SchOS/kernel/fs/file.c

#include "core/types.h"
#include "core/defs.h"
#include "uapi/stat.h"
#include "core/param.h"
#include "fs/fs.h"
#include "core/spinlock.h"
#include "core/sleeplock.h"
#include "fs/file.h"
#include "fs/fsstat.h"
#include "uapi/posix.h"
#include "uapi/errno.h"
#include "core/kmem.h"

extern struct devsw devsw[];

// Open file descriptions, shared by every process. NFILE of them in the
// kernel image and more from a slab cache once those are all open; a closed
// one goes on the free list. This was NFILE and then ENFILE, which a few
// processes each holding a hundred descriptors -- an editor, a build, a
// shell -- were already a third of the way to.
struct {
  struct spinlock lock;
  struct file file[NFILE];
  struct file *free;
  struct kmem_cache *more;
  uint n;
  uint64 serial;        // the last ino handed out
} ftable;

void
fileinit(void)
{
  initlock(&ftable.lock, "ftable");
  for(int i = NFILE - 1; i >= 0; i--){
    ftable.file[i].fnext = ftable.free;
    ftable.free = &ftable.file[i];
  }
  ftable.n = NFILE;
}

uint
file_slots(void)
{
  return ftable.n;
}

struct file*
filealloc(void)
{
  struct file *f;

  acquire(&ftable.lock);
  if((f = ftable.free) != 0){
    ftable.free = f->fnext;
  } else {
    // The slab allocator takes only its own lock and kalloc's.
    if(ftable.more == 0)
      ftable.more = kmem_cache_create("file", sizeof(struct file));
    if(ftable.more == 0 || (f = kmem_cache_alloc(ftable.more)) == 0){
      release(&ftable.lock);
      return 0;
    }
    memset(f, 0, sizeof *f);
    ftable.n++;
  }
  f->ref = 1;
  f->ino = ++ftable.serial;
  f->path = 0;
  FSSTAT_TAKE(fsstat.file_used, fsstat.file_peak);   // 0 -> 1
  // Cleared here rather than left to the caller: every creator sets
  // readable/writable/off, but only open() sets nonblock, so a struct file
  // recycled from a non-blocking device would hand its flag to the next
  // pipe or socket that happened to land in this slot.
  f->nonblock = 0;
  release(&ftable.lock);
  return f;
}

struct file*
filedup(struct file *f)
{
  acquire(&ftable.lock);
  if(f->ref < 1)
    panic("filedup");
  f->ref++;
  release(&ftable.lock);
  return f;
}

void
fileclose(struct file *f)
{
  struct file ff;

  acquire(&ftable.lock);
  if(f->ref < 1)
    panic("fileclose");
  if(--f->ref > 0){
    release(&ftable.lock);
    return;
  }
  // Its flock() and OFD locks go with it, before the slot can be handed to
  // someone else: they are keyed by this address.
  lock_close_file(f);
  ff = *f;
  f->ref = 0;
  f->path = 0;
  f->type = FD_NONE;
  f->fnext = ftable.free;
  ftable.free = f;
  FSSTAT_DROP(fsstat.file_used);                          // 1 -> 0
  release(&ftable.lock);

  if(ff.path)
    kmfree(ff.path);

  if(ff.type == FD_MEMFD)
    shm_detach(ff.memfd_seg);           // the last mapper frees the pages
  else if(ff.type == FD_EVENT)
    ev_close(ff.ev);
  else if(ff.type == FD_PIPE)
    pipeclose(ff.pipe, ff.writable);
  else if(ff.type == FD_SOCKET)
    sockclose(ff.sock);
  else if(ff.type == FD_INODE){
    // Give a device a chance to put the hardware back before the inode goes.
    // Done here rather than in exit() because a program that closes the
    // framebuffer without exiting has finished with it just as thoroughly.
    if(ff.ip && ff.ip->type == T_DEV && ff.ip->major >= 0 &&
       ff.ip->major < NDEV && devsw[ff.ip->major].close)
      devsw[ff.ip->major].close(ff.ip);

    begin_op();
    iput(ff.ip);
    end_op();
  }
}

// What /proc/<pid>/fd/N points at, in Linux's spelling: the path, or
// "socket:[ino]", "pipe:[ino]", "anon_inode:[eventpoll]". Returns the length.
// ftable.lock is what keeps `path` alive: fileclose() takes it from the file
// under the lock and frees it after.
int
file_describe(struct file *f, char *buf, int cap)
{
  const char *pre = 0, *s = 0;
  int n = 0;

