static void * minix_mmap_emulator(void *addrhint, size_t size, int prot, int flags, int fd, off_t off)
{
void *ret;
int mapflags;
size_t s;
mapflags = flags & (MAP_FIXED);
#ifdef MINIXVERBOSE
if(addrhint) {
fprintf(stderr, "0x%lx-0x%lx requested\n", addrhint,
(char *) addrhint + size);
}
#endif
if((ret = minix_mmap(addrhint, size, PROT_READ|PROT_WRITE,
MAP_ANON|MAP_PRIVATE|MAP_PREALLOC|mapflags, -1, 0)) == MAP_FAILED) {
return ret;
}
if(!(mapflags & MAP_ANON)) {
if((s=pread(fd, ret, size, off)) <= 0) {
munmap(ret, size);
return MAP_FAILED;
}
}
#ifdef MINIXVERBOSE
fprintf(stderr, "0x%lx-0x%lx mapped\n",
ret, (char *) ret + size);
#endif
return ret;
}
static void
lmfs_alloc_block(struct buf *bp)
{
ASSERT(!bp->data);
ASSERT(bp->lmfs_bytes == 0);
if((bp->data = minix_mmap(0, fs_block_size,
PROT_READ|PROT_WRITE, MAP_PREALLOC|MAP_ANON, -1, 0)) == MAP_FAILED) {
free_unused_blocks();
if((bp->data = minix_mmap(0, fs_block_size, PROT_READ|PROT_WRITE,
MAP_PREALLOC|MAP_ANON, -1, 0)) == MAP_FAILED) {
panic("libminixfs: could not allocate block");
}
}
assert(bp->data);
bp->lmfs_bytes = fs_block_size;
bp->lmfs_needsetcache = 1;
}
void *alloc_contig(size_t len, int flags, phys_bytes *phys)
{
vir_bytes buf;
int mmapflags = MAP_PREALLOC|MAP_CONTIG|MAP_ANON;
if(flags & AC_LOWER16M)
mmapflags |= MAP_LOWER16M;
if(flags & AC_LOWER1M)
mmapflags |= MAP_LOWER1M;
if(flags & AC_ALIGN64K)
mmapflags |= MAP_ALIGN64K;
/* First try to get memory with minix_mmap. This is guaranteed
* to be page-aligned, and we can tell VM it has to be
* pre-allocated and contiguous.
*/
errno = 0;
buf = (vir_bytes) minix_mmap(0, len, PROT_READ|PROT_WRITE, mmapflags, -1, 0);
/* If that failed, maybe we're not running in paged mode.
* If that's the case, ENXIO will be returned.
* Memory returned with malloc() will be preallocated and
* contiguous, so fallback on that, and ask for a little extra
* so we can page align it ourselves.
*/
if(buf == (vir_bytes) MAP_FAILED) {
u32_t align = 0;
if(errno != (_SIGN ENXIO)) {
return NULL;
}
if(flags & AC_ALIGN4K)
align = 4*1024;
if(flags & AC_ALIGN64K)
align = 64*1024;
if(len + align < len)
return NULL;
len += align;
if(!(buf = (vir_bytes) malloc(len))) {
return NULL;
}
if(align)
buf += align - (buf % align);
}
/* Get physical address, if requested. */
if(phys != NULL && sys_umap_data_fb(SELF, buf, len, phys) != OK)
panic("sys_umap_data_fb failed");
return (void *) buf;
}
void basic_regression(void)
{
void *block;
#define BLOCKSIZE (PAGE_SIZE*10)
block = minix_mmap(0, BLOCKSIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANON, -1, 0);
if(block == MAP_FAILED) { e(1); exit(1); }
memset(block, 0, BLOCKSIZE);
/* shrink from bottom */
minix_munmap(block, PAGE_SIZE);
}
int
readblock(int b, int blocksize, u32_t seed, char *data)
