SchOS/kernel/fs/procfs.c

// /proc as a real filesystem.
//
// It used to be a handful of character devices: init ran `mknod("/proc/uptime",
// 6, 2)` and friends, so every file was a real inode on the root filesystem
// with a major/minor pair behind it. That works for a fixed set of files and
// cannot work at all for the thing /proc is mostly *for* -- one directory per
// process. You cannot mknod a directory for a process that does not exist yet
// and will not exist in a minute, which is why `ps` and `top` have never
// worked: they call dirtree_flagread("/proc") and get a list with no pids in
// it.
//
// So this is a synthetic filesystem behind the same `struct fs_ops` vtable
// SchFS uses. Nothing is stored: every directory listing is generated from the
// process table at the moment it is asked for, and every file's contents are
// generated when it is read.
//
// The global files (meminfo, uptime, pci, ...) keep their existing generators
// in kernel/dev/procfs.c; this layer only decides what exists and hands the
// reads over. That split is deliberate -- the generators are about content and
// this file is about namespace, and they change for different reasons.

#include "core/types.h"
#include "core/defs.h"
#include "core/klog.h"
#include "core/param.h"
#include "arch/memlayout.h"
#include "arch/mmu.h"
#include "core/proc.h"
#include "core/spinlock.h"
#include "core/sleeplock.h"
#include "fs/fs.h"
#include "fs/file.h"
#include "core/kmem.h"
#include "uapi/stat.h"
#include "core/uaccess.h"
#include "uapi/errno.h"

// ---------------------------------------------------------------------------
// Inode numbering
//
// An inode number has to survive being handed out, stored in a struct inode,
// and given back later with no other context -- so it encodes the whole
// identity of the object.
//
//   1                       the root, /proc
//   PROCFS_GLOBAL | id      a global file, id being the PROC_* minor that
//                           kernel/dev/procfs.c already uses
//   pid << 8                the directory /proc/<pid>
//   (pid << 8) | fileid     a file inside it
//
// pid is bounded by NPROC and file ids by PF_MAX, so the ranges cannot
// collide. Shifting by 8 rather than packing tightly keeps the numbers
// readable in a debugger, which matters more than density here.
// ---------------------------------------------------------------------------

#define PROCFS_ROOT     1ULL
#define PROCFS_GLOBAL   0x10000ULL

#define PF_DIR      0        // the directory itself
#define PF_STAT     1
#define PF_STATUS   2
#define PF_CMDLINE  3
#define PF_STATM    4
#define PF_EXE      5
#define PF_SCHEDSTAT 6
#define PF_MAX      6

#define INO_PID(ino)    ((int)((ino) >> 8))
#define INO_FILE(ino)   ((int)((ino) & 0xFF))
#define INO_IS_GLOBAL(ino) (((ino) & PROCFS_GLOBAL) != 0)

// The global files, by the name they appear under and the PROC_* id that
// kernel/dev/procfs.c generates them from.
struct global_file { const char *name; int id; };

static const struct global_file globals[] = {
  { "meminfo",     1  },
  { "uptime",      2  },
  { "cpuinfo",     3  },
  { "version",     4  },
  { "stat",        5  },
  { "loadavg",     6  },
  { "mounts",      7  },
  { "filesystems", 8  },
  { "schfs",       9  },
  { "sysrq",       10 },
  { "schinit",     16 },
  { "pci",         11 },
  { "interrupts",  12 },
  { "devices",     13 },
  { "fsstat",      14 },
  { "netif",       15 },
  // The nameservers DHCP supplied, in resolv.conf's own syntax. See
  // net_resolv() and cmd/resolvconf.
  { "resolv.conf", 18 },
  // "quiet" or "verbose" -- how this machine was booted. See cmd/schinitd.
  { "bootmode",    19 },
  { "bootdebug",   30 },
  // What the slab allocator is holding. See include/core/kmem.h.
  { "slab",        20 },
  // Disks and partitions, in Linux's format. See blkdev_info().
  { "partitions",  21 },
  { "kmsg",        17 },
};
#define NGLOBALS ((int)(sizeof(globals) / sizeof(globals[0])))

