Linux内核情景分析之异常访问,用户堆栈的扩展

==================== arch/i386/mm/fault.c 106 152 ====================
106  asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long error_code)
107  {
108 struct task_struct *tsk;
109 struct mm_struct *mm;
110 struct vm_area_struct * vma;
111 unsigned long address;
112 unsigned long page;
113 unsigned long fixup;
114 int write;
115 siginfo_t info;
116
117 /* get the address */
118 __asm__("movl %%cr2,%0":"=r" (address));//得到失败的线性地址
119
120 tsk = current;//获取当前描述符
121
122 /*
123  * We fault-in kernel-space virtual memory on-demand. The
124  * ‘reference‘ page table is init_mm.pgd.
125  *
126  * NOTE! We MUST NOT take any locks for this case. We may
127  * be in an interrupt or a critical region, and should
128  * only copy the information from the master page table,
129  * nothing more.
130  */
 //如果大于3G,表示缺页异常时,访问的是内核空间,很有可能是访问了非连续的内核空间,转到vmalloc_fault处理
131 if (address >= TASK_SIZE)
132 goto vmalloc_fault;
133
134 mm = tsk->mm;
135 info.si_code = SEGV_MAPERR;
136
137 /*
138  缺页异常发生在中断时,是错误不可以的,表示是内核线程,不可以对其页表进行修改
140  */
141 if (in_interrupt() || !mm)
142 goto no_context;
143
144 down(&mm->mmap_sem);
145
146 vma = find_vma(mm, address);//查到end大于address的地址
147 if (!vma)//是否在行现地址内,不在转为错误处理
148 goto bad_area;
//在线性区内，跳到正常处理部分,可能是由于权限问题,也有可能是对应的物理地址没有分配2种情况
149 if (vma->vm_start <= address)
150 goto good_area;
151 if (!(vma->vm_flags & VM_GROWSDOWN))//如果发生在一空洞上方的区间不是堆栈区,那么此地址是由于撤销映射留下的,进行错误处理
152 goto bad_area;

220  /*
221   * Something tried to access memory that isn‘t in our memory map..
222   * Fix it, but check if it‘s kernel or user first..
223   */
224  bad_area:
225 up(&mm->mmap_sem);
226 //用户态的错误处理
227  bad_area_nosemaphore:
228 /* User mode accesses just cause a SIGSEGV */
229 if (error_code & 4) {//判断错误发生在用户态
230 tsk->thread.cr2 = address;
231 tsk->thread.error_code = error_code;
232 tsk->thread.trap_no = 14;
233 info.si_signo = SIGSEGV;//强制发送SIGEGV信号

if (!(vma->vm_flags & VM_GROWSDOWN))
152 goto bad_area;
153 if (error_code & 4) {
154 /*
155  * 还要检查异常地址是否紧挨着esp指针,如果远超过32,那就是非法越界,错误处理
32是因为pusha(一次把32个字节压入栈中)
159  */
160 if (address + 32 < regs->esp)
161 goto bad_area;
162 }
//扩展堆栈
163 if (expand_stack(vma, address))
164 goto bad_area;

[do_page_fault()]
165  /*
166   * Ok, we have a good vm_area for this memory access, so
167   * we can handle it..
168   */
169  good_area:
170 info.si_code = SEGV_ACCERR;
171 write = 0;
172 switch (error_code & 3) {
173 default:  /* 3: write, present */
174  #ifdef TEST_VERIFY_AREA
175 if (regs->cs == KERNEL_CS)
176 printk("WP fault at %08lx\n", regs->eip);
177  #endif
178 /* fall through */
179 case 2: /* write, not present */
180 if (!(vma->vm_flags & VM_WRITE))//堆栈段可读可写,调到196行
181 goto bad_area;
182 write++;
183 break;
184 case 1: /* read, present */

185 goto bad_area;
186 case 0: /* read, not present */
187 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
188 goto bad_area;
189 }
190
191 /*
192  * If for any reason at all we couldn‘t handle the fault,
193  * make sure we exit gracefully rather than endlessly redo
194  * the fault.
195  */
196 switch (handle_mm_fault(mm, vma, address, write)) {
197 case 1:
198 tsk->min_flt++;
199 break;
200 case 2:
201 tsk->maj_flt++;
202 break;
203 case 0:
204 goto do_sigbus;
205 default:
206 goto out_of_memory;
207 }

[do_page_fault()>handle_mm_fault()]
1189 /*
1190  * By the time we get here, we already hold the mm semaphore
1191  */
1192 int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct * vma,
1193 unsigned long address, int write_access)
1194 {
1195 int ret = -1;
1196 pgd_t *pgd;
1197 pmd_t *pmd;
1198
1199 pgd = pgd_offset(mm, address);//获取该地址所在页面目录项的指针(页表的地址)
1200 pmd = pmd_alloc(pgd, address);//分配pmd目录
1201
1202 if (pmd) {
1203 pte_t * pte = pte_alloc(pmd, address);//分配pte表现
1204 if (pte)
//分配物理地址
1205 ret = handle_pte_fault(mm, vma, address, write_access, pte);
1206 }
1207 return ret;
1208 }