{
u64_t offset;
int fd;
char *mmapdata;
int pread_first = random() % 2;
get_fd_offset(b, blocksize, &offset, &fd);
if(pread_first) {
if(pread(fd, data, blocksize, offset) < blocksize) {
perror("pread");
return -1;
}
}
if((mmapdata = minix_mmap(NULL, blocksize, PROT_READ, MAP_PRIVATE | MAP_FILE,
fd, offset)) == MAP_FAILED) {
perror("mmap");
return -1;
}
if(!pread_first) {
if(pread(fd, data, blocksize, offset) < blocksize) {
perror("pread");
return -1;
}
}
if(memcmp(mmapdata, data, blocksize)) {
fprintf(stderr, "readblock: mmap, pread mismatch\n");
return -1;
}
if(minix_munmap(mmapdata, blocksize) < 0) {
perror("munmap");
return -1;
}
return blocksize;
}
int
main(int argc, char *argv[])
{
#define CHUNKSIZE 8192
#define CHUNKS1 3
#define CHUNKS2 2
#define CHUNKS (CHUNKS1+CHUNKS2)
#define LARGESIZE 262144
int i, fd;
char *v[CHUNKS];
#define STARTV 0x90000000
char *vaddr = (char *) STARTV;
ssize_t l;
pid_t f;
printf("Test 44 ");
fflush(stdout);
for(i = 0; i < CHUNKS; i++) {
v[i] = minix_mmap(vaddr, CHUNKSIZE, PROT_READ|PROT_WRITE, 0,
-1, 0);
if(v[i] == MAP_FAILED) {
perror("minix_mmap");
fprintf(stderr, "minix_mmap failed\n");
exit(1);
}
if(v[i] != vaddr) {
fprintf(stderr,
"minix_mmap said 0x%p but i wanted 0x%p\n",
v[i], vaddr);
exit(1);
}
vaddr += CHUNKSIZE;
}
#define DEV_ZERO "/dev/zero"
if((fd=open(DEV_ZERO, O_RDONLY)) < 0) {
perror("open");
fprintf(stderr, "open failed for %s\n", DEV_ZERO);
exit(1);
}
#define TOTAL1 (CHUNKS1*CHUNKSIZE)
/* Make single read cross region boundary. */
if((l=read(fd, v[0], TOTAL1)) != TOTAL1) {
fprintf(stderr, "read %d but expected %d\n", l, TOTAL1);
exit(1);
}
/* Force single copy to cross region boundary. */
{
char *t;
t = v[CHUNKS1]+CHUNKSIZE-2;
if((l=read(fd, t, CHUNKSIZE)) != CHUNKSIZE) {
fprintf(stderr, "read %d but expected %d\n", l, CHUNKSIZE);
exit(1);
}
}
/* Now start a child to test bogus memory access */
if((f = fork()) == -1) {
perror("fork");
exit(1);
}
if(f > 0) {
int st;
/* Parent waits. */
if(waitpid(f, &st, 0) < 0) {
perror("waitpid");
exit(1);
}
if(!WIFEXITED(st)) {
fprintf(stderr, "child not signaled\n");
exit(1);
}
if(WEXITSTATUS(st) != 0) {
fprintf(stderr, "child exited with nonzero status\n");
exit(1);
}
} else {
/* Child performs bogus read */
int res;
char *buf = v[CHUNKS-1];
errno = 0;
res = read(fd, buf, LARGESIZE);
if(res >= 0) {
fprintf(stderr, "res %d\n", res);
exit(1);
}
if(errno != EFAULT) {
fprintf(stderr, "errno %d\n", errno);
exit(1);
}
exit(0);
}
printf("ok\n");
exit(0);
}
/* minix-without-mmap version of _rtld_map_object() */
Obj_Entry *
_rtld_map_object_fallback(const char *path, int fd, const struct stat *sb)
{
Obj_Entry *obj;
Elf_Ehdr *ehdr;
Elf_Phdr *phdr;
#if defined(__HAVE_TLS_VARIANT_I) || defined(__HAVE_TLS_VARIANT_II)
Elf_Phdr *phtls;