  acquire(&ftable.lock);
  if(f->ref > 0){
    switch(f->type){
    case FD_INODE:  s = f->path; if(!s) pre = "inode:["; break;
    case FD_SOCKET: pre = "socket:["; break;
    case FD_PIPE:   pre = "pipe:["; break;
    case FD_MEMFD:  s = "/memfd: (deleted)"; break;
    case FD_EVENT:  s = ev_kind_name(f->ev); break;
    default:        break;
    }
  }
  if(pre){
    char num[24];
    int k = 0;
    uint64 v = f->type == FD_INODE && f->ip ? f->ip->inum : f->ino;
    do { num[k++] = '0' + v % 10; v /= 10; } while(v);
    while(*pre && n < cap) buf[n++] = *pre++;
    while(k && n < cap) buf[n++] = num[--k];
    if(n < cap) buf[n++] = ']';
  } else if(s){
    while(*s && n < cap) buf[n++] = *s++;
  }
  release(&ftable.lock);
  return n;
}

int
filestat(struct file *f, struct stat *st)
{
  // A memfd has no inode, so the size is the only field with an answer. It is
  // the field that matters: mmap length and ftruncate round-trips are checked
  // against it, and a caller that fstat()s to find out how big the buffer is
  // gets the truth.
  if(f->type == FD_MEMFD){
    memset(st, 0, sizeof(*st));
    st->type = T_FILE;
    st->size = memfd_size(f->memfd_seg);
    st->nlink = 1;
    return 0;
  }
  if(f->type == FD_INODE){
    ilock(f->ip);
    stati(f->ip, st);
    iunlock(f->ip);
    return 0;
  }
  // An eventfd, timerfd, signalfd or epoll set: Linux calls these anonymous
  // inodes, and fstat succeeds on them with nothing much to say. A program
  // that fstat()s every descriptor it is handed should not fail on one.
  if(f->type == FD_EVENT){
    memset(st, 0, sizeof(*st));
    st->nlink = 1;
    return 0;
  }
  return -1;
}

int
fileread(struct file *f, char *addr, int n)
{
  int r;

  if(f->readable == 0)
    return -1;

  if(f->type == FD_EVENT)
    return ev_read(f, addr, n);

  // A memfd is a file: read() and write() move bytes in and out of the shared
  // pages, and the descriptor carries its own offset like any other. mmap is
  // what it is usually for, but a descriptor that refuses read() is a surprise
  // nobody needs.
  if(f->type == FD_MEMFD){
    r = memfd_rw(f->memfd_seg, 1, (uint64)addr, f->off, n, 0);
    if(r > 0)
      f->off += r;
    return r;
  }

  // O_NONBLOCK on a pipe or a socket. Reported here rather than inside
  // piperead/sockread so that the blocking implementations stay unchanged:
  // this is a check of the same condition their sleep loops wait on, made one
  // level up. Writes still block -- fcntl documents that.
  if(f->type == FD_PIPE){
    // POLLHUP counts as readable here, and leaving it out was a bug: a pipe
    // whose writers have all gone is not "nothing yet", it is end of file, and
    // the answer to a read is 0. Told EAGAIN instead, a caller that polls and
    // reads spins forever waiting for an end that has already happened.
    // sockreadable() has always included the equivalent condition.
    if(f->nonblock &&
       !(pipe_poll(f->pipe, f->readable, f->writable) & (POLLIN | POLLHUP)))
      return -EAGAIN;
    return piperead(f->pipe, addr, n);
  }

  if(f->type == FD_SOCKET){
    if(f->nonblock && !sockreadable(f->sock))
      return -EAGAIN;
    return sockread(f->sock, addr, n);
  }

  if(f->type == FD_INODE){
    ilock(f->ip);
    if(f->ip->type == 3){
      if(f->ip->major < 0 || f->ip->major >= NDEV || !devsw[f->ip->major].read){
        iunlock(f->ip);
        return -1;
      }

      // O_NONBLOCK. Without this a game that opens an input device
      // non-blocking and drains it once per frame stops dead on the first
      // frame -- it blocks in read() waiting for a key, and the result looks
      // exactly like a hung renderer rather than a blocking read.
      if(f->nonblock && devsw[f->ip->major].ready &&
         !devsw[f->ip->major].ready(f->ip)){
        iunlock(f->ip);
        return -EAGAIN;
      }

      r = devsw[f->ip->major].read(f->ip, f->off, addr, n);
      if(r > 0)
        f->off += r;
      iunlock(f->ip);
      return r;
    }

    if((f->ip->rwmode & 2) == 0){
      cprintf("SchOS: permission denied (read)\n");
      iunlock(f->ip);
      return -1;
    }

