SchOS/kernel/fs/fs.c

// SCHFS - File system implementation.
//
// Five layers:
//   + Blocks: allocator for raw disk blocks.
//   + Log: crash recovery for multi-step updates.
//   + Files: inode allocator, reading, writing, metadata.
//   + Directories: inode with special contents (list of other inodes!)
//   + Names: paths like /usr/rtm/schfs/fs.c for convenient naming.
//
// Derived from "New File System For XV6" by Qi Ma, Leyi Pan, Ao Sun and
// Peiran Xu (MIT License), which itself extends xv6's fs.c. The HTree directory
// index, the Ext4-style extents, the inode bitmap and the symlink support all
// come from there. See NOTICE.
//
// Changes from the original xv6-based fs.c:
//   1. CRC32c checksums on every inode and the superblock.
//   2. Timestamps: atime/mtime/ctime written on alloc, read, write, trunc.
//   3. Proper extent tree for T_EXTENT inodes: struct fsv1_extent packed into
//      addrs[], with an indirect extent block for larger files.
//      Replaces the lossy "addr*256|len" encoding that overflowed on large
//      physical block numbers.
//   4. itrunc for T_EXTENT fixed (original leaked the indirect block and had
//      an off-by-one loop termination).
//   5. readi/writei honour the user_dst/user_src flag via copyout/copyin.
//   6. tree_search uses OVERFLOW_DIR_START/END instead of bare magic numbers.
//   7. contiguous_block_allocation: removed the dangerous mid-transaction
//      end_op()/begin_op() calls.
//   8. divesymlink: fixed one-byte buffer overrun in the "short read" check.
//   9. updatesb: writes CRC32c checksum before flushing to disk.
//  10. Dead stubs (check_log_full, check_free, cal_block_index, tree_add)
//      removed.
// This FS is a joke lol, suck my dick chinese people lol

#include "types.h"
#include "x86.h"
#include "defs.h"
#include "param.h"
#include "stat.h"
#include "spinlock.h"
#include "proc.h"
#include "sleeplock.h"
#include "fs.h"
#include "buf.h"
#include "fsstat.h"
#include "file.h"
#include "fat.h"
#include "errno.h"

extern uint uptime(void);

#define min(a, b) ((a) < (b) ? (a) : (b))

// ===========================================================================
// CRC32c (Castagnoli polynomial 0x82F63B78)
// Must be byte-for-byte identical to the copy in mkfs.c.
// ===========================================================================

static uint crc32c_table[256];
static int  crc32c_ready = 0;

static void
crc32c_init(void)
{
  uint i, j, crc;
  for (i = 0; i < 256; i++) {
    crc = i;
    for (j = 0; j < 8; j++)
      crc = (crc >> 1) ^ (crc & 1 ? 0x82F63B78U : 0);
    crc32c_table[i] = crc;
  }
  crc32c_ready = 1;
}

static uint
crc32c(uint crc, const void *buf, uint len)
{
  const uchar *p = (const uchar *)buf;
  crc = ~crc;
  while (len--)
    crc = (crc >> 8) ^ crc32c_table[(crc ^ *p++) & 0xFF];
  return ~crc;
}

// Compute CRC32c of a dinode.  The checksum field itself is zeroed for the
// computation so the result is stable regardless of what was stored there.
static uint
inode_checksum(struct dinode *dip)
{
  uint saved    = dip->checksum;
  dip->checksum = 0;
  uint ck       = crc32c(0, dip, sizeof(*dip));
  dip->checksum = saved;
  return ck;
}

// ===========================================================================
// Superblock
// ===========================================================================

struct superblock sb;
struct spinlock  sblock;

void
readsb(int dev, struct superblock *sb)
{
  struct buf *bp = bread(dev, 1);
  memmove(sb, bp->data, sizeof(*sb));
  brelse(bp);
}

static void
updatesb(int dev, struct superblock *sb)
{
  // Recompute checksum with the checksum field zeroed.
  sb->checksum = 0;
  sb->checksum = crc32c(0, sb, sizeof(*sb));

  struct buf *bp = bread(dev, 1);
  memmove(bp->data, sb, sizeof(*sb));
  log_write(bp);
  brelse(bp);
}

void
fsinit(int dev)
{
  if (!crc32c_ready)
    crc32c_init();

  initlock(&sblock, "superblock");
  readsb(dev, &sb);

  if (sb.magic != FSMAGIC)
    panic("fsinit: invalid file system");

