exec: Linux's 2 MB of arguments; mmap: merge neighbouring mappings
exec took 32 arguments and 64 environment variables, in fixed arrays, so a glob over more than 31 files failed with E2BIG. It now reads the caller's argv and envp where they are, through the uaccess helpers and a bounce page -- a second thread rewriting the array after a check cannot slip a kernel address in -- and sizes the new stack to fit them. The limits are Linux's: ARG_MAX, 2 MB of strings and pointers, and 128 KB for any one string (USER_STR_MAX follows it). Every mmap took one of a process's 256 VMA slots, and zsh takes one per 16 KB heap block, so `x=($(<file))` over 30000 words ran out after 4 MB of heap and exited 1 without a word. Neighbouring anonymous mappings now share a record, as Linux merges them -- the bump allocator puts each new one where the last ended -- and munmap trims or splits it. Partial unmaps of device or segment mappings are refused with EINVAL instead of kfree()ing pages that are not the process's, and a failed allocation half way through a mapping no longer leaks what it had. Checked under QEMU: touch over 3000 names, /bin/echo with 50000 arguments, E2BIG past 2 MB, zsh splitting 100000 words, fork loops. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
This commit is contained in:
Nexoniarz
parent
eab13241a0
commit
51983144e3
7 changed files with 366 additions and 133 deletions
|
|
@ -1079,6 +1079,23 @@ memory. Three translations from the truth.
|
|||
|
||||
`NVMA` is 256 and the failure returns `-ENOMEM`.
|
||||
|
||||
256 ran out too, later, and silently: `x=($(<file))` over 30000 words exited
|
||||
zsh with status 1 and no message. zsh takes one 16 KB mapping per heap block,
|
||||
so 256 slots was 4 MB of heap. Neighbouring anonymous mappings are merged into
|
||||
one record now, as Linux merges them. Addresses come from a bump allocator, so
|
||||
a new mapping nearly always starts where the last one ended. `munmap` trims or
|
||||
splits the record. The slots now bound how many *separate* regions a process
|
||||
holds, not how many `mmap` calls it has made, and 100000 words split fine.
|
||||
|
||||
### exec took 32 arguments
|
||||
|
||||
`MAXARG` was 32 and `MAXENV` 64, in fixed arrays, so `rm *.o` in any build
|
||||
directory failed with `E2BIG`. exec now reads the caller's arrays where they
|
||||
are, through the uaccess helpers and a bounce page, so a second thread
|
||||
rewriting them cannot slip a kernel address in after a check. It sizes the new
|
||||
stack to fit them. The limits are Linux's: 2 MB of strings and pointers
|
||||
(`ARG_MAX`), and 128 KB per string.
|
||||
|
||||
### And underneath all of it: a small write costs 58 ms
|
||||
|
||||
With everything above fixed, `compinit` still appeared to hang. It was not
|
||||
|
|
|
|||
|
|
@ -194,6 +194,9 @@ int ibusy(uint dev, uint64 except);
|
|||
// Path lookup that follows a final symlink too; 0 with *err set on failure.
|
||||
// `start` 0 means the cwd. In fs.c.
|
||||
struct inode *nameifollow(struct inode *start, char *path, int *err);
|
||||
// execve(2) with the argv and envp arrays still in the caller's memory. In
|
||||
// exec.c; exec() is the same for arrays the kernel holds.
|
||||
int exec_user(char *path, addr_t uargv, addr_t uenvp);
|
||||
|
||||
// table mapping major device number to
|
||||
// device functions
|
||||
|
|
|
|||
|
|
@ -42,8 +42,11 @@
|
|||
#else
|
||||
#define ROOTDEV 1 // device number of file system root disk
|
||||
#endif
|
||||
#define MAXARG 32 // max exec arguments
|
||||
#define MAXENV 64 // max environment variables passed through exec
|
||||
// exec's limits, Linux's: ARG_MAX bytes of argument and environment strings
|
||||
// and their pointers together, and no one string over ARG_STRLEN_MAX (Linux's
|
||||
// MAX_ARG_STRLEN). There is no limit on the count beyond that.
|
||||
#define ARG_MAX (2 * 1024 * 1024)
|
||||
#define ARG_STRLEN_MAX (128 * 1024)
|
||||
#define MAXPATH 128 // max path
|
||||
#define MAXOPBLOCKS 30 // max # of blocks any FS op writes
|
||||
#define LOGSIZE (MAXOPBLOCKS*3) // max data blocks in on-disk log
|
||||
|
|
|
|||
|
|
@ -39,7 +39,7 @@ int copy_to_user(addr_t uaddr, const void *src, uint64 n);
|
|||
#define put_user(x, uaddr) copy_to_user((addr_t)(uaddr), &(x), sizeof(x))
|
||||
|
||||
// Longest string fetchstr()/strlen_user() will scan before giving up.
