| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de) |
| 4 | * Copyright (C) 2002 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com) |
| 5 | */ |
| 6 | |
| 7 | #include <linux/mm.h> |
| 8 | #include <linux/sched/signal.h> |
| 9 | #include <linux/slab.h> |
| 10 | |
| 11 | #include <shared/irq_kern.h> |
| 12 | #include <asm/pgalloc.h> |
| 13 | #include <asm/sections.h> |
| 14 | #include <asm/mmu_context.h> |
| 15 | #include <as-layout.h> |
| 16 | #include <os.h> |
| 17 | #include <skas.h> |
| 18 | #include <stub-data.h> |
| 19 | |
| 20 | /* Ensure the stub_data struct covers the allocated area */ |
| 21 | static_assert(sizeof(struct stub_data) == STUB_DATA_PAGES * UM_KERN_PAGE_SIZE); |
| 22 | |
| 23 | static spinlock_t mm_list_lock; |
| 24 | static struct list_head mm_list; |
| 25 | |
| 26 | void enter_turnstile(struct mm_id *mm_id) __acquires(turnstile) |
| 27 | { |
| 28 | struct mm_context *ctx = container_of(mm_id, struct mm_context, id); |
| 29 | |
| 30 | mutex_lock(&ctx->turnstile); |
| 31 | } |
| 32 | |
| 33 | void exit_turnstile(struct mm_id *mm_id) __releases(turnstile) |
| 34 | { |
| 35 | struct mm_context *ctx = container_of(mm_id, struct mm_context, id); |
| 36 | |
| 37 | mutex_unlock(lock: &ctx->turnstile); |
| 38 | } |
| 39 | |
| 40 | int init_new_context(struct task_struct *task, struct mm_struct *mm) |
| 41 | { |
| 42 | struct mm_id *new_id = &mm->context.id; |
| 43 | unsigned long stack = 0; |
| 44 | int ret = -ENOMEM; |
| 45 | |
| 46 | mutex_init(&mm->context.turnstile); |
| 47 | spin_lock_init(&mm->context.sync_tlb_lock); |
| 48 | |
| 49 | stack = __get_free_pages(GFP_KERNEL | __GFP_ZERO, ilog2(STUB_DATA_PAGES)); |
| 50 | if (stack == 0) |
| 51 | goto out; |
| 52 | |
| 53 | new_id->stack = stack; |
| 54 | new_id->syscall_data_len = 0; |
| 55 | new_id->syscall_fd_num = 0; |
| 56 | |
| 57 | scoped_guard(spinlock_irqsave, &mm_list_lock) { |
| 58 | /* Insert into list, used for lookups when the child dies */ |
| 59 | list_add(new: &mm->context.list, head: &mm_list); |
| 60 | } |
| 61 | |
| 62 | ret = start_userspace(new_id); |
| 63 | if (ret < 0) |
| 64 | goto out_free; |
| 65 | |
| 66 | /* Ensure the new MM is clean and nothing unwanted is mapped */ |
| 67 | unmap(new_id, 0, STUB_START); |
| 68 | |
| 69 | return 0; |
| 70 | |
| 71 | out_free: |
| 72 | free_pages(new_id->stack, ilog2(STUB_DATA_PAGES)); |
| 73 | out: |
| 74 | return ret; |
| 75 | } |
| 76 | |
| 77 | void destroy_context(struct mm_struct *mm) |
| 78 | { |
| 79 | struct mm_context *mmu = &mm->context; |
| 80 | |
| 81 | /* |
| 82 | * If init_new_context wasn't called, this will be |
| 83 | * zero, resulting in a kill(0), which will result in the |
| 84 | * whole UML suddenly dying. Also, cover negative and |
| 85 | * 1 cases, since they shouldn't happen either. |
| 86 | * |
| 87 | * Negative cases happen if the child died unexpectedly. |
| 88 | */ |
| 89 | if (mmu->id.pid >= 0 && mmu->id.pid < 2) { |
| 90 | printk(KERN_ERR "corrupt mm_context - pid = %d\n" , |
| 91 | mmu->id.pid); |
| 92 | return; |
| 93 | } |
| 94 | |
| 95 | scoped_guard(spinlock_irqsave, &mm_list_lock) |
| 96 | list_del(entry: &mm->context.list); |
| 97 | |
| 98 | if (mmu->id.pid > 0) { |
| 99 | os_kill_ptraced_process(mmu->id.pid, 1); |
| 100 | mmu->id.pid = -1; |
| 101 | } |
| 102 | |
| 103 | if (using_seccomp && mmu->id.sock) |
| 104 | os_close_file(mmu->id.sock); |
| 105 | |
| 106 | free_pages(mmu->id.stack, ilog2(STUB_DATA_PAGES)); |
| 107 | } |
| 108 | |
| 109 | static irqreturn_t mm_sigchld_irq(int irq, void* dev) |
| 110 | { |
| 111 | struct mm_context *mm_context; |
| 112 | pid_t pid; |
| 113 | |
| 114 | guard(spinlock)(l: &mm_list_lock); |
| 115 | |
| 116 | while ((pid = os_reap_child()) > 0) { |
| 117 | /* |
| 118 | * A child died, check if we have an MM with the PID. This is |
| 119 | * only relevant in SECCOMP mode (as ptrace will fail anyway). |
| 120 | * |
| 121 | * See wait_stub_done_seccomp for more details. |
| 122 | */ |
| 123 | list_for_each_entry(mm_context, &mm_list, list) { |
| 124 | if (mm_context->id.pid == pid) { |
| 125 | struct stub_data *stub_data; |
| 126 | printk("Unexpectedly lost MM child! Affected tasks will segfault." ); |
| 127 | |
| 128 | /* Marks the MM as dead */ |
| 129 | mm_context->id.pid = -1; |
| 130 | |
| 131 | stub_data = (void *)mm_context->id.stack; |
| 132 | stub_data->futex = FUTEX_IN_KERN; |
| 133 | #if IS_ENABLED(CONFIG_SMP) |
| 134 | os_futex_wake(&stub_data->futex); |
| 135 | #endif |
| 136 | |
| 137 | /* |
| 138 | * NOTE: Currently executing syscalls by |
| 139 | * affected tasks may finish normally. |
| 140 | */ |
| 141 | break; |
| 142 | } |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | return IRQ_HANDLED; |
| 147 | } |
| 148 | |
| 149 | static int __init init_child_tracking(void) |
| 150 | { |
| 151 | int err; |
| 152 | |
| 153 | spin_lock_init(&mm_list_lock); |
| 154 | INIT_LIST_HEAD(list: &mm_list); |
| 155 | |
| 156 | err = request_irq(SIGCHLD_IRQ, mm_sigchld_irq, 0, "SIGCHLD" , NULL); |
| 157 | if (err < 0) |
| 158 | panic(fmt: "Failed to register SIGCHLD IRQ: %d" , err); |
| 159 | |
| 160 | return 0; |
| 161 | } |
| 162 | early_initcall(init_child_tracking) |
| 163 | |