#endif
size_t phsize;
Elf_Phdr *phlimit;
Elf_Phdr *segs[2];
int nsegs;
caddr_t mapbase = MAP_FAILED;
size_t mapsize = 0;
int mapflags;
Elf_Off base_offset;
#ifdef MAP_ALIGNED
Elf_Addr base_alignment;
#endif
Elf_Addr base_vaddr;
Elf_Addr base_vlimit;
Elf_Addr text_vlimit;
int text_flags;
caddr_t base_addr;
Elf_Off data_offset;
Elf_Addr data_vaddr;
Elf_Addr data_vlimit;
int data_flags;
caddr_t data_addr;
#if defined(__HAVE_TLS_VARIANT_I) || defined(__HAVE_TLS_VARIANT_II)
Elf_Addr tls_vaddr = 0; /* Noise GCC */
#endif
Elf_Addr phdr_vaddr;
size_t phdr_memsz;
#if defined(__HAVE_TLS_VARIANT_I) || defined(__HAVE_TLS_VARIANT_II)
caddr_t gap_addr;
size_t gap_size;
#endif
int i;
#ifdef RTLD_LOADER
Elf_Addr clear_vaddr;
caddr_t clear_addr;
size_t nclear;
#endif
if (sb != NULL && sb->st_size < (off_t)sizeof (Elf_Ehdr)) {
_rtld_error("%s: not ELF file (too short)", path);
return NULL;
}
obj = _rtld_obj_new();
obj->path = xstrdup(path);
obj->pathlen = strlen(path);
if (sb != NULL) {
obj->dev = sb->st_dev;
obj->ino = sb->st_ino;
}
#ifdef __minix
ehdr = minix_mmap(NULL, _rtld_pagesz, PROT_READ|PROT_WRITE,
MAP_PREALLOC|MAP_ANON, -1, (off_t)0);
Pread(ehdr, _rtld_pagesz, fd, 0);
#if MINIXVERBOSE
fprintf(stderr, "minix mmap for header: 0x%lx\n", ehdr);
#endif
#else
ehdr = mmap(NULL, _rtld_pagesz, PROT_READ, MAP_FILE | MAP_SHARED, fd,
(off_t)0);
#endif
obj->ehdr = ehdr;
if (ehdr == MAP_FAILED) {
_rtld_error("%s: read error: %s", path, xstrerror(errno));
goto bad;
}
/* Make sure the file is valid */
if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0) {
_rtld_error("%s: not ELF file (magic number bad)", path);
goto bad;
}
if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
_rtld_error("%s: invalid ELF class %x; expected %x", path,
ehdr->e_ident[EI_CLASS], ELFCLASS);
goto bad;
}
/* Elf_e_ident includes class */
if (ehdr->e_ident[EI_VERSION] != EV_CURRENT ||
ehdr->e_version != EV_CURRENT ||
ehdr->e_ident[EI_DATA] != ELFDEFNNAME(MACHDEP_ENDIANNESS)) {
_rtld_error("%s: unsupported file version", path);
goto bad;
}
if (ehdr->e_type != ET_EXEC && ehdr->e_type != ET_DYN) {
_rtld_error("%s: unsupported file type", path);
goto bad;
}
switch (ehdr->e_machine) {
ELFDEFNNAME(MACHDEP_ID_CASES)
default:
_rtld_error("%s: unsupported machine", path);
goto bad;
}
/*
* We rely on the program header being in the first page. This is
* not strictly required by the ABI specification, but it seems to
* always true in practice. And, it simplifies things considerably.
*/
assert(ehdr->e_phentsize == sizeof(Elf_Phdr));
assert(ehdr->e_phoff + ehdr->e_phnum * sizeof(Elf_Phdr) <=
_rtld_pagesz);
/*
* Scan the program header entries, and save key information.
*
* We rely on there being exactly two load segments, text and data,
* in that order.