|
||||
#define USER_STR_MAX 32768
|
||||
#define USER_STR_MAX (128 * 1024) // = ARG_STRLEN_MAX: an argument is one
|
||||
|
||||
// Length of the NUL-terminated string at uaddr (excluding the NUL),
|
||||
// validating every page it spans. -EFAULT if it runs off mapped memory
|
||||
|
|
|
|||
280
kernel/fs/exec.c
280
kernel/fs/exec.c
|
|
@ -15,6 +15,7 @@
|
|||
#include "sleeplock.h"
|
||||
#include "fs.h"
|
||||
#include "file.h"
|
||||
#include "uaccess.h"
|
||||
|
||||
// Binaries with entry points at 0x100000+ are musl/Linux-ABI.
|
||||
// SchOS-native binaries (init) are linked at 0x4000.
|
||||
|
|
@ -22,39 +23,178 @@
|
|||
extern int linux_syscall_map[];
|
||||
extern int linux_syscall_map_size;
|
||||
|
||||
// Where the argument strings come from: an array in the kernel (the first
|
||||
// exec, of /bin/schinit), or one in the calling process's memory.
|
||||
//
|
||||
// The user case reads every pointer and every string through the uaccess
|
||||
// helpers, and copies through a kernel buffer, rather than being handed
|
||||
// pointers somebody checked earlier. Another thread of the same process can
|
||||
// rewrite the array between a check and a use, and a pointer swapped for a
|
||||
// kernel address afterwards would have copied kernel memory into the new
|
||||
// program's stack.
|
||||
struct argsrc {
|
||||
int user;
|
||||
char **k;
|
||||
addr_t u;
|
||||
};
|
||||
|
||||
// Pointer i of the array, 0 past its end; -1 on a fault.
|
||||
static int
|
||||
arg_ptr(struct argsrc *s, uint64 i, addr_t *p)
|
||||
{
|
||||
if (s->user)
|
||||
return s->u ? fetchaddr(s->u + i * sizeof(addr_t), p) : (*p = 0, 0);
|
||||
*p = s->k ? (addr_t)s->k[i] : 0;
|
||||
return 0;
|
||||
}
|
||||
|
||||
static long
|
||||
arg_len(struct argsrc *s, addr_t p)
|
||||
{
|
||||
return s->user ? strlen_user(p) : (long)strlen((char *)p);
|
||||
}
|
||||
|
||||
// Count an array and its bytes, against Linux's limits: ARG_MAX in all,
|
||||
// pointers included, and ARG_STRLEN_MAX for any one string.
|
||||
static int
|
||||
arg_measure(struct argsrc *s, uint64 *count, uint64 *bytes, uint64 already)
|
||||
{
|
||||
for (uint64 i = 0; ; i++) {
|
||||
addr_t p;
|
||||
if (arg_ptr(s, i, &p) < 0)
|
||||
return -EFAULT;
|
||||
if (p == 0) {
|
||||
*count = i;
|
||||
return 0;
|
||||
}
|
||||
long n = arg_len(s, p);
|
||||
if (n < 0)
|
||||
return n == -EFAULT ? -EFAULT : -E2BIG;
|
||||
if (n + 1 > ARG_STRLEN_MAX)
|
||||
return -E2BIG;
|
||||
*bytes += (uint64)n + 1;
|
||||
if (*bytes + (already + i + 1) * sizeof(addr_t) > ARG_MAX)
|
||||
return -E2BIG;
|
||||
}
|
||||
}
|
||||
|
||||
// The new stack's pointer table, written in order a page at a time.
|
||||
struct tbl {
|
||||
pml4e_t *pgdir;
|
||||
uint64 *buf;
|
||||
addr_t at; // where buf[0] goes in the new address space
|
||||
int n;
|
||||
};
|
||||
|
||||
static int
|
||||
tbl_flush(struct tbl *t)
|
||||
{
|
||||
if (t->n && copyout(t->pgdir, t->at, (char *)t->buf, t->n * 8) < 0)
|
||||
return -1;
|
||||
t->at += (addr_t)t->n * 8;
|
||||
t->n = 0;
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
tbl_put(struct tbl *t, uint64 v)
|
||||
{
|
||||
t->buf[t->n++] = v;
|
||||
return t->n == PGSIZE / 8 ? tbl_flush(t) : 0;
|
||||
}
|
||||
|
||||
// Copy `count` strings to `*at` onward in the new address space, each NUL
|
||||
// terminated whatever happened to it since it was measured, and put each
|
||||
// one's address in the table. The strings must not have grown: the room for
|
||||
// them was measured once, and running past it would write over the table.