    // The "recycle bin" check that used to sit here is gone.
    //
    // It made a file that still existed, still had a name and still had blocks
    // refuse every read with a printf -- a state no POSIX program is prepared
    // for. It was reachable only through `delete`, a NO_LINUX syscall from the
    // third-party filesystem this file derives from, which the decision to keep
    // only Linux numbering had already orphaned.
    //
    // The on-disk `showmode` field stays for now: removing it changes the inode
    // layout, and this whole file is being replaced by SchFS. It is simply
    // never consulted.

    if((r = readi(f->ip, 0, (uint64)addr, f->off, n)) > 0)
      f->off += r;
    iunlock(f->ip);
    return r;
  }

  panic("fileread");
}

int
filewrite(struct file *f, char *addr, int n)
{
  int r;

  if(f->writable == 0)
    return -1;

  if(f->type == FD_EVENT)
    return ev_write(f, addr, n);

  if(f->type == FD_MEMFD){
    r = memfd_rw(f->memfd_seg, 1, (uint64)addr, f->off, n, 1);
    if(r > 0)
      f->off += r;
    return r;
  }

  if(f->type == FD_PIPE)
    return pipewrite(f->pipe, addr, n);

  if(f->type == FD_SOCKET)
    return sockwrite(f->sock, addr, n);

  if(f->type == FD_INODE){
    ilock(f->ip);

    if(f->ip->type == 3){
      if(f->ip->major < 0 || f->ip->major >= NDEV || !devsw[f->ip->major].write){
        iunlock(f->ip);
        return -1;
      }
      r = devsw[f->ip->major].write(f->ip, f->off, addr, n);
      if(r > 0)
        f->off += r;
      iunlock(f->ip);
      return r;
    }

    if((f->ip->rwmode & 1) == 0){
      cprintf("SchOS: permission denied (write)\n");
      iunlock(f->ip);
      return -1;
    }

    // Recycle-bin check removed here too; see the note in fileread().
    iunlock(f->ip);

    // How much may go in one call is the filesystem's business. v1 is bounded
    // by its log; SchFS is not bounded at all and would rather have the lot,
    // because it makes one transaction out of whatever it is given.
    uint max = f->ip->ops->max_write;
    int i = 0;
    while(i < n){
      int n1 = n - i;
      if(max && (uint)n1 > max)
        n1 = (int)max;

      begin_op();
      ilock(f->ip);
      if ((r = writei(f->ip, 1, (uint64)(addr + i), f->off, n1)) > 0)
        f->off += r;

      iunlock(f->ip);
      end_op();

      if(r < 0)
        break;
      i += r;
      // Short means full. v1 and SchFS never wrote short, which is why this
      // was a panic; FAT does, when the volume or its root directory fills.
      // write(2) reports what went in, and the next call gets the error.
      if(r != n1)
        break;
    }
    if(i > 0)
      return i;
    return r < 0 ? r : -1;
  }
  panic("filewrite");
}

// poll(2) readiness for one open file. Returns a POLL* bitmask restricted
// to the requested events (plus error/hangup conditions, which POSIX says
// are always reported). Pipes and the console tty have real state; other
// inodes (regular files, block-ish devices) are always ready, per POSIX.
int
file_poll(struct file *f, int events)
{
  if(f->type == FD_EVENT)
    return ev_poll(f, events);

  if(f->type == FD_PIPE)
    return pipe_poll(f->pipe, f->readable, f->writable) &
           (events | POLLERR | POLLHUP);

  if(f->type == FD_SOCKET){
    int rev = 0;
    if(sockreadable(f->sock))
      rev |= POLLIN;
    // A stream socket is writable whenever it has a peer. The buffer can
    // still be full, in which case the write blocks -- same latitude poll()
    // has always had, and the same answer a pipe gives.
    if(sock_hup(f->sock))
      rev |= POLLHUP;
    else
      rev |= POLLOUT;
    return rev & (events | POLLERR | POLLHUP);
  }

  if(f->type == FD_INODE){
    // An input device is only readable when an event is waiting. Reporting it
    // always-ready -- which is what the generic answer below does -- turns a
    // poll loop into a busy loop that blocks in read().
    if(f->ip && f->ip->type == T_DEV && f->ip->major == INPUT_MAJOR){
      int rev = 0;
      if((events & POLLIN) && f->readable && input_ready(f->ip->minor))
        rev |= POLLIN;
      if(events & POLLOUT)
        rev |= POLLOUT;
      return rev;
    }

    if(f->ip && f->ip->type == T_DEV && f->ip->major == CONSOLE){
      int rev = 0;
      if((events & POLLIN) && f->readable && console_input_ready(f->ip ? f->ip->minor : 1))
        rev |= POLLIN;
      if((events & POLLOUT) && f->writable)
        rev |= POLLOUT;
      return rev;
    }
    return events & (POLLIN | POLLOUT);
  }
  return POLLNVAL;
}