create (); if (t->pagedir == NULL) goto done; process_activate (); /* Open executable file. */ file = filesys_open (file_name); if (file == NULL) { printf ("load: %s: open failed\n", file_name); goto done; } /* Read and verify executable header. */ if (file_read (file, &ehdr, sizeof ehdr) != sizeof ehdr || memcmp (ehdr.e_ident, "\177ELF\1\1\1", 7) || ehdr.e_type != 2 || ehdr.e_machine != 3 || ehdr.e_version != 1 || ehdr.e_phentsize != sizeof (struct Elf32_Phdr) || ehdr.e_phnum > 1024) { printf ("load: %s: error loading executable\n", file_name); goto done; } /* Read program headers. */ file_ofs = ehdr.e_phoff; for (i = 0; i < ehdr.e_phnum; i++) { struct Elf32_Phdr phdr; if (file_ofs < 0 || file_ofs > file_length (file)) goto done; file_seek (file, file_ofs); if (file_read (file, &phdr, sizeof phdr) != sizeof phdr) goto done; file_ofs += sizeof phdr; switch (phdr.p_type) { case PT_NULL: case PT_NOTE: case PT_PHDR: case PT_STACK: default: /* Ignore this segment. */ break; case PT_DYNAMIC: case PT_INTERP: case PT_SHLIB: goto done; case PT_LOAD: if (validate_segment (&phdr, file)) { bool writable = (phdr.p_flags & PF_W) != 0; uint32_t file_page = phdr.p_offset & ~PGMASK; uint32_t mem_page = phdr.p_vaddr & ~PGMASK; uint32_t page_offset = phdr.p_vaddr & PGMASK; uint32_t read_bytes, zero_bytes; if (phdr.p_filesz > 0) { /* Normal segment. Read initial part from disk and zero the rest. */ read_bytes = page_offset + phdr.p_filesz; zero_bytes = (ROUND_UP (page_offset + phdr.p_memsz, PGSIZE) - read_bytes); } else { /* Entirely zero. Don't read anything from disk. */ read_bytes = 0; zero_bytes = ROUND_UP (page_offset + phdr.p_memsz, PGSIZE); } if (!load_segment (file, file_page, (void *) mem_page, read_bytes, zero_bytes, writable)) goto done; } else goto done; break; } } /* Set up stack. */ if (!setup_stack (esp)) goto done; /* Start address. */ *eip = (void (*) (void)) ehdr.e_entry; success = true; done: /* We arrive here whether the load is successful or not. */ #ifndef VM file_close (file); #endif lock_release (&filesys_lock); return success; } /* load() helpers. */ /* Checks whether PHDR describes a valid, loadable segment in FILE and returns true if so, false otherwise. */ static bool validate_segment (const struct Elf32_Phdr *phdr, struct file *file) { /* p_offset and p_vaddr must have the same page offset. */ if ((phdr->p_offset & PGMASK) != (phdr->p_vaddr & PGMASK)) return false; /* p_offset must point within FILE. */ if (phdr->p_offset > (Elf32_Off) file_length (file)) return false; /* p_memsz must be at least as big as p_filesz. */ if (phdr->p_memsz < phdr->p_filesz) return false; /* The segment must not be empty. */ if (phdr->p_memsz == 0) return false; /* The virtual memory region must both start and end within the user address space range. */ if (!is_user_vaddr ((void *) phdr->p_vaddr)) return false; if (!is_user_vaddr ((void *) (phdr->p_vaddr + phdr->p_memsz))) return false; /* The region cannot "wrap around" across the kernel virtual address space. */ if (phdr->p_vaddr + phdr->p_memsz < phdr->p_vaddr) return false; /* Disallow mapping page 0. Not only is it a bad idea to map page 0, but if we allowed it then user code that passed a null pointer to system calls could quite likely panic the kernel by way of null pointer assertions in memcpy(), etc. */ if (phdr->p_vaddr < PGSIZE) return false; /* It's okay. */ return true; } /* Loads a segment starting at offset OFS in FILE at address UPAGE. In total, READ_BYTES + ZERO_BYTES bytes of virtual memory are initialized, as follows: - READ_BYTES bytes at UPAGE must be read from FILE starting at offset OFS. - ZERO_BYTES bytes at UPAGE + READ_BYTES must be zeroed. The pages initialized by this function must be writable by the user process if WRITABLE is true, read-only otherwise. Return true if successful, false if a memory allocation error or disk read error occurs. */ static bool load_segment (struct file *file, off_t ofs, uint8_t *upage, uint32_t read_bytes, uint32_t zero_bytes, bool writable) { ASSERT ((read_bytes + zero_bytes) % PGSIZE == 0); ASSERT (pg_ofs (upage) == 0); ASSERT (ofs % PGSIZE == 0); while (read_bytes > 0 || zero_bytes > 0) { /* Calculate how to fill this page. We will read PAGE_READ_BYTES bytes from FILE and zero the final PAGE_ZERO_BYTES bytes. */ size_t page_read_bytes = read_bytes < PGSIZE ? read_bytes : PGSIZE; size_t page_zero_bytes = PGSIZE - page_read_bytes; /* Add the page metadata to the SPT to be lazy loaded later on */ if (page_insert_file (file, ofs, upage, page_read_bytes, page_zero_bytes, writable, PAGE_FILE) == NULL) return false; /* Advance. */ read_bytes -= page_read_bytes; zero_bytes -= page_zero_bytes; ofs += PGSIZE; upage += PGSIZE; } return true; } /* Create a minimal stack by mapping a zeroed page at the top of user virtual memory. */ static bool setup_stack (void **esp) { uint8_t *kpage; bool success = false; void *upage = ((uint8_t *) PHYS_BASE) - PGSIZE; kpage = get_usr_kpage (PAL_ZERO, upage); if (kpage != NULL) { success = install_page (upage, kpage, true); if (success) *esp = PHYS_BASE; else free_usr_kpage (kpage); } return success; } /* Claims a page from the user pool for ownership by the current thread and returns its kernel address, updating the frame table if VM is enabled. Requires the intended virtual address for where the page will be installed. */ static void * get_usr_kpage (enum palloc_flags flags, void *upage) { void *page; #ifdef VM struct thread *t = thread_current (); if (pagedir_get_page (t->pagedir, upage) != NULL) return NULL; else page = frame_alloc (flags, upage, t); pagedir_set_accessed (t->pagedir, upage, true); #else page = palloc_get_page (flags | PAL_USER); #endif return page; } /* Frees a page belonging to a user process given its kernel address, updating the frame table if VM is enabled. */ static void free_usr_kpage (void *kpage) { #ifdef VM frame_free (kpage); #else palloc_free_page (kpage); #endif } /* Adds a mapping from user virtual address UPAGE to kernel virtual address KPAGE to the page table. If WRITABLE is true, the user process may modify the page; otherwise, it is read-only. UPAGE must not already be mapped. KPAGE should probably be a page obtained from the user pool with palloc_get_page(). Returns true on success, false if UPAGE is already mapped or if memory allocation fails. */ bool install_page (void *upage, void *kpage, bool writable) { struct thread *t = thread_current (); /* Verify that there's not already a page at that virtual address, then map our page there. */ return (pagedir_get_page (t->pagedir, upage) == NULL && pagedir_set_page (t->pagedir, upage, kpage, writable)); }