|
||||
static int
|
||||
arg_copy(struct argsrc *s, uint64 count, pml4e_t *pgdir, addr_t *at,
|
||||
addr_t end, char *bounce, struct tbl *t)
|
||||
{
|
||||
for (uint64 i = 0; i < count; i++) {
|
||||
addr_t p;
|
||||
if (arg_ptr(s, i, &p) < 0 || p == 0)
|
||||
return -1;
|
||||
long n = arg_len(s, p);
|
||||
if (n < 0 || *at + (addr_t)n + 1 > end)
|
||||
return -1;
|
||||
if (tbl_put(t, *at) < 0)
|
||||
return -1;
|
||||
for (long done = 0; done < n; ) {
|
||||
long c = n - done > PGSIZE ? PGSIZE : n - done;
|
||||
if (s->user) {
|
||||
if (copy_from_user(bounce, p + done, c) < 0)
|
||||
return -1;
|
||||
} else {
|
||||
memmove(bounce, (char *)p + done, c);
|
||||
}
|
||||
if (copyout(pgdir, *at + done, bounce, c) < 0)
|
||||
return -1;
|
||||
done += c;
|
||||
}
|
||||
char nul = 0;
|
||||
if (copyout(pgdir, *at + n, &nul, 1) < 0)
|
||||
return -1;
|
||||
*at += (addr_t)n + 1;
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int do_exec(char *path, struct argsrc *av, struct argsrc *ev);
|
||||
|
||||
// envp may be null, which means "no environment" -- the same thing this
|
||||
// function used to do unconditionally.
|
||||
int
|
||||
exec(char *path, char **argv, char **envp)
|
||||
{
|
||||
struct argsrc a = { 0, argv, 0 }, e = { 0, envp, 0 };
|
||||
return do_exec(path, &a, &e);
|
||||
}
|
||||
|
||||
// execve(2): the arrays are in the caller's memory. See struct argsrc.
|
||||
int
|
||||
exec_user(char *path, addr_t uargv, addr_t uenvp)
|
||||
{
|
||||
struct argsrc a = { 1, 0, uargv }, e = { 1, 0, uenvp };
|
||||
return do_exec(path, &a, &e);
|
||||
}
|
||||
|
||||
static int
|
||||
do_exec(char *path, struct argsrc *av, struct argsrc *ev)
|
||||
{
|
||||
uint64 _t0 = rdtsc();
|
||||
fsstat.exec_calls++;
|
||||
char *s, *last;
|
||||
int i, off;
|
||||
uint64 argc, envc, sz, sp;
|
||||
uint64 sz, sp;
|
||||
struct elfhdr elf;
|
||||
struct inode *ip;
|
||||
struct proghdr ph;
|
||||
pml4e_t *pgdir, *oldpgdir;
|
||||
int slot;
|
||||
|
||||
// The three pointer arrays live on the heap, not on the kernel stack.
|
||||
// The arguments are measured before anything is loaded, so that too many
|
||||
// of them is E2BIG with nothing to undo -- and so the stack can be sized to
|
||||
// hold them.
|
||||
//
|
||||
// Together they are about 1.8 KB, and KSTACKSIZE is 4096 -- with the call
|
||||
// chain under this function running exec -> loaduvm -> readi -> SchFS ->
|
||||
// btree, there is not 1.8 KB to spare. Putting them here overflowed the
|
||||
// stack on the first exec after the environment was added, and the canary
|
||||
// in param.h caught it on pid 1 before it could corrupt anything.
|
||||
//
|
||||
// One page holds all three with room over.
|
||||
// There used to be a fixed 32 arguments and 64 variables, in arrays on the
|
||||
// kernel stack and then in one page. `rm *.o` in a build directory, or any
|
||||
// glob over more than 31 files, failed with E2BIG. Now the limit is Linux's:
|
||||
// 2 MB of strings and pointers together.
|
||||
uint64 argc = 0, envc = 0, strbytes = 0;
|
||||
int r;
|
||||
if ((r = arg_measure(av, &argc, &strbytes, 0)) < 0 ||
|
||||
(r = arg_measure(ev, &envc, &strbytes, argc + 1)) < 0)
|
||||
return r;
|
||||
|
||||
// Two pages, off the kernel stack, which has no room for either: a bounce
|
||||
// buffer for the strings and a batch of pointer-table entries.