// Per-process files. `statm` and `status` are here because ps reads them; a
// missing file is not an error to it, but a present one is cheaper than the
// fallback paths it takes without them.
static const struct global_file pidfiles[] = {
  { "stat",    PF_STAT    },
  { "status",  PF_STATUS  },
  { "cmdline", PF_CMDLINE },
  { "statm",   PF_STATM   },
  { "exe",     PF_EXE     },
  // Three numbers about how the scheduler has treated this process, in the
  // order Linux writes them: weighted CPU nanoseconds consumed, then -- where
  // Linux reports run-queue wait -- this kernel's nice level, then the number
  // of slices it has been given. The middle field differs on purpose and is
  // documented in docs/architecture/scheduler.md; nothing here reads it, and
  // inventing a wait time that is not measured would be worse than saying
  // something true in its place.
  { "schedstat", PF_SCHEDSTAT },
};
#define NPIDFILES ((int)(sizeof(pidfiles) / sizeof(pidfiles[0])))

// ---------------------------------------------------------------------------
// Content
// ---------------------------------------------------------------------------

// A tiny append-only formatter. procfs files are small and generated whole,
// so there is no reason for anything cleverer.
struct sb { char *p; int n, cap; };

static void sbputc(struct sb *s, char c)
{ if (s->n < s->cap - 1) s->p[s->n++] = c; }

static void sbputs(struct sb *s, const char *t)
{ while (*t) sbputc(s, *t++); }

static void sbputi(struct sb *s, int64 v)
{
  char tmp[24];
  int i = 0, neg = 0;
  uint64 u;
  if (v < 0) { neg = 1; u = (uint64)(-v); } else u = (uint64)v;
  do { tmp[i++] = '0' + (int)(u % 10); u /= 10; } while (u);
  if (neg) sbputc(s, '-');
  while (i--) sbputc(s, tmp[i]);
}

// Linux's /proc/<pid>/stat, as far as anything here reads it.
//
// The field order is not negotiable: ps counts fields by position, so a
// missing one silently shifts everything after it. Fields SchOS does not
// track are zero, which is what Linux itself reports for things a given
// process has never done.
static char
state_char(int st)
{
  switch (st) {
  case RUNNING:  return 'R';
  case RUNNABLE: return 'R';
  case SLEEPING: return 'S';
  case ZOMBIE:   return 'Z';
  case EMBRYO:   return 'R';
  default:       return '?';
  }
}

static void
gen_pid_stat(struct sb *s, const struct procinfo *pi)
{
  int i;

  sbputi(s, pi->pid);
  sbputs(s, " (");
  sbputs(s, pi->name);
  sbputs(s, ") ");
  sbputc(s, state_char(pi->state));
  sbputc(s, ' ');
  sbputi(s, pi->ppid);   // 4  ppid
  sbputc(s, ' ');
  sbputi(s, pi->pgid);   // 5  pgrp
  sbputc(s, ' ');
  sbputi(s, pi->sid);    // 6  session
  // 7 tty_nr. Linux's dev_t: major 12 bits above minor's 8. /dev/ttyN is
  // major 1, minor 10+N in devfs, so a process whose ctty is 1 reports
  // (1 << 8) | 11. This was hardcoded 0, and that is why plain `ps` printed a
  // header and nothing else: ps with no arguments shows only the processes on
  // *this* terminal, and nothing could ever match. `ps -e` always worked.
  sbputc(s, ' ');
  sbputi(s, pi->ctty ? ((1 << 8) | (10 + pi->ctty)) : 0);
  sbputs(s, " -1 0"); // 8 tpgid, 9 flags

  // 10..13: minflt cminflt majflt cmajflt. There is no fault accounting, and
  // a fabricated number is worse than none -- top would draw a plausible graph
  // of something nobody measured.
  for (i = 0; i < 4; i++)
    sbputs(s, " 0");

