xref: /minix/minix/servers/vm/utility.c (revision 433d6423)
1 
2 /* This file contains some utility routines for VM.  */
3 
4 #define _SYSTEM		1
5 
6 #define brk _brk	/* get rid of no previous prototype warning */
7 
8 #include <minix/callnr.h>
9 #include <minix/com.h>
10 #include <minix/config.h>
11 #include <minix/const.h>
12 #include <minix/ds.h>
13 #include <minix/endpoint.h>
14 #include <minix/minlib.h>
15 #include <minix/type.h>
16 #include <minix/ipc.h>
17 #include <minix/sysutil.h>
18 #include <minix/syslib.h>
19 #include <minix/type.h>
20 #include <minix/bitmap.h>
21 #include <string.h>
22 #include <errno.h>
23 #include <env.h>
24 #include <unistd.h>
25 #include <assert.h>
26 #include <sys/param.h>
27 #include <sys/mman.h>
28 #include <sys/resource.h>
29 
30 #include "proto.h"
31 #include "glo.h"
32 #include "util.h"
33 #include "region.h"
34 #include "sanitycheck.h"
35 
36 #include <machine/archtypes.h>
37 #include "kernel/const.h"
38 #include "kernel/config.h"
39 #include "kernel/type.h"
40 #include "kernel/proc.h"
41 
42 /*===========================================================================*
43  *                              get_mem_chunks                               *
44  *===========================================================================*/
45 void get_mem_chunks(
46 struct memory *mem_chunks)                      /* store mem chunks here */
47 {
48 /* Initialize the free memory list from the kernel-provided memory map.  Translate
49  * the byte offsets and sizes in this list to clicks, properly truncated.
50  */
51   phys_bytes base, size, limit;
52   int i;
53   struct memory *memp;
54 
55   /* Initialize everything to zero. */
56   memset(mem_chunks, 0, NR_MEMS*sizeof(*mem_chunks));
57 
58   /* Obtain and parse memory from kernel environment. */
59   /* XXX Any memory chunk in excess of NR_MEMS is silently ignored. */
60   for(i = 0; i < MIN(MAXMEMMAP, NR_MEMS); i++) {
61   	mem_chunks[i].base = kernel_boot_info.memmap[i].mm_base_addr;
62   	mem_chunks[i].size = kernel_boot_info.memmap[i].mm_length;
63   }
64 
65   /* Round physical memory to clicks. Round start up, round end down. */
66   for (i = 0; i < NR_MEMS; i++) {
67         memp = &mem_chunks[i];          /* next mem chunk is stored here */
68         base = mem_chunks[i].base;
69         size = mem_chunks[i].size;
70         limit = base + size;
71         base = (phys_bytes) (CLICK_CEIL(base));
72         limit = (phys_bytes) (CLICK_FLOOR(limit));
73         if (limit <= base) {
74                 memp->base = memp->size = 0;
75         } else {
76                 memp->base = base >> CLICK_SHIFT;
77                 memp->size = (limit - base) >> CLICK_SHIFT;
78         }
79   }
80 }
81 
82 /*===========================================================================*
83  *                              vm_isokendpt                           	     *
84  *===========================================================================*/
85 int vm_isokendpt(endpoint_t endpoint, int *procn)
86 {
87         *procn = _ENDPOINT_P(endpoint);
88         if(*procn < 0 || *procn >= NR_PROCS)
89 		return EINVAL;
90         if(*procn >= 0 && endpoint != vmproc[*procn].vm_endpoint)
91                 return EDEADEPT;
92         if(*procn >= 0 && !(vmproc[*procn].vm_flags & VMF_INUSE))
93                 return EDEADEPT;
94         return OK;
95 }
96 
97 
98 /*===========================================================================*
99  *                              do_info                                      *
100  *===========================================================================*/
101 int do_info(message *m)
102 {
103 	struct vm_stats_info vsi;
104 	struct vm_usage_info vui;
105 	static struct vm_region_info vri[MAX_VRI_COUNT];
106 	struct vmproc *vmp;
107 	vir_bytes addr, size, next, ptr;
108 	int r, pr, dummy, count, free_pages, largest_contig;
109 
110 	if (vm_isokendpt(m->m_source, &pr) != OK)
111 		return EINVAL;
112 	vmp = &vmproc[pr];
113 
114 	ptr = (vir_bytes) m->m_lsys_vm_info.ptr;
115 
116 	switch(m->m_lsys_vm_info.what) {
117 	case VMIW_STATS:
118 		vsi.vsi_pagesize = VM_PAGE_SIZE;
119 		vsi.vsi_total = total_pages;