|
||||
char *bounce = kalloc();
|
||||
uint64 *scratch = (uint64*)kalloc();
|
||||
if(!scratch)
|
||||
return -1;
|
||||
uint64 *ustack = scratch; /* MAXARG+1 */
|
||||
uint64 *estack = ustack + MAXARG + 1; /* MAXENV+1 */
|
||||
uint64 *stack_slots = estack + MAXENV + 1; /* the rest */
|
||||
_Static_assert((MAXARG + 1 + MAXENV + 1 + MAXARG + MAXENV + 32) * 8 <= 4096,
|
||||
"exec scratch does not fit in one page");
|
||||
if(!bounce || !scratch){
|
||||
if (bounce) kfree(bounce);
|
||||
if (scratch) kfree((char*)scratch);
|
||||
return -ENOMEM;
|
||||
}
|
||||
uint64 max_sz = 0;
|
||||
uint64 rand_va = 0;
|
||||
|
||||
|
|
@ -105,91 +245,65 @@ exec(char *path, char **argv, char **envp)
|
|||
end_op();
|
||||
ip = 0;
|
||||
|
||||
// The stack: 63 pages plus a guard, as always, and on top of that room for
|
||||
// the arguments, the pointer table and AT_RANDOM.
|
||||
uint64 nslots = 1 + argc + 1 + envc + 1 + 20;
|
||||
uint64 argpages = (strbytes + nslots * 8 + 64 + PGSIZE - 1) / PGSIZE;
|
||||
uint64 stackpages = 64 + argpages;
|
||||
sz = PGROUNDUP(max_sz);
|
||||
if((sz = allocuvm(pgdir, sz, sz + 64*PGSIZE)) == 0)
|
||||
if((sz = allocuvm(pgdir, sz, sz + stackpages*PGSIZE)) == 0)
|
||||
goto bad;
|
||||
clearpteu(pgdir, (char*)(sz - 64*PGSIZE));
|
||||
sp = sz;
|
||||
clearpteu(pgdir, (char*)(sz - stackpages*PGSIZE));
|
||||
|
||||
for(argc = 0; argv[argc]; argc++) {
|
||||
if(argc >= MAXARG)
|
||||
goto bad;
|
||||
sp -= strlen(argv[argc]) + 1;
|
||||
sp &= ~7;
|
||||
if(copyout(pgdir, sp, argv[argc], strlen(argv[argc]) + 1) < 0)
|
||||
goto bad;
|
||||
ustack[argc] = sp;
|
||||
}
|
||||
// From the top: the strings, then 16 random bytes, then the table that
|
||||
// points at them -- argc, argv, 0, envp, 0, the auxiliary vector -- with
|
||||
// the stack pointer at argc, 16-byte aligned, as the ELF ABI requires.
|
||||
addr_t strbase = (sz - strbytes) & ~0xFULL;
|
||||
rand_va = (strbase - 16) & ~0xFULL;
|
||||
sp = (rand_va - nslots * 8) & ~0xFULL;
|
||||
|
||||
// The environment, copied the same way and for the same reason: these
|
||||
// strings live in the *old* address space, which is still mapped, and will
|
||||
// not be once the new pgdir is installed.
|
||||
for(envc = 0; envp && envp[envc]; envc++) {
|
||||
if(envc >= MAXENV)
|
||||
goto bad;
|
||||
sp -= strlen(envp[envc]) + 1;
|
||||
sp &= ~7;
|
||||
if(copyout(pgdir, sp, envp[envc], strlen(envp[envc]) + 1) < 0)
|
||||
goto bad;
|
||||
estack[envc] = sp;
|
||||
}
|
||||
|
||||
// Push 16 pseudo-random bytes for AT_RANDOM.
|
||||
{
|
||||
uint8 randbuf[16];
|
||||
uint rseed = (uint)(uintptr_t)myproc() ^ (uint)myproc()->pid;
|
||||
uint rseed = (uint)(uintptr_t)myproc() ^ (uint)myproc()->pid ^ (uint)rdtsc();
|
||||
for(int ri = 0; ri < 16; ri++){
|
||||
rseed = rseed * 1664525u + 1013904223u;
|
||||
randbuf[ri] = (uint8)(rseed >> 16);
|
||||
}
|
||||
sp -= 16; sp &= ~0xFUL;
|
||||
rand_va = sp;
|
||||
if(copyout(pgdir, rand_va, (char*)randbuf, 16) < 0) goto bad;
|
||||
}
|
||||
|
||||
slot = 0;
|
||||
stack_slots[slot++] = argc;
|
||||
for(i = 0; i < argc; i++) {
|
||||
stack_slots[slot++] = ustack[i];
|
||||
}
|
||||
stack_slots[slot++] = 0; // end of argv
|
||||
|
||||
struct tbl t = { pgdir, scratch, sp, 0 };
|
||||
addr_t at = strbase;
|
||||
if (tbl_put(&t, argc) < 0)
|
||||
goto bad;
|
||||
if (arg_copy(av, argc, pgdir, &at, sz, bounce, &t) < 0 || tbl_put(&t, 0) < 0)
|
||||
goto bad;
|
||||
// The environment used to be a single zero here, with the comment "(no
|
||||
// environment)". Nothing on SchOS had one: not a shell, not a musl program
|
||||
// that had just called execve with a perfectly good envp, not anything. So
|
||||
// $TERM was unset, $PATH was whatever each program invented, and a variable
|
||||
// exported in a shell did not survive the exec that was supposed to carry it.