  // Validate superblock checksum.  A zero checksum means an old image that
  // predates this field; skip validation so we stay backward-compatible.
  if (sb.checksum != 0) {
    uint stored   = sb.checksum;
    sb.checksum   = 0;
    uint computed = crc32c(0, &sb, sizeof(sb));
    sb.checksum   = stored;
    if (computed != stored)
      panic("fsinit: superblock checksum mismatch");
  }

  initlog(dev);
}

// ===========================================================================
// Block allocator
// ===========================================================================

static void
bzero(int dev, int bno)
{
  struct buf *bp = bread(dev, bno);
  memset(bp->data, 0, BSIZE);
  // Direct write (not logged): the bitmap log_write in balloc is the commit
  // point. If we crash before commit, the bitmap reverts and this block stays
  // free — the zero content is harmless. Logging every new block would burn
  // most of LOGSIZE when tree_expand pre-allocates hash-tree blocks.
  bwrite(bp);
  brelse(bp);
}

static uint
balloc(uint dev)
{
  int b, bi, m;
  struct buf *bp = 0;

  for (b = 0; b < sb.size; b += BPB) {
    bp = bread(dev, BBLOCK(b, sb));
    for (bi = 0; bi < BPB && b + bi < sb.size; bi++) {
      m = 1 << (bi % 8);
      if ((bp->data[bi / 8] & m) == 0) {
        bp->data[bi / 8] |= m;
        log_write(bp);
        brelse(bp);
        acquire(&sblock);
        sb.freeblocks--;
        release(&sblock);
        updatesb(dev, &sb);
        bzero(dev, b + bi);
        return b + bi;
      }
    }
    brelse(bp);
  }
  panic("balloc: out of blocks");
}

static void
bfree(int dev, uint b)
{
  struct buf *bp;
  int bi, m;

  bp = bread(dev, BBLOCK(b, sb));
  bi = b % BPB;
  m  = 1 << (bi % 8);
  if ((bp->data[bi / 8] & m) == 0)
    panic("bfree: freeing free block");
  bp->data[bi / 8] &= ~m;
  log_write(bp);
  brelse(bp);
  acquire(&sblock);
  sb.freeblocks++;
  release(&sblock);
  updatesb(dev, &sb);
}

// ===========================================================================
// Inode table
// ===========================================================================

struct {
  struct spinlock lock;
  struct inode    inode[NINODE];
} itable;

void
iinit(void)
{
  initlock(&itable.lock, "itable");
  for (int i = 0; i < NINODE; i++)
    initsleeplock(&itable.inode[i].lock, "inode");
}

struct inode *iget(uint dev, uint64 inum);

// ===========================================================================
// Filesystem dispatch
//
// Every entry point below the syscall layer goes through ip->ops, which iget()
// picks from the device number. The v1 implementations keep their original
// bodies and are simply named v1_*; the public functions are the thin wrappers
// at the bottom of this file.
//
// The one rule worth stating: a filesystem's hooks are only ever called on
// inodes belonging to that filesystem, so none of them needs to check.
// ===========================================================================

static void          v1_iload(struct inode *);
static void          v1_istore(struct inode *);
static void          v1_ifree(struct inode *);
static int           v1_read(struct inode *, int, uint64, uint64, uint);
static int           v1_write(struct inode *, int, uint64, uint64, uint);
static void          v1_truncate(struct inode *, uint64);
static void          v1_itrunc_all(struct inode *);
static void          v1_itrunc_to(struct inode *, uint64);
static struct inode *v1_dirlookup(struct inode *, char *, uint *);
static int           v1_dirlink(struct inode *, char *, uint64);
static int           v1_dirunlink(struct inode *, char *, uint);
static int           v1_dirempty(struct inode *);
static struct inode *v1_ialloc(uint, short);
static int           v1_getdents(struct inode *, char *, int, uint64 *);