120 		memstats(&dummy, &free_pages, &largest_contig);
121 		vsi.vsi_free = free_pages;
122 		vsi.vsi_largest = largest_contig;
123 
124 		get_stats_info(&vsi);
125 
126 		addr = (vir_bytes) &vsi;
127 		size = sizeof(vsi);
128 
129 		break;
130 
131 	case VMIW_USAGE:
132 		if(m->m_lsys_vm_info.ep < 0)
133 			get_usage_info_kernel(&vui);
134 		else if (vm_isokendpt(m->m_lsys_vm_info.ep, &pr) != OK)
135 			return EINVAL;
136 		else get_usage_info(&vmproc[pr], &vui);
137 
138 		addr = (vir_bytes) &vui;
139 		size = sizeof(vui);
140 
141 		break;
142 
143 	case VMIW_REGION:
144 		if (vm_isokendpt(m->m_lsys_vm_info.ep, &pr) != OK)
145 			return EINVAL;
146 
147 		count = MIN(m->m_lsys_vm_info.count, MAX_VRI_COUNT);
148 		next = m->m_lsys_vm_info.next;
149 
150 		count = get_region_info(&vmproc[pr], vri, count, &next);
151 
152 		m->m_lsys_vm_info.count = count;
153 		m->m_lsys_vm_info.next = next;
154 
155 		addr = (vir_bytes) vri;
156 		size = sizeof(vri[0]) * count;
157 
158 		break;
159 
160 	default:
161 		return EINVAL;
162 	}
163 
164 	if (size == 0)
165 		return OK;
166 
167 	/* Make sure that no page faults can occur while copying out. A page
168 	 * fault would cause the kernel to send a notify to us, while we would
169 	 * be waiting for the result of the copy system call, resulting in a
170 	 * deadlock. Note that no memory mapping can be undone without the
171 	 * involvement of VM, so we are safe until we're done.
172 	 */
173 	r = handle_memory_once(vmp, ptr, size, 1 /*wrflag*/);
174 	if (r != OK) return r;
175 
176 	/* Now that we know the copy out will succeed, perform the actual copy
177 	 * operation.
178 	 */
179 	return sys_datacopy(SELF, addr,
180 		(vir_bytes) vmp->vm_endpoint, ptr, size);
181 }
182 
183 /*===========================================================================*
184  *				swap_proc_slot	     			     *
185  *===========================================================================*/
186 int swap_proc_slot(struct vmproc *src_vmp, struct vmproc *dst_vmp)
187 {
188 	struct vmproc orig_src_vmproc, orig_dst_vmproc;
189 
190 #if LU_DEBUG
191 	printf("VM: swap_proc: swapping %d (%d) and %d (%d)\n",
192 	    src_vmp->vm_endpoint, src_vmp->vm_slot,
193 	    dst_vmp->vm_endpoint, dst_vmp->vm_slot);
194 #endif
195 
196 	/* Save existing data. */
197 	orig_src_vmproc = *src_vmp;
198 	orig_dst_vmproc = *dst_vmp;
199 
200 	/* Swap slots. */
201 	*src_vmp = orig_dst_vmproc;
202 	*dst_vmp = orig_src_vmproc;
203 
204 	/* Preserve endpoints and slot numbers. */
205 	src_vmp->vm_endpoint = orig_src_vmproc.vm_endpoint;
206 	src_vmp->vm_slot = orig_src_vmproc.vm_slot;
207 	dst_vmp->vm_endpoint = orig_dst_vmproc.vm_endpoint;
208 	dst_vmp->vm_slot = orig_dst_vmproc.vm_slot;
209 
210 #if LU_DEBUG
211 	printf("VM: swap_proc: swapped %d (%d) and %d (%d)\n",
212 	    src_vmp->vm_endpoint, src_vmp->vm_slot,
213 	    dst_vmp->vm_endpoint, dst_vmp->vm_slot);
214 #endif
215 
216 	return OK;
217 }
218 
219 /*===========================================================================*
220  *			      swap_proc_dyn_data	     		     *
221  *===========================================================================*/
222 int swap_proc_dyn_data(struct vmproc *src_vmp, struct vmproc *dst_vmp)
223 {
224 	int is_vm;
225 	int r;
226 
227 	is_vm = (dst_vmp->vm_endpoint == VM_PROC_NR);
228 
229         /* For VM, transfer memory regions above the stack first. */
230         if(is_vm) {
231 #if LU_DEBUG
232 		printf("VM: swap_proc_dyn_data: tranferring regions above the stack from old VM (%d) to new VM (%d)\n",
233 			src_vmp->vm_endpoint, dst_vmp->vm_endpoint);
234 #endif
235 		r = pt_map_in_range(src_vmp, dst_vmp, VM_STACKTOP, 0);
236 		if(r != OK) {
237 			printf("swap_proc_dyn_data: pt_map_in_range failed\n");
238 			return r;
239 		}
240         }
241 
242 #if LU_DEBUG
243 	printf("VM: swap_proc_dyn_data: swapping regions' parents for %d (%d) and %d (%d)\n",
244 	    src_vmp->vm_endpoint, src_vmp->vm_slot,
245 	    dst_vmp->vm_endpoint, dst_vmp->vm_slot);
246 #endif
247 
248 	/* Swap vir_regions' parents. */
249 	map_setparent(src_vmp);
250 	map_setparent(dst_vmp);
251 
252 	/* For regular processes, transfer regions above the stack now.