|
||||
for(i = 0; i < envc; i++)
|
||||
stack_slots[slot++] = estack[i];
|
||||
stack_slots[slot++] = 0; // end of envp
|
||||
if (arg_copy(ev, envc, pgdir, &at, sz, bounce, &t) < 0 || tbl_put(&t, 0) < 0)
|
||||
goto bad;
|
||||
|
||||
// Auxiliary vector — musl reads these for page size, clock, etc.
|
||||
stack_slots[slot++] = 6; // AT_PAGESZ
|
||||
stack_slots[slot++] = 4096;
|
||||
stack_slots[slot++] = 17; // AT_CLKTCK
|
||||
stack_slots[slot++] = USER_HZ; // what times() is scaled to, not a literal
|
||||
stack_slots[slot++] = 23; // AT_SECURE = 0 (not setuid)
|
||||
stack_slots[slot++] = 0;
|
||||
stack_slots[slot++] = 11; // AT_UID
|
||||
stack_slots[slot++] = 0;
|
||||
stack_slots[slot++] = 12; // AT_EUID
|
||||
stack_slots[slot++] = 0;
|
||||
stack_slots[slot++] = 13; // AT_GID
|
||||
stack_slots[slot++] = 0;
|
||||
stack_slots[slot++] = 14; // AT_EGID
|
||||
stack_slots[slot++] = 0;
|
||||
stack_slots[slot++] = 16; // AT_HWCAP
|
||||
stack_slots[slot++] = 0;
|
||||
stack_slots[slot++] = 25; // AT_RANDOM
|
||||
stack_slots[slot++] = rand_va;
|
||||
stack_slots[slot++] = 0; // AT_NULL
|
||||
stack_slots[slot++] = 0;
|
||||
|
||||
sp -= slot * sizeof(uint64);
|
||||
sp &= ~0xF;
|
||||
|
||||
if(copyout(pgdir, sp, (char*)stack_slots, slot * sizeof(uint64)) < 0)
|
||||
goto bad;
|
||||
{
|
||||
static const uint64 fixed[] = {
|
||||
6, 4096, // AT_PAGESZ
|
||||
17, USER_HZ, // AT_CLKTCK: what times() is scaled to, not a literal
|
||||
23, 0, // AT_SECURE: not setuid
|
||||
11, 0, 12, 0, // AT_UID, AT_EUID
|
||||
13, 0, 14, 0, // AT_GID, AT_EGID
|
||||
16, 0, // AT_HWCAP
|
||||
};
|
||||
for (int k = 0; k < (int)NELEM(fixed); k++)
|
||||
if (tbl_put(&t, fixed[k]) < 0)
|
||||
goto bad;
|
||||
if (tbl_put(&t, 25) < 0 || tbl_put(&t, rand_va) < 0 || // AT_RANDOM
|
||||
tbl_put(&t, 0) < 0 || tbl_put(&t, 0) < 0 || // AT_NULL
|
||||
tbl_flush(&t) < 0)
|
||||
goto bad;
|
||||
}
|
||||
|
||||
for(last=s=path; *s; s++)
|
||||
if(*s == '/')
|
||||
|
|
@ -277,11 +391,13 @@ exec(char *path, char **argv, char **envp)
|
|||
switchuvm(myproc());
|
||||
freevm(oldpgdir);
|
||||
kfree((char*)scratch);
|
||||
kfree(bounce);
|
||||
fsstat.exec_cycles += rdtsc() - _t0;
|
||||
return 0;
|
||||
|
||||
bad:
|
||||
kfree((char*)scratch);
|
||||
kfree(bounce);
|
||||
if(pgdir)
|
||||
freevm(pgdir);
|
||||
if(ip){
|
||||
|
|
|
|||
|
|
@ -987,34 +987,20 @@ addr_t sys_unlink(void)
|
|||
|
||||
addr_t sys_exec(void)
|
||||
{
|
||||
char *path, *argv[MAXARG], *envp[MAXENV];
|
||||
addr_t uargv, uenvp, uarg;
|
||||
char *path;
|
||||
addr_t uargv, uenvp;
|
||||
if (argstr(0, &path) < 0 || argaddr(1, &uargv) < 0) return -EFAULT;
|
||||
memset(argv, 0, sizeof(argv));
|
||||
for (int i = 0; ; i++) {
|
||||
if (i >= NELEM(argv)) return -E2BIG;
|
||||
if (fetchaddr(uargv + sizeof(addr_t) * i, &uarg) < 0) return -EFAULT;
|
||||
if (!uarg) { argv[i] = 0; break; }
|
||||
if (fetchstr(uarg, &argv[i]) < 0) return -EFAULT;
|
||||
}
|
||||
|
||||
// The third argument. This is Linux's execve(2), so musl has been passing
|
||||
// an environment here all along -- it was simply never read, and every
|
||||
// program on the system started with an empty one.