*/
phdr = (Elf_Phdr *) ((caddr_t)ehdr + ehdr->e_phoff);
#if defined(__HAVE_TLS_VARIANT_I) || defined(__HAVE_TLS_VARIANT_II)
phtls = NULL;
#endif
phsize = ehdr->e_phnum * sizeof(phdr[0]);
obj->phdr = NULL;
phdr_vaddr = EA_UNDEF;
phdr_memsz = 0;
phlimit = phdr + ehdr->e_phnum;
nsegs = 0;
while (phdr < phlimit) {
switch (phdr->p_type) {
case PT_INTERP:
obj->interp = (void *)(uintptr_t)phdr->p_vaddr;
dbg(("%s: PT_INTERP %p", obj->path, obj->interp));
break;
case PT_LOAD:
if (nsegs < 2)
segs[nsegs] = phdr;
++nsegs;
#if ELFSIZE == 64
#define PRImemsz PRIu64
#else
#define PRImemsz PRIu32
#endif
dbg(("%s: %s %p phsize %" PRImemsz, obj->path, "PT_LOAD",
(void *)(uintptr_t)phdr->p_vaddr, phdr->p_memsz));
break;
case PT_PHDR:
phdr_vaddr = phdr->p_vaddr;
phdr_memsz = phdr->p_memsz;
dbg(("%s: %s %p phsize %" PRImemsz, obj->path, "PT_PHDR",
(void *)(uintptr_t)phdr->p_vaddr, phdr->p_memsz));
break;
case PT_DYNAMIC:
obj->dynamic = (void *)(uintptr_t)phdr->p_vaddr;
dbg(("%s: %s %p phsize %" PRImemsz, obj->path, "PT_DYNAMIC",
(void *)(uintptr_t)phdr->p_vaddr, phdr->p_memsz));
break;
#if defined(__HAVE_TLS_VARIANT_I) || defined(__HAVE_TLS_VARIANT_II)
case PT_TLS:
phtls = phdr;
dbg(("%s: %s %p phsize %" PRImemsz, obj->path, "PT_TLS",
(void *)(uintptr_t)phdr->p_vaddr, phdr->p_memsz));
break;
#endif
}
++phdr;
}
phdr = (Elf_Phdr *) ((caddr_t)ehdr + ehdr->e_phoff);
obj->entry = (void *)(uintptr_t)ehdr->e_entry;
if (!obj->dynamic) {
_rtld_error("%s: not dynamically linked", path);
goto bad;
}
if (nsegs != 2) {
_rtld_error("%s: wrong number of segments (%d != 2)", path,
nsegs);
goto bad;
}
/*
* Map the entire address space of the object as a file
* region to stake out our contiguous region and establish a
* base for relocation. We use a file mapping so that
* the kernel will give us whatever alignment is appropriate
* for the platform we're running on.
*
* We map it using the text protection, map the data segment
* into the right place, then map an anon segment for the bss
* and unmap the gaps left by padding to alignment.
*/
#ifdef MAP_ALIGNED
base_alignment = segs[0]->p_align;
#endif
base_offset = round_down(segs[0]->p_offset);
base_vaddr = round_down(segs[0]->p_vaddr);
base_vlimit = round_up(segs[1]->p_vaddr + segs[1]->p_memsz);
text_vlimit = round_up(segs[0]->p_vaddr + segs[0]->p_memsz);
text_flags = protflags(segs[0]->p_flags);
data_offset = round_down(segs[1]->p_offset);
data_vaddr = round_down(segs[1]->p_vaddr);
data_vlimit = round_up(segs[1]->p_vaddr + segs[1]->p_filesz);
data_flags = protflags(segs[1]->p_flags);
#ifdef RTLD_LOADER
clear_vaddr = segs[1]->p_vaddr + segs[1]->p_filesz;
#endif
obj->textsize = text_vlimit - base_vaddr;
obj->vaddrbase = base_vaddr;
obj->isdynamic = ehdr->e_type == ET_DYN;
#if defined(__HAVE_TLS_VARIANT_I) || defined(__HAVE_TLS_VARIANT_II)
if (phtls != NULL) {
++_rtld_tls_dtv_generation;