253 	 * In case of rollback, we need to skip this step. To sandbox the
254 	 * new instance and prevent state corruption on rollback, we share all
255 	 * the regions between the two instances as COW.
256 	 */
257 	if(!is_vm) {
258 		struct vir_region *vr;
259 		vr = map_lookup(dst_vmp, VM_STACKTOP, NULL);
260 		if(vr && !map_lookup(src_vmp, VM_STACKTOP, NULL)) {
261 #if LU_DEBUG
262 			printf("VM: swap_proc_dyn_data: tranferring regions above the stack from %d to %d\n",
263 				src_vmp->vm_endpoint, dst_vmp->vm_endpoint);
264 #endif
265 			r = map_proc_copy_from(src_vmp, dst_vmp, vr);
266 			if(r != OK) {
267 				return r;
268 			}
269 		}
270 	}
271 
272 	return OK;
273 }
274 
275 void *mmap(void *addr, size_t len, int f, int f2, int f3, off_t o)
276 {
277 	void *ret;
278 	phys_bytes p;
279 
280 	assert(!addr);
281 	assert(!(len % VM_PAGE_SIZE));
282 
283 	ret = vm_allocpages(&p, VMP_SLAB, len/VM_PAGE_SIZE);
284 
285 	if(!ret) return MAP_FAILED;
286 	memset(ret, 0, len);
287 	return ret;
288 }
289 
290 int munmap(void * addr, size_t len)
291 {
292 	vm_freepages((vir_bytes) addr, roundup(len, VM_PAGE_SIZE)/VM_PAGE_SIZE);
293 	return 0;
294 }
295 
296 int brk(void *addr)
297 {
298 	/* brk is a special case function to allow vm itself to
299 	   allocate memory in it's own (cacheable) HEAP */
300 	vir_bytes target = roundup((vir_bytes)addr, VM_PAGE_SIZE), v;
301 	extern char _end;
302 	extern char *_brksize;
303 	static vir_bytes prevbrk = (vir_bytes) &_end;
304 	struct vmproc *vmprocess = &vmproc[VM_PROC_NR];
305 
306 	for(v = roundup(prevbrk, VM_PAGE_SIZE); v < target;
307 		v += VM_PAGE_SIZE) {
308 		phys_bytes mem, newpage = alloc_mem(1, 0);
309 		if(newpage == NO_MEM) return -1;
310 		mem = CLICK2ABS(newpage);
311 		if(pt_writemap(vmprocess, &vmprocess->vm_pt,
312 			v, mem, VM_PAGE_SIZE,
313 			  ARCH_VM_PTE_PRESENT
314 			| ARCH_VM_PTE_USER
315 			| ARCH_VM_PTE_RW
316 #if defined(__arm__)
317 			| ARM_VM_PTE_CACHED
318 #endif
319 			, 0) != OK) {
320 			free_mem(newpage, 1);
321 			return -1;
322 		}
323 		prevbrk = v + VM_PAGE_SIZE;
324 	}
325 
326         _brksize = (char *) addr;
327 
328         if(sys_vmctl(SELF, VMCTL_FLUSHTLB, 0) != OK)
329         	panic("flushtlb failed");
330 
331 	return 0;
332 }
333 
334 /*===========================================================================*
335  *				do_getrusage		     		     *
336  *===========================================================================*/
337 int do_getrusage(message *m)
338 {
339 	int res, slot;
340 	struct vmproc *vmp;
341 	struct rusage r_usage;
342 	if ((res = vm_isokendpt(m->m_source, &slot)) != OK)
343 		return ESRCH;
344 
345 	vmp = &vmproc[slot];
346 
347 	if ((res = sys_datacopy(m->m_source, m->m_lc_vm_rusage.addr,
348 		SELF, (vir_bytes) &r_usage, (vir_bytes) sizeof(r_usage))) < 0)
349 		return res;
350 
351 	r_usage.ru_maxrss = vmp->vm_total_max;
352 	r_usage.ru_minflt = vmp->vm_minor_page_fault;
353 	r_usage.ru_majflt = vmp->vm_major_page_fault;
354 
355 	return sys_datacopy(SELF, (vir_bytes) &r_usage, m->m_source,
356 		m->m_lc_vm_rusage.addr, (vir_bytes) sizeof(r_usage));
357 }
358