|
||||
// The arrays are read where they are, by exec_user(), rather than copied
|
||||
// into fixed arrays here: there used to be room for 32 arguments and 64
|
||||
// variables, and `rm *` in a directory of 40 files failed with E2BIG.
|
||||
//
|
||||
// A null envp is legal and means exactly that. It is also what the
|
||||
// first-party exec() wrapper passes, so old callers keep their old
|
||||
// behaviour without knowing about this.
|
||||
memset(envp, 0, sizeof(envp));
|
||||
if (argaddr(2, &uenvp) == 0 && uenvp) {
|
||||
for (int i = 0; ; i++) {
|
||||
if (i >= NELEM(envp)) return -E2BIG;
|
||||
if (fetchaddr(uenvp + sizeof(addr_t) * i, &uarg) < 0) return -EFAULT;
|
||||
if (!uarg) { envp[i] = 0; break; }
|
||||
if (fetchstr(uarg, &envp[i]) < 0) return -EFAULT;
|
||||
}
|
||||
}
|
||||
return exec(path, argv, envp);
|
||||
// envp is Linux's third argument, which musl had been passing all along
|
||||
// while nothing read it. A null envp is legal and means "none"; it is also
|
||||
// what the first-party exec() wrapper passes.
|
||||
if (argaddr(2, &uenvp) < 0)
|
||||
uenvp = 0;
|
||||
return exec_user(path, uargv, uenvp);
|
||||
}
|
||||
|
||||
// ftruncate/truncate both grow and shrink. Growing used to be quietly dropped
|
||||
|
|
|
|||
|
|
@ -31,6 +31,51 @@ bad_user_range(uint64 va, uint64 length)
|
|||
return length == 0 || va + length < va || va + length > KERNBASE;
|
||||
}
|
||||
|
||||
// Anonymous memory -- pages this process owns outright, with no segment or
|
||||
// device behind them. These are the only mappings that may be merged, trimmed
|
||||
// or split, because for them the page table is the whole truth: fork copies
|
||||
// and munmap frees whatever is present in a range and skips the holes.
|
||||
static int
|
||||
vma_anon(const struct vma *v)
|
||||
{
|
||||
return v->used && !v->no_free && v->shmid < 0;
|
||||
}
|
||||
|
||||
static struct vma *
|
||||
vma_free_slot(struct proc *p)
|
||||
{
|
||||
for (int i = 0; i < NVMA; i++)
|
||||
if (!p->vma[i].used)
|
||||
return &p->vma[i];
|
||||
return 0;
|
||||
}
|
||||
|
||||
// Record [va, va+len) as anonymous memory.
|
||||
//
|
||||
// Every mapping used to take a slot of its own, so a process could hold NVMA
|
||||
// of them and no more -- and zsh takes one per 16 KB heap block, so splitting
|
||||
// 30000 words into an array failed after 4 MB of heap with a silent exit 1.
|
||||
// Addresses come from a bump allocator, so a new mapping almost always starts
|
||||
// where the last one ended, and it extends that one instead. Linux merges
|
||||
// neighbouring anonymous mappings for the same reason.
|
||||
static int
|
||||
vma_note_anon(struct proc *p, uint64 va, uint64 len)
|
||||
{
|
||||
for (int i = 0; i < NVMA; i++) {
|
||||
struct vma *v = &p->vma[i];
|
||||
if (vma_anon(v) && v->addr + v->length == va) {
|
||||
v->length += len;
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
struct vma *v = vma_free_slot(p);
|
||||
if (!v)
|
||||
return -ENOMEM;
|
||||
v->addr = va; v->length = len; v->used = 1;
|
||||
v->no_free = 0; v->shmid = -1;
|
||||
return 0;
|
||||
}
|
||||
|
||||
addr_t
|
||||
sys_sbrk(void)
|
||||
{
|
||||
|
|
@ -90,15 +135,15 @@ sys_mmap(void)
|
|||
return va;
|
||||
}
|
||||
|
||||
struct vma *v = 0;
|
||||
for (int i = 0; i < NVMA; i++) {
|
||||
if (!myproc()->vma[i].used) { v = &myproc()->vma[i]; break; }
|
||||
}
|
||||
// A slot of its own, for the mappings that cannot be merged: a segment,
|
||||
// device memory, the framebuffer. Anonymous memory goes through
|
||||
// vma_note_anon() instead and usually needs none.