static const struct fs_ops v1_ops = {
  .name      = "schfs-v1",
  .iload     = v1_iload,
  .istore    = v1_istore,
  .ifree     = v1_ifree,
  .read      = v1_read,
  .write     = v1_write,
  .truncate  = v1_truncate,
  .dirlookup = v1_dirlookup,
  .dirlink   = v1_dirlink,
  .dirunlink = v1_dirunlink,
  .dirempty  = v1_dirempty,
  .ialloc    = v1_ialloc,
  .getdents  = v1_getdents,
  // A write has to fit in the on-disk log, with room for the inode block, the
  // bitmap block and the log header. This is the constant filewrite used to
  // apply to every filesystem, SchFS included.
  .max_write = ((LOGSIZE-1-1-2) / 2) * 512,
};

// Defined in fs/schfs/vfs.c, which is only linked when SchFS is built in.
extern const struct fs_ops schfs_ops;
extern const struct fs_ops fat_ops;
// Defined in fs/procfs.c -- the synthetic /proc.
extern const struct fs_ops procfs_ops;
extern const struct fs_ops devfs_ops;
extern const struct fs_ops sysfs_ops;

// Is this filesystem read-only?
//
// A mounted snapshot is. Callers must ask *before* they start mutating: the
// create path allocates an inode and then panics if the dirlink fails, which is
// a reasonable assumption about a writable filesystem and a crash on a
// read-only one.
int
fs_readonly(uint dev)
{
  return SCHFS_IS_SNAP_DEV(dev) || (FAT_IS_DEV(dev) && fat_readonly(dev));
}

static const struct fs_ops *
ops_for(uint dev)
{
  // Every mounted SchFS volume, and every mounted snapshot. One table serves
  // them all: the operations take an inode, an inode carries its device, and
  // kmount.c turns that device into the right volume.
  if (SCHFS_IS_ANY_DEV(dev)) return &schfs_ops;
  if (dev == PROCFS_DEV) return &procfs_ops;
  if (dev == DEVFS_DEV)  return &devfs_ops;
  if (dev == SYSFS_DEV)  return &sysfs_ops;
  if (FAT_IS_DEV(dev))   return &fat_ops;
  return &v1_ops;
}

// ===========================================================================
// Inode bitmap helpers
// ===========================================================================

static int
searchibmap(uint dev)
{
  int b, bi, m;
  struct buf *bp = 0;

  for (b = 0; b < sb.ninodes; b += BPB) {
    bp = bread(dev, IBBLOCK(b, sb));
    for (bi = 0; bi < BPB && b + bi < sb.ninodes; bi++) {
      m = 1 << (bi % 8);
      if ((bp->data[bi / 8] & m) == 0) {
        bp->data[bi / 8] |= m;
        log_write(bp);
        brelse(bp);
        acquire(&sblock);
        sb.freeinodes--;
        release(&sblock);
        updatesb(dev, &sb);
        return b + bi;
      }
    }
    brelse(bp);
  }
  panic("searchibmap: out of free inodes");
}

static void
ifree(int dev, int inum)
{
  struct buf *bp;
  int bi, m;

  bp = bread(dev, IBBLOCK(inum, sb));
  bi = inum % BPB;
  m  = 1 << (bi % 8);
  if ((bp->data[bi / 8] & m) == 0)
    panic("ifree: freeing free inode");
  bp->data[bi / 8] &= ~m;
  log_write(bp);
  brelse(bp);
  acquire(&sblock);
  sb.freeinodes++;
  release(&sblock);
  updatesb(dev, &sb);
}

// ===========================================================================
// Timestamp helper
// ===========================================================================