  // 14..17: utime stime cutime cstime, in ticks, which is the unit the field
  // is defined in. These were zero too, which is why `ps` and `top` showed
  // 00:00:00 for everything no matter how long it had been running.
  sbputc(s, ' '); sbputi(s, (int64)pi->utime);
  sbputc(s, ' '); sbputi(s, (int64)pi->stime);
  sbputc(s, ' '); sbputi(s, (int64)pi->cutime);
  sbputc(s, ' '); sbputi(s, (int64)pi->cstime);

  sbputs(s, " 20 0 1 0 0 ");  // priority nice num_threads itrealvalue starttime
  sbputi(s, (int64)pi->sz);   // 23 vsize, in bytes
  sbputc(s, ' ');
  sbputi(s, (int64)(pi->sz / PGSIZE));  // 24 rss, in pages
  sbputs(s, " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n");
}

static void
gen_pid_status(struct sb *s, const struct procinfo *pi)
{
  sbputs(s, "Name:\t");   sbputs(s, pi->name);            sbputc(s, '\n');
  sbputs(s, "State:\t");  sbputc(s, state_char(pi->state));
  sbputs(s, "\nTgid:\t"); sbputi(s, pi->pid);
  sbputs(s, "\nPid:\t");  sbputi(s, pi->pid);
  sbputs(s, "\nPPid:\t"); sbputi(s, pi->ppid);
  sbputs(s, "\nUid:\t");  sbputi(s, pi->uid);
  sbputc(s, '\t');        sbputi(s, pi->uid);
  sbputc(s, '\t');        sbputi(s, pi->uid);
  sbputc(s, '\t');        sbputi(s, pi->uid);
  sbputs(s, "\nVmSize:\t"); sbputi(s, (int64)(pi->sz / 1024));
  sbputs(s, " kB\nThreads:\t1\n");

  // Where Linux keeps them, under the names Linux uses. The two are worth
  // telling apart: a process with a high voluntary count is waiting on
  // something, and one with a high involuntary count is being preempted --
  // which is the difference between "this is slow" and "this is competing".
  sbputs(s, "voluntary_ctxt_switches:\t");
  sbputi(s, (int64)pi->nvcsw);
  sbputs(s, "\nnonvoluntary_ctxt_switches:\t");
  sbputi(s, (int64)pi->nivcsw);
  sbputc(s, '\n');
}

static void
gen_pid_statm(struct sb *s, const struct procinfo *pi)
{
  int64 pages = (int64)(pi->sz / PGSIZE);
  sbputi(s, pages); sbputc(s, ' ');
  sbputi(s, pages); sbputs(s, " 0 0 0 0 0\n");
}

// argv is not kept after exec, so the best honest answer is the process name.
// Linux separates arguments with NUL and ps turns those into spaces; one
// NUL-terminated string is a valid degenerate case of that.
static void
gen_pid_cmdline(struct sb *s, const struct procinfo *pi)
{
  sbputs(s, pi->name);
  sbputc(s, '\0');
}

// Generate the whole of one file into `buf`. Returns its length.
static int
gen_file(uint64 ino, char *buf, int cap)
{
  struct sb s = { buf, 0, cap };
  struct procinfo pi;

  if (INO_IS_GLOBAL(ino))
    return proc_gen_global((int)(ino & 0xFF), buf, cap);

  if (proc_snapshot_pid(INO_PID(ino), &pi) < 0)
    return 0;                    // it exited between lookup and read

  switch (INO_FILE(ino)) {
  case PF_STAT:    gen_pid_stat(&s, &pi);    break;
  case PF_STATUS:  gen_pid_status(&s, &pi);  break;
  case PF_CMDLINE: gen_pid_cmdline(&s, &pi); break;
  case PF_STATM:   gen_pid_statm(&s, &pi);   break;
  case PF_EXE:     sbputs(&s, pi.exe[0] ? pi.exe : pi.name); break;
  case PF_SCHEDSTAT:
    sbputi(&s, (int64)pi.vruntime);   sbputc(&s, ' ');
    sbputi(&s, (int64)pi.nice);       sbputc(&s, ' ');
    sbputi(&s, (int64)(pi.nvcsw + pi.nivcsw));
    sbputc(&s, '\n');
    break;
  default: return 0;
  }
  return s.n;
}