|
||||
//
|
||||
// -ENOMEM, not -1. A bare -1 reaches musl as -EPERM, so a process that
|
||||
// had simply used all NVMA slots was told "operation not permitted" --
|
||||
// which sends you looking for a permission model this system does not
|
||||
// have. malloc() turns either into NULL, but only one of them is a clue.
|
||||
if (!v) return -ENOMEM;
|
||||
struct vma *v = vma_free_slot(myproc());
|
||||
|
||||
// Ignore non-zero hint addresses (not MAP_FIXED); always pick a fresh VA.
|
||||
// Bump-allocated: each mapping gets its own disjoint, exactly-length-sized
|
||||
|
|
@ -139,6 +184,8 @@ sys_mmap(void)
|
|||
// it is refused with the errno that says the range is wrong.
|
||||
if (offset + length > have)
|
||||
return -EINVAL;
|
||||
if (!v)
|
||||
return -ENOMEM;
|
||||
|
||||
shm_ref(id);
|
||||
for (uint64 o = 0; o < length; o += PGSIZE) {
|
||||
|
|
@ -176,10 +223,14 @@ sys_mmap(void)
|
|||
|
||||
for (uint64 a = va; a < va + length; a += PGSIZE) {
|
||||
char *mem = kalloc();
|
||||
if (!mem) return -ENOMEM;
|
||||
if (!mem) {
|
||||
vm_unmap_range(myproc()->pgdir, va, a - va);
|
||||
return -ENOMEM;
|
||||
}
|
||||
memset(mem, 0, PGSIZE);
|
||||
if (mappages(myproc()->pgdir, (void*)a, PGSIZE, V2P((uint64)mem), PTE_W | PTE_U) < 0) {
|
||||
kfree(mem);
|
||||
vm_unmap_range(myproc()->pgdir, va, a - va);
|
||||
return -ENOMEM;
|
||||
}
|
||||
}
|
||||
|
|
@ -190,8 +241,11 @@ sys_mmap(void)
|
|||
readi(f->ip, 0, va, offset, length);
|
||||
iunlock(f->ip);
|
||||
|
||||
v->addr = va; v->length = length; v->used = 1;
|
||||
v->no_free = 0; v->shmid = -1;
|
||||
// A private copy, so from here it is anonymous memory like any other.
|
||||
if (vma_note_anon(myproc(), va, length) < 0) {
|
||||
vm_unmap_range(myproc()->pgdir, va, length);
|
||||
return -ENOMEM;
|
||||
}
|
||||
return va;
|
||||
}
|
||||
|
||||
|
|
@ -206,6 +260,8 @@ sys_mmap(void)
|
|||
uint64 len = PGROUNDUP(length);
|
||||
if (!pciuser_mmio_ok(phys, len))
|
||||
return -EACCES;
|
||||
if (!v)
|
||||
return -ENOMEM;
|
||||
// PTE_PWT: device registers are not memory, and a cached write to a
|
||||
// doorbell arrives whenever the cache feels like it, which for a
|
||||
// driver is indistinguishable from hardware that ignores you.