// Returns the current time in seconds since the epoch.
//
// It used to return uptime() under a comment saying to replace it with a real
// wall-clock source "if your kernel has one". It has one -- see wallclock() in
// kernel/core/time.c -- and the same fix went into SchFS's know(), where the
// symptom was visible: every file created after boot was stamped 1970, and the
// timestamps advanced a hundred times too fast because uptime() counts 10 ms
// ticks and they were being read as seconds.
static uint
now_sec(void)
{
  uint64 sec;
  wallclock(&sec, 0);
  return (uint)sec;
}

// ===========================================================================
// Inode allocation / update
// ===========================================================================

struct inode *
v1_ialloc(uint dev, short type)
{
  int inum = searchibmap(dev);

  struct buf    *bp  = bread(dev, IBLOCK(inum, sb));
  struct dinode *dip = (struct dinode *)bp->data + inum % IPB;

  memset(dip, 0, sizeof(*dip));
  dip->type      = type;
  dip->rwmode    = RW;
  dip->supermode = 0;
  dip->showmode  = 1;

  uint ts        = now_sec();
  dip->atime     = ts;
  dip->mtime     = ts;
  dip->ctime     = ts;

  dip->checksum  = inode_checksum(dip);

  log_write(bp);
  brelse(bp);
  return iget(dev, inum);
}

void
v1_istore(struct inode *ip)
{
  struct buf    *bp  = bread(ip->dev, IBLOCK(ip->inum, sb));
  struct dinode *dip = (struct dinode *)bp->data + ip->inum % IPB;

  dip->type      = ip->type;
  dip->major     = ip->major;
  dip->minor     = ip->minor;
  // v1's on-disk field is a short and cannot be widened without changing the
  // format, so the narrowing is the format's, not an accident of this line.
  // The in-memory count is an int now because SchFS needs the range.
  dip->nlink     = (short)ip->nlink;
  dip->size      = ip->size;
  dip->rwmode    = ip->rwmode;
  dip->supermode = ip->supermode;
  dip->showmode  = ip->showmode;
  dip->atime     = ip->atime;
  dip->mtime     = ip->mtime;
  dip->ctime     = ip->ctime;

  memmove(dip->addrs, ip->addrs, sizeof(ip->addrs));

  dip->checksum  = inode_checksum(dip);

  log_write(bp);
  brelse(bp);
}

// ===========================================================================
// iget / idup / ilock / iunlock / iput
// ===========================================================================

struct inode *
iget(uint dev, uint64 inum)
{
  struct inode *ip, *empty;

  acquire(&itable.lock);
  empty = 0;
  for (ip = &itable.inode[0]; ip < &itable.inode[NINODE]; ip++) {
    if (ip->ref > 0 && ip->dev == dev && ip->inum == inum) {
      ip->ref++;
      release(&itable.lock);
      return ip;
    }
    if (empty == 0 && ip->ref == 0)
      empty = ip;
  }
  if (empty == 0)
    panic("iget: no inodes");

  ip        = empty;
  ip->dev   = dev;
  ip->inum  = inum;
  ip->ref   = 1;
  FSSTAT_TAKE(fsstat.inode_used, fsstat.inode_peak);   // 0 -> 1
  ip->valid = 0;
  ip->ops   = ops_for(dev);
  release(&itable.lock);
  return ip;
}

// Is (dev, inum) in the inode cache with a reference? For a filesystem that
// keeps its own table of inode numbers and wants to know which it can drop.
int
iheld(uint dev, uint64 inum)
{
  int held = 0;
  acquire(&itable.lock);
  for (struct inode *ip = &itable.inode[0]; ip < &itable.inode[NINODE]; ip++)
    if (ip->ref > 0 && ip->dev == dev && ip->inum == inum) {
      held = 1;
      break;
    }
  release(&itable.lock);
  return held;
}