// ---------------------------------------------------------------------------
// fs_ops
// ---------------------------------------------------------------------------

static void
procfs_iload(struct inode *ip)
{
  uint64 ino = ip->inum;

  ip->valid = 1;
  ip->nlink = 1;
  ip->major = ip->minor = 0;
  // Read-only, with two exceptions: /proc/sysrq is the deliberate write target
  // (`echo c > /proc/sysrq` to crash on purpose), and /proc/schinit is where
  // PID 1 publishes its job table. Marking everything 2 made filewrite()
  // refuse with "permission denied", which the USB suite caught immediately --
  // three of its checks drive the disk through sysrq.
  ip->rwmode = (INO_IS_GLOBAL(ino) &&
                ((ino & 0xFF) == 10 || (ino & 0xFF) == 16)) ? 3 : 2;
  ip->showmode = 1;
  ip->atime = ip->mtime = ip->ctime = 0;

  if (ino == PROCFS_ROOT || (!INO_IS_GLOBAL(ino) && INO_FILE(ino) == PF_DIR)) {
    ip->type = T_DIR;
    ip->size = 0;
    ip->nlink = 2;
    return;
  }

  // /proc/<pid>/exe is a symlink, as it is on Linux.
  //
  // It was a regular file holding the path as text, which reads the same from
  // `cat` and is wrong to everything that matters: readlink() refused it with
  // EINVAL, and `ps -o comm` -- which readlinks it -- fell back to the process
  // name in brackets and printed `[init]` for everything.
  //
  // This one file has to report a real size, and every other procfs file
  // deliberately reports zero. readlink(2) bounds its copy by ip->size, so a
  // symlink whose size is 0 resolves to the empty string -- which is worse
  // than the regular file was, because it looks like it worked. Generating it
  // here costs one small render per stat() of one file.
  if (!INO_IS_GLOBAL(ino) && INO_FILE(ino) == PF_EXE) {
    char path[64];
    int len = gen_file(ino, path, sizeof(path));
    ip->type = T_SYMLINK;
    ip->size = len > 0 ? (uint64)len : 0;
    ip->nlink = 1;
    return;
  }

  ip->type = T_FILE;
  // Size is reported as zero.
  //
  // A synthetic file has no length until it is generated, and generating it
  // here -- on every stat(), for every process -- to fill in a number almost
  // nobody reads would be absurd. Linux does exactly the same thing, which is
  // why `ls -l /proc` shows a column of zeroes, and why every tool that reads
  // /proc reads until EOF instead of trusting st_size.
  ip->size = 0;
}

static void procfs_istore(struct inode *ip) { (void)ip; }
static void procfs_ifree(struct inode *ip)  { (void)ip; }
static void procfs_truncate(struct inode *ip, uint64 n) { (void)ip; (void)n; }

static int
procfs_read(struct inode *ip, int user_dst, uint64 dst, uint64 off, uint n)
{
  char *buf;
  int len, r;

  if (ip->type == T_DIR)
    return -EISDIR;

  // The 2048-byte scratch used to be a local. KSTACKSIZE is 4096, so that was
  // half the kernel stack in one array, on a path already several frames deep
  // -- a generator with any locals of its own, or one more frame below it,
  // runs off the end. /proc/schfs is what finally did it: the canary at the
  // bottom of the stack came back zeroed. Nothing about SchFS caused this; it
  // only pushed an existing overrun over the edge.
  //
  // A whole page for a 2 KB buffer is not wasteful here -- kalloc's unit is a
  // page either way, and this is not a hot path.
  // /proc/kmsg is 128 KB and comes out of a ring, so it does not go through
  // the render-the-whole-file-then-slice path below. See klog_read().
  if (INO_IS_GLOBAL(ip->inum) && (int)(ip->inum & 0xFF) == 17) {
    char *k;
    int   got;

    if (n > PGSIZE)
      n = PGSIZE;
    if ((k = kalloc()) == 0)
      return -ENOMEM;
    got = klog_read(off, k, (int)n);
    if (got <= 0) {
      kfree(k);
      return 0;
    }
    r = got;
    if (user_dst) {
      if (copy_to_user(dst, k, got) < 0)
        r = -EFAULT;
    } else {
      memmove((char *)dst, k, got);
    }
    kfree(k);
    return r;
  }

