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 struct mutex *__get_turnstile(struct mm_id *mm_id) 27 { 28 struct mm_context *ctx = container_of(mm_id, struct mm_context, id); 29 30 return &ctx->turnstile; 31 } 32 33 void enter_turnstile(struct mm_id *mm_id) 34 { 35 mutex_lock(__get_turnstile(mm_id)); 36 } 37 38 void exit_turnstile(struct mm_id *mm_id) 39 { 40 mutex_unlock(__get_turnstile(mm_id)); 41 } 42 43 int init_new_context(struct task_struct *task, struct mm_struct *mm) 44 { 45 struct mm_id *new_id = &mm->context.id; 46 unsigned long stack = 0; 47 int ret = -ENOMEM; 48 49 mutex_init(&mm->context.turnstile); 50 spin_lock_init(&mm->context.sync_tlb_lock); 51 52 stack = __get_free_pages(GFP_KERNEL | __GFP_ZERO, ilog2(STUB_DATA_PAGES)); 53 if (stack == 0) 54 goto out; 55 56 new_id->stack = stack; 57 new_id->syscall_data_len = 0; 58 new_id->syscall_fd_num = 0; 59 60 scoped_guard(spinlock_irqsave, &mm_list_lock) { 61 /* Insert into list, used for lookups when the child dies */ 62 list_add(&mm->context.list, &mm_list); 63 } 64 65 ret = start_userspace(new_id); 66 if (ret < 0) 67 goto out_free; 68 69 /* Ensure the new MM is clean and nothing unwanted is mapped */ 70 unmap(new_id, 0, STUB_START); 71 72 return 0; 73 74 out_free: 75 free_pages(new_id->stack, ilog2(STUB_DATA_PAGES)); 76 out: 77 return ret; 78 } 79 80 void destroy_context(struct mm_struct *mm) 81 { 82 struct mm_context *mmu = &mm->context; 83 84 /* 85 * If init_new_context wasn't called, this will be 86 * zero, resulting in a kill(0), which will result in the 87 * whole UML suddenly dying. Also, cover negative and 88 * 1 cases, since they shouldn't happen either. 89 * 90 * Negative cases happen if the child died unexpectedly. 91 */ 92 if (mmu->id.pid >= 0 && mmu->id.pid < 2) { 93 printk(KERN_ERR "corrupt mm_context - pid = %d\n", 94 mmu->id.pid); 95 return; 96 } 97 98 scoped_guard(spinlock_irqsave, &mm_list_lock) 99 list_del(&mm->context.list); 100 101 if (mmu->id.pid > 0) { 102 os_kill_ptraced_process(mmu->id.pid, 1); 103 mmu->id.pid = -1; 104 } 105 106 if (using_seccomp && mmu->id.sock) 107 os_close_file(mmu->id.sock); 108 109 free_pages(mmu->id.stack, ilog2(STUB_DATA_PAGES)); 110 } 111 112 static irqreturn_t mm_sigchld_irq(int irq, void* dev) 113 { 114 struct mm_context *mm_context; 115 pid_t pid; 116 117 guard(spinlock)(&mm_list_lock); 118 119 while ((pid = os_reap_child()) > 0) { 120 /* 121 * A child died, check if we have an MM with the PID. This is 122 * only relevant in SECCOMP mode (as ptrace will fail anyway). 123 * 124 * See wait_stub_done_seccomp for more details. 125 */ 126 list_for_each_entry(mm_context, &mm_list, list) { 127 if (mm_context->id.pid == pid) { 128 struct stub_data *stub_data; 129 printk("Unexpectedly lost MM child! Affected tasks will segfault."); 130 131 /* Marks the MM as dead */ 132 mm_context->id.pid = -1; 133 134 stub_data = (void *)mm_context->id.stack; 135 stub_data->futex = FUTEX_IN_KERN; 136 #if IS_ENABLED(CONFIG_SMP) 137 os_futex_wake(&stub_data->futex); 138 #endif 139 140 /* 141 * NOTE: Currently executing syscalls by 142 * affected tasks may finish normally. 143 */ 144 break; 145 } 146 } 147 } 148 149 return IRQ_HANDLED; 150 } 151 152 static int __init init_child_tracking(void) 153 { 154 int err; 155 156 spin_lock_init(&mm_list_lock); 157 INIT_LIST_HEAD(&mm_list); 158 159 err = request_irq(SIGCHLD_IRQ, mm_sigchld_irq, 0, "SIGCHLD", NULL); 160 if (err < 0) 161 panic("Failed to register SIGCHLD IRQ: %d", err); 162 163 return 0; 164 } 165 early_initcall(init_child_tracking) 166