|
||||
|
|
@ -230,6 +286,8 @@ sys_mmap(void)
|
|||
return -ENODEV;
|
||||
if (length > fbsz)
|
||||
length = fbsz;
|
||||
if (!v)
|
||||
return -ENOMEM;
|
||||
if (mappages(myproc()->pgdir, (void*)va, length, fb, PTE_W | PTE_U) < 0)
|
||||
return -1;
|
||||
is_fb = 1;
|
||||
|
|
@ -240,11 +298,22 @@ sys_mmap(void)
|
|||
} else {
|
||||
for (uint64 a = va; a < va + length; a += PGSIZE) {
|
||||
char *mem = kalloc();
|
||||
if (!mem) return -1;
|
||||
if (!mem) {
|
||||
vm_unmap_range(myproc()->pgdir, va, a - va);
|
||||
return -ENOMEM;
|
||||
}
|
||||
memset(mem, 0, PGSIZE);
|
||||
if (mappages(myproc()->pgdir, (void*)a, PGSIZE, V2P((uint64)mem), PTE_W | PTE_U) < 0)
|
||||
return -1;
|
||||
if (mappages(myproc()->pgdir, (void*)a, PGSIZE, V2P((uint64)mem), PTE_W | PTE_U) < 0) {
|
||||
kfree(mem);
|
||||
vm_unmap_range(myproc()->pgdir, va, a - va);
|
||||
return -ENOMEM;
|
||||
}
|
||||
}
|
||||
if (vma_note_anon(myproc(), va, length) < 0) {
|
||||
vm_unmap_range(myproc()->pgdir, va, length);
|
||||
return -ENOMEM;
|
||||
}
|
||||
return va;
|
||||
}
|
||||
|
||||
v->addr = va; v->length = length; v->used = 1;
|
||||
|
|
@ -256,27 +325,66 @@ addr_t
|
|||
sys_munmap(void)
|
||||
{
|
||||
uint64 addr, length;
|
||||
struct proc *p = myproc();
|
||||
if (argaddr(0, &addr) < 0 || argaddr(1, &length) < 0)
|
||||
return -1;
|
||||
length = PGROUNDUP(length);
|
||||
if (bad_user_range(addr, length))
|
||||
if (bad_user_range(addr, length) || (addr % PGSIZE))
|
||||
return -EINVAL;
|
||||
uint64 end = addr + length;
|
||||
|
||||
// A segment or device mapping goes only as a whole, by its own address.
|
||||
// Part of one is refused: unmapping it the anonymous way below would kfree
|
||||
// pages that are the framebuffer's or another process's.
|
||||
for (int i = 0; i < NVMA; i++) {
|
||||
if (myproc()->vma[i].used && myproc()->vma[i].addr == addr) {
|
||||
if (myproc()->vma[i].shmid >= 0)
|
||||
shm_detach(myproc()->vma[i].shmid);
|
||||
if (myproc()->vma[i].no_free)
|
||||
vm_unmap_norelease(myproc()->pgdir, addr, myproc()->vma[i].length);
|
||||
struct vma *v = &p->vma[i];
|
||||
if (!v->used || vma_anon(v))
|
||||
continue;
|
||||
if (v->addr == addr) {
|
||||
if (v->shmid >= 0)
|
||||
shm_detach(v->shmid);
|
||||
if (v->no_free)
|
||||
vm_unmap_norelease(p->pgdir, addr, v->length);
|
||||
else
|
||||
vm_unmap_range(myproc()->pgdir, addr, myproc()->vma[i].length);
|
||||
myproc()->vma[i].used = myproc()->vma[i].addr = myproc()->vma[i].length = 0;
|
||||
myproc()->vma[i].no_free = 0;
|
||||
myproc()->vma[i].shmid = -1;
|
||||
vm_unmap_range(p->pgdir, addr, v->length);
|
||||
v->used = v->addr = v->length = 0;
|
||||
v->no_free = 0;
|
||||
v->shmid = -1;
|
||||
return 0;
|
||||
}
|
||||
if (addr < v->addr + v->length && end > v->addr)
|
||||
return -EINVAL;
|
||||
}
|
||||
// Partial / untracked unmap — free pages anyway
|
||||
vm_unmap_range(myproc()->pgdir, addr, length);
|
||||
|
||||
// Anonymous memory: trim every mapping the range touches, splitting one
|
||||
// it falls in the middle of. With no slot for the split, the record keeps
|
||||
// covering the hole, which is harmless -- fork and munmap skip pages that
|
||||
// are not there, and the bump allocator never hands the addresses out
|
||||
// again.
|
||||
for (int i = 0; i < NVMA; i++) {
|
||||
struct vma *v = &p->vma[i];
|
||||
if (!vma_anon(v))
|
||||
continue;
|
||||
uint64 vs = v->addr, ve = v->addr + v->length;
|
||||
if (end <= vs || addr >= ve)
|
||||
continue;
|
||||
if (addr <= vs && end >= ve) {
|
||||
v->used = v->addr = v->length = 0;
|
||||
} else if (addr <= vs) {
|
||||
v->addr = end;
|
||||
v->length = ve - end;
|
||||
} else if (end >= ve) {
|
||||
v->length = addr - vs;
|
||||
} else {
|
||||
struct vma *w = vma_free_slot(p);
|
||||
if (w) {
|
||||
w->addr = end; w->length = ve - end; w->used = 1;
|
||||
w->no_free = 0; w->shmid = -1;
|
||||
v->length = addr - vs;
|
||||
}
|
||||
}
|
||||
}
|
||||
vm_unmap_range(p->pgdir, addr, length);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Reference in a new issue