  // The shared buffer in kernel/dev/procfs.c, 64 KB, held until the copy out
  // is done. It was 2 KB from kmalloc, which cut /proc/cpuinfo off at the
  // eighteenth processor.
  int cap;
  buf = proc_genbuf_get(&cap);
  len = gen_file(ip->inum, buf, cap);
  if (off >= (uint64)len) {
    proc_genbuf_put();
    return 0;
  }
  if (off + n > (uint64)len)
    n = (uint)((uint64)len - off);

  r = (int)n;
  if (user_dst) {
    if (copy_to_user(dst, buf + off, n) < 0)
      r = -EFAULT;
  } else {
    memmove((char *)dst, buf + off, n);
  }

  proc_genbuf_put();
  return r;
}

static int
procfs_write(struct inode *ip, int user_src, uint64 src, uint64 off, uint n)
{
  // Only /proc/sysrq accepts writes, and its handler lives with the other
  // generators. Everything else is read-only, and says so.
  if (INO_IS_GLOBAL(ip->inum) && (ip->inum & 0xFF) == 10)
    return proc_write_sysrq(user_src, src, n);
  // /proc/schinit is the job table, and PID 1 is the one that knows it. The
  // kernel holds the text rather than generating it: walking another
  // process's memory to answer a read is not something a procfs generator
  // should be doing, and PID 1 already knows when the table changed.
  if (INO_IS_GLOBAL(ip->inum) && (ip->inum & 0xFF) == 16)
    return proc_write_schinit(user_src, src, n);
  (void)off;
  return -EROFS;
}

// Decimal string -> pid, or -1 if it is not all digits. This is also what
// makes a name like "meminfo" fall through to the global table.
static int
name_to_pid(const char *name)
{
  int v = 0;

  if (!name || !*name)
    return -1;
  while (*name) {
    if (*name < '0' || *name > '9')
      return -1;
    v = v * 10 + (*name++ - '0');
    if (v > 1 << 20)
      return -1;
  }
  return v;
}

static struct inode *
procfs_dirlookup(struct inode *dp, char *name, uint *poff)
{
  struct procinfo pi;
  int i, pid;

  if (poff)
    *poff = 0;

  if (namecmp(name, ".") == 0)
    return iget(dp->dev, dp->inum);
  if (namecmp(name, "..") == 0)
    return iget(dp->dev, PROCFS_ROOT);   // /proc/<pid>/.. is /proc

  // Inside /proc/<pid>/
  if (dp->inum != PROCFS_ROOT) {
    for (i = 0; i < NPIDFILES; i++)
      if (namecmp(name, (char *)pidfiles[i].name) == 0)
        return iget(dp->dev, (dp->inum & ~0xFFULL) | pidfiles[i].id);
    return 0;
  }

  // In /proc itself: a number is a process, a word is a global file.
  pid = name_to_pid(name);
  if (pid > 0) {
    if (proc_snapshot_pid(pid, &pi) < 0)
      return 0;
    return iget(dp->dev, ((uint64)pid << 8) | PF_DIR);
  }

  for (i = 0; i < NGLOBALS; i++)
    if (namecmp(name, (char *)globals[i].name) == 0)
      return iget(dp->dev, PROCFS_GLOBAL | globals[i].id);
  return 0;
}

// Nothing in /proc can be created, removed or linked. Saying so with an error
// is better than a silent no-op: a program that thinks it wrote a file and
// finds it missing later has a much harder problem to debug.
static int procfs_dirlink(struct inode *dp, char *name, uint64 inum)
{ (void)dp; (void)name; (void)inum; return -EROFS; }
static int procfs_dirunlink(struct inode *dp, char *name, uint off)
{ (void)dp; (void)name; (void)off; return -EROFS; }
static int procfs_dirempty(struct inode *dp) { (void)dp; return 0; }
static struct inode *procfs_ialloc(uint dev, short type)
{ (void)dev; (void)type; return 0; }

// One getdents64 record.
struct linux_dirent64 {
  uint64 d_ino;
  int64  d_off;
  ushort d_reclen;
  uchar  d_type;
  char   d_name[];
};