obj->tlsindex = ++_rtld_tls_max_index;
obj->tlssize = phtls->p_memsz;
obj->tlsalign = phtls->p_align;
obj->tlsinitsize = phtls->p_filesz;
tls_vaddr = phtls->p_vaddr;
}
#endif
obj->phdr_loaded = false;
for (i = 0; i < nsegs; i++) {
if (phdr_vaddr != EA_UNDEF &&
segs[i]->p_vaddr <= phdr_vaddr &&
segs[i]->p_memsz >= phdr_memsz) {
obj->phdr_loaded = true;
break;
}
if (segs[i]->p_offset <= ehdr->e_phoff &&
segs[i]->p_memsz >= phsize) {
phdr_vaddr = segs[i]->p_vaddr + ehdr->e_phoff;
phdr_memsz = phsize;
obj->phdr_loaded = true;
break;
}
}
if (obj->phdr_loaded) {
obj->phdr = (void *)(uintptr_t)phdr_vaddr;
obj->phsize = phdr_memsz;
} else {
Elf_Phdr *buf;
buf = xmalloc(phsize);
if (buf == NULL) {
_rtld_error("%s: cannot allocate program header", path);
goto bad;
}
memcpy(buf, phdr, phsize);
obj->phdr = buf;
obj->phsize = phsize;
}
dbg(("%s: phdr %p phsize %zu (%s)", obj->path, obj->phdr, obj->phsize,
obj->phdr_loaded ? "loaded" : "allocated"));
/* Unmap header if it overlaps the first load section. */
if (base_offset < _rtld_pagesz) {
munmap(ehdr, _rtld_pagesz);
obj->ehdr = MAP_FAILED;
}
/*
* Calculate log2 of the base section alignment.
*/
mapflags = 0;
#ifdef MAP_ALIGNED
if (base_alignment > _rtld_pagesz) {
unsigned int log2 = 0;
for (; base_alignment > 1; base_alignment >>= 1)
log2++;
mapflags = MAP_ALIGNED(log2);
}
mapflags = MAP_ALIGNED(log2);
}
#endif
#ifdef RTLD_LOADER
base_addr = obj->isdynamic ? NULL : (caddr_t)base_vaddr;
#else
base_addr = NULL;
#endif
mapsize = base_vlimit - base_vaddr;
#ifndef __minix
mapbase = mmap(base_addr, mapsize, text_flags,
mapflags | MAP_FILE | MAP_PRIVATE, fd, base_offset);
#else
mapbase = minix_mmap(base_addr, mapsize, PROT_READ|PROT_WRITE,
MAP_ANON | MAP_PREALLOC, -1, 0);
#if MINIXVERBOSE
fprintf(stderr, "minix mmap for whole block: 0x%lx-0x%lx\n", mapbase, mapbase+mapsize);
#endif
Pread(mapbase, obj->textsize, fd, 0);
#endif
if (mapbase == MAP_FAILED) {
_rtld_error("mmap of entire address space failed: %s",
xstrerror(errno));
goto bad;
}
/* Overlay the data segment onto the proper region. */
data_addr = mapbase + (data_vaddr - base_vaddr);
#ifdef __minix
Pread(data_addr, data_vlimit - data_vaddr, fd, data_offset);
/*===========================================================================*
* m_block_ioctl *
*===========================================================================*/
static int m_block_ioctl(dev_t minor, unsigned int request, endpoint_t endpt,
cp_grant_id_t grant)
{
/* I/O controls for the block devices of the memory driver. Currently there is
* one I/O control specific to the memory driver:
* - MIOCRAMSIZE: to set the size of the RAM disk.
*/
struct device *dv;
u32_t ramdev_size;
int s;
void *mem;
if (request != MIOCRAMSIZE)
return EINVAL;
/* Someone wants to create a new RAM disk with the given size.
* A ramdisk can be created only once, and only on RAM disk device.