[do_page_fault()>handle_mm_fault()>pte_alloc()]
120  extern inline pte_t * pte_alloc(pmd_t * pmd, unsigned long address)
121  {
122 address = (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);//给定地址转换为页表的下标,用于定位页表项
124 if (pmd_none(*pmd))//如果pmd所指向的页表为空,那就转到getnew分配
125 goto getnew;
126 if (pmd_bad(*pmd))
127 goto fix;
128 return (pte_t *)pmd_page(*pmd) + address;
129  getnew:
130  {
131 unsigned long page = (unsigned long) get_pte_fast();//从缓冲池获取(释放页表,并非一定会释放物理地址)
132
133 if (!page)
134 return get_pte_slow(pmd, address);
135 set_pmd(pmd, __pmd(_PAGE_TABLE + __pa(page)));//写入中间pmd
136 return (pte_t *)page + address;
137  }
138  fix:
139 __handle_bad_pmd(pmd);
140 return NULL;
141  }

[do_page_fault()>handle_mm_fault()>handle_pte_fault()]
1135 /*
1136  * These routines also need to handle stuff like marking pages dirty
1137  * and/or accessed for architectures that don‘t do it in hardware (most
1138  * RISC architectures).  The early dirtying is also good on the i386.
1139  *
1140  * There is also a hook called "update_mmu_cache()" that architectures
1141  * with external mmu caches can use to update those (ie the Sparc or
1142  * PowerPC hashed page tables that act as extended TLBs).
1143  *
1144  * Note the "page_table_lock". It is to protect against kswapd removing
1147  * we can drop the lock early.
1148  *
1149  * The adding of pages is protected by the MM semaphore (which we hold),
1150  * so we don‘t need to worry about a page being suddenly been added into
1151  * our VM.
1152  */
1153 static inline int handle_pte_fault(struct mm_struct *mm,
1154 struct vm_area_struct * vma, unsigned long address,
1155 int write_access, pte_t * pte)
1156 {
1157 pte_t entry;
1158
1159 /*
1160  * We need the page table lock to synchronize with kswapd
1161  * and the SMP-safe atomic PTE updates.
1162  */
1163 spin_lock(&mm->page_table_lock);
1164 entry = *pte;//pte对应的物理地址当然没有,所以为null
1165 if (!pte_present(entry)) {//检查其对应的物理地址否为空
1166 /*
1167  * If it truly wasn‘t present, we know that kswapd
1168  * and the PTE updates will not touch it later. So
1169  * drop the lock.
1170  */
1171 spin_unlock(&mm->page_table_lock);
1172 if (pte_none(entry))//页表项内容为0，表明进程未访问过该页 
1173 return do_no_page(mm, vma, address, write_access, pte);//调用do_no_page分配
//否则换出
1174 return do_swap_page(mm, vma, address, pte, pte_to_swp_entry(entry), write_access);
1175 }
1176
1177 if (write_access) {
1178 if (!pte_write(entry))
1179 return do_wp_page(mm, vma, address, pte, entry);
1180
1181 entry = pte_mkdirty(entry);
1182 }
1183 entry = pte_mkyoung(entry);
1184 establish_pte(vma, address, pte, entry);
1185 spin_unlock(&mm->page_table_lock);
1186 return 1;
1187 }
1145  * pages from under us. Note that kswapd only ever _removes_ pages, never
1146  * adds them. As such, once we have noticed that the page is not present

[do_page_fault()>handle_mm_fault()>handle_pte_fault()>do_no_page()]
1080 /*
1081  * do_no_page() tries to create a new page mapping. It aggressively
1082  * tries to share with existing pages, but makes a separate copy if
1083  * the "write_access" parameter is true in order to avoid the next
1084  * page fault.
1085  *
1086  * As this is called only for pages that do not currently exist, we
1087  * do not need to flush old virtual caches or the TLB.
1088  *
1089  * This is called with the MM semaphore held.
1090  */
1091 static int do_no_page(struct mm_struct * mm, struct vm_area_struct * vma,
1092 unsigned long address, int write_access, pte_t *page_table)
1093 {
1094 struct page * new_page;
1095 pte_t entry;
1096
1097 if (!vma->vm_ops || !vma->vm_ops->nopage)
1098 return do_anonymous_page(mm, vma, page_table, write_access, address);//只是其封装而已
    ......
==================== mm/memory.c 1133 1133 ====================
1133 }

[do_page_fault()>handle_mm_fault()>handle_pte_fault()>do_no_page()>do_anonymous_page()]
1058 /*
1059  * This only needs the MM semaphore
1060  */
1061 static int do_anonymous_page(struct mm_struct * mm, struct vm_area_struct * vma, pte_t *page_table,
int write_access, unsigned long addr)
1062 {
1063 struct page *page = NULL;
//如果引起页面异常是一次读操作,那么由mk_pte构建的映射表项要通过pte_wrprotect修正,只读属性
//同时对于只读的页面,一律映射ZERO_PAGE同一个物理内存页面,也即是内容全部为0,只有可写才独立分配内存
1064 pte_t entry = pte_wrprotect(mk_pte(ZERO_PAGE(addr), vma->vm_page_prot));
1065 if (write_access) {
1066 page = alloc_page(GFP_HIGHUSER);//分配独立物理页面
1067 if (!page)
1068 return -1;
1069 clear_user_highpage(page, addr);
//下面相同.只可写属性
1070 entry = pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
1071 mm->rss++;
1072 flush_page_to_ram(page);
1073 }
//虚拟页面到物理内存页面映射建立
1074 set_pte(page_table, entry);
1075 /* No need to invalidate - it was non-present before */
1076 update_mmu_cache(vma, addr, entry);
1077 return 1; /* Minor fault */
1078 }