#define DT_DIR_  4
#define DT_REG_  8

static int
emit(char *out, int cap, int used, uint64 ino, int type, const char *name)
{
  struct linux_dirent64 *de;
  int nlen = 0, reclen;

  while (name[nlen]) nlen++;
  reclen = (int)sizeof(*de) + nlen + 1;
  reclen = (reclen + 7) & ~7;          // 8-byte aligned, as getdents64 requires
  if (used + reclen > cap)
    return -1;

  de = (struct linux_dirent64 *)(out + used);
  de->d_ino    = ino;
  de->d_off    = used + reclen;
  de->d_reclen = (ushort)reclen;
  de->d_type   = (uchar)type;
  memmove(de->d_name, name, nlen);
  de->d_name[nlen] = 0;
  return reclen;
}

static void
utoa(uint v, char *buf)
{
  char t[12];
  int i = 0, j = 0;
  do { t[i++] = '0' + (v % 10); v /= 10; } while (v);
  while (i) buf[j++] = t[--i];
  buf[j] = 0;
}

// The listing is generated fresh each time, and *off is an index into it.
//
// A process that exits between two getdents calls therefore shifts everything
// after it up by one, and one entry can be missed. That is a real race and it
// is the same one Linux has -- /proc has no stable ordering to offer, and
// anything that wants a consistent snapshot has to re-read.
static int
procfs_getdents(struct inode *dp, char *out, int cap, uint64 *off)
{
  struct procinfo pi;
  char name[16];
  int used = 0, idx = 0, i, r;
  uint64 want = *off;

  #define EMIT(ino, type, nm) do {                            \
      if (idx++ >= (int)want) {                               \
        r = emit(out, cap, used, (ino), (type), (nm));        \
        if (r < 0) goto full;                                 \
        used += r; (*off)++;                                  \
      }                                                       \
    } while (0)

  if (dp->inum == PROCFS_ROOT) {
    EMIT(PROCFS_ROOT, DT_DIR_, ".");
    EMIT(PROCFS_ROOT, DT_DIR_, "..");
    for (i = 0; i < NGLOBALS; i++)
      EMIT(PROCFS_GLOBAL | globals[i].id, DT_REG_, globals[i].name);
    for (i = 0; proc_snapshot(i, &pi) == 0; i++) {
      utoa((uint)pi.pid, name);
      EMIT(((uint64)pi.pid << 8) | PF_DIR, DT_DIR_, name);
    }
  } else {
    EMIT(dp->inum, DT_DIR_, ".");
    EMIT(PROCFS_ROOT, DT_DIR_, "..");
    for (i = 0; i < NPIDFILES; i++)
      EMIT((dp->inum & ~0xFFULL) | pidfiles[i].id, DT_REG_, pidfiles[i].name);
  }

full:
  #undef EMIT
  return used;
}

const struct fs_ops procfs_ops = {
  .name      = "procfs",
  .iload     = procfs_iload,
  .istore    = procfs_istore,
  .ifree     = procfs_ifree,
  .read      = procfs_read,
  .write     = procfs_write,
  .truncate  = procfs_truncate,
  .dirlookup = procfs_dirlookup,
  .dirlink   = procfs_dirlink,
  .dirunlink = procfs_dirunlink,
  .dirempty  = procfs_dirempty,
  .ialloc    = procfs_ialloc,
  .getdents  = procfs_getdents,
};

// Graft this filesystem over /proc.
//
// Called by init once it has made the directory. The mknod'd files that used
// to live there are simply covered: path resolution stops at the mount point,
// so they are unreachable and harmless. Removing them from init is a separate,
// later cleanup -- doing both at once would make a boot failure ambiguous
// between "the mount is wrong" and "the nodes are missing".
void
procfs_mount(void)
{
  struct inode *dp = namei("/proc");

  if (dp == 0) {
    cprintf("procfs: /proc does not exist; not mounted\n");
    return;
  }
  if (mount_add(dp->dev, dp->inum, PROCFS_DEV, PROCFS_ROOT, "/proc", "proc", "proc") < 0)
    cprintf("procfs: /proc already has something mounted on it\n");
  iput(dp);
}