*/
if ((dv = m_block_part(minor)) == NULL) return ENXIO;
if((minor < RAM_DEV_FIRST || minor > RAM_DEV_LAST) && minor != RAM_DEV_OLD) {
printf("MEM: MIOCRAMSIZE: %d not a ramdisk\n", minor);
return EINVAL;
}
/* Get request structure */
s= sys_safecopyfrom(endpt, grant, 0, (vir_bytes)&ramdev_size,
sizeof(ramdev_size));
if (s != OK)
return s;
if(m_vaddrs[minor] && !cmp64(dv->dv_size, cvul64(ramdev_size))) {
return(OK);
}
/* openct is 1 for the ioctl(). */
if(openct[minor] != 1) {
printf("MEM: MIOCRAMSIZE: %d in use (count %d)\n",
minor, openct[minor]);
return(EBUSY);
}
if(m_vaddrs[minor]) {
u32_t size;
if(ex64hi(dv->dv_size)) {
panic("huge old ramdisk");
}
size = ex64lo(dv->dv_size);
minix_munmap((void *) m_vaddrs[minor], size);
m_vaddrs[minor] = (vir_bytes) NULL;
}
#if DEBUG
printf("MEM:%d: allocating ramdisk of size 0x%x\n", minor, ramdev_size);
#endif
/* Try to allocate a piece of memory for the RAM disk. */
if((mem = minix_mmap(NULL, ramdev_size, PROT_READ|PROT_WRITE,
MAP_PREALLOC|MAP_ANON, -1, 0)) == MAP_FAILED) {
printf("MEM: failed to get memory for ramdisk\n");
return(ENOMEM);
}
m_vaddrs[minor] = (vir_bytes) mem;
dv->dv_size = cvul64(ramdev_size);
return(OK);
}
/*===========================================================================*
* do_shmget *
*===========================================================================*/
int do_shmget(message *m)
{
struct shm_struct *shm;
long key, size, old_size;
int flag;
int id;
key = m->SHMGET_KEY;
old_size = size = m->SHMGET_SIZE;
flag = m->SHMGET_FLAG;
if ((shm = shm_find_key(key))) {
if (!check_perm(&shm->shmid_ds.shm_perm, who_e, flag))
return EACCES;
if ((flag & IPC_CREAT) && (flag & IPC_EXCL))
return EEXIST;
if (size && shm->shmid_ds.shm_segsz < size)
return EINVAL;
id = shm->id;
} else { /* no key found */
if (!(flag & IPC_CREAT))
return ENOENT;
if (size <= 0)
return EINVAL;
/* round up to a multiple of PAGE_SIZE */
if (size % I386_PAGE_SIZE)
size += I386_PAGE_SIZE - size % I386_PAGE_SIZE;
if (size <= 0)
return EINVAL;
if (shm_list_nr == MAX_SHM_NR)
return ENOMEM;
/* TODO: shmmni should be changed... */
if (identifier == SHMMNI)
return ENOSPC;
shm = &shm_list[shm_list_nr];
memset(shm, 0, sizeof(struct shm_struct));
shm->page = (vir_bytes) minix_mmap(0, size,
PROT_READ|PROT_WRITE,
MAP_CONTIG|MAP_PREALLOC|MAP_ANON|MAP_IPC_SHARED,
-1, 0);
if (shm->page == (vir_bytes) MAP_FAILED)
return ENOMEM;
shm->phys = vm_getphys(SELF_E, (void *) shm->page);
memset((void *)shm->page, 0, size);
shm->shmid_ds.shm_perm.cuid =
shm->shmid_ds.shm_perm.uid = getnuid(who_e);
shm->shmid_ds.shm_perm.cgid =
shm->shmid_ds.shm_perm.gid = getngid(who_e);
shm->shmid_ds.shm_perm.mode = flag & 0777;
shm->shmid_ds.shm_segsz = old_size;
shm->shmid_ds.shm_atime = 0;
shm->shmid_ds.shm_dtime = 0;
shm->shmid_ds.shm_ctime = time(NULL);
shm->shmid_ds.shm_cpid = getnpid(who_e);
shm->shmid_ds.shm_lpid = 0;
shm->shmid_ds.shm_nattch = 0;
shm->id = id = identifier++;
shm->key = key;
shm_list_nr++;
}
m->SHMGET_RETID = id;
return OK;
}