// Is anything on dev held, other than `except` (a mounted root, which the
// mount itself holds)? For umount.
int
ibusy(uint dev, uint64 except)
{
  int busy = 0;
  acquire(&itable.lock);
  for (struct inode *ip = &itable.inode[0]; ip < &itable.inode[NINODE]; ip++)
    if (ip->ref > 0 && ip->dev == dev && ip->inum != except) {
      busy = 1;
      break;
    }
  release(&itable.lock);
  return busy;
}

struct inode *
idup(struct inode *ip)
{
  acquire(&itable.lock);
  ip->ref++;
  release(&itable.lock);
  return ip;
}

static void
v1_iload(struct inode *ip)
{
  struct buf    *bp  = bread(ip->dev, IBLOCK(ip->inum, sb));
  struct dinode *dip = (struct dinode *)bp->data + ip->inum % IPB;

  // Validate checksum.  Zero means an old image without checksums; skip.
  if (dip->checksum != 0) {
    uint computed = inode_checksum(dip);
    if (computed != dip->checksum)
      panic("ilock: inode checksum mismatch");
  }

  ip->type      = dip->type;
  ip->major     = dip->major;
  ip->minor     = dip->minor;
  ip->nlink     = dip->nlink;
  ip->size      = dip->size;
  ip->rwmode    = dip->rwmode;
  ip->supermode = dip->supermode;
  ip->showmode  = dip->showmode;
  ip->atime     = dip->atime;
  ip->mtime     = dip->mtime;
  ip->ctime     = dip->ctime;

  memmove(ip->addrs, dip->addrs, sizeof(ip->addrs));
  brelse(bp);
  ip->valid = 1;

  if (ip->type == 0)
    panic("ilock: no type");
}

void
ilock(struct inode *ip)
{
  if (ip == 0 || ip->ref < 1)
    panic("ilock");

  acquiresleep(&ip->lock);

  if (ip->valid == 0)
    ip->ops->iload(ip);
}

void
iunlock(struct inode *ip)
{
  if (ip == 0 || !holdingsleep(&ip->lock) || ip->ref < 1)
    panic("iunlock");
  releasesleep(&ip->lock);
}

void
iput(struct inode *ip)
{
  acquire(&itable.lock);

  if (ip->ref == 1 && ip->valid && ip->nlink == 0) {
    acquiresleep(&ip->lock);
    release(&itable.lock);

    ip->ops->ifree(ip);
    ip->valid = 0;

    releasesleep(&ip->lock);
    acquire(&itable.lock);
  }

  if (--ip->ref == 0)
    FSSTAT_DROP(fsstat.inode_used);
  release(&itable.lock);
}

void
iunlockput(struct inode *ip)
{
  iunlock(ip);
  iput(ip);
}

// ===========================================================================
// Extent-tree block mapping  (T_EXTENT inodes)
//
// Inline extents: addrs[0..NDIRECT-1] hold packed struct fsv1_extent.
//   MAX_INLINE_EXTENTS = NDIRECT*sizeof(uint)/sizeof(fsv1_extent) = 4.
// Indirect extent block: addrs[NDIRECT] → physical block containing
//   BSIZE/sizeof(fsv1_extent) = 85 more extents.
// addrs[NDIRECT+1] and addrs[NDIRECT+2] are unused for T_EXTENT (future use).
// ===========================================================================

// Find the physical address for logical block bn.
// Allocates a new extent entry (and physical block) if alloc != 0.
static uint
bmap_extent(struct inode *ip, uint bn, int alloc)
{
  struct fsv1_extent *ext = (struct fsv1_extent *)ip->addrs;
  int n = 0;

  // --- Search inline extents ---
  for (n = 0; n < (int)MAX_INLINE_EXTENTS; n++) {
    if (ext[n].len == 0)
      break;
    if (bn >= ext[n].start_block && bn < ext[n].start_block + ext[n].len)
      return ext[n].start_blkaddr + (bn - ext[n].start_block);
  }

  // --- Search indirect extent block ---
  if (ip->addrs[NDIRECT] != 0) {
    struct buf          *bp   = bread(ip->dev, ip->addrs[NDIRECT]);
    struct fsv1_extent *iext = (struct fsv1_extent *)bp->data;
    int nind = BSIZE / sizeof(struct fsv1_extent);

    for (int i = 0; i < nind; i++) {
      if (iext[i].len == 0)
        break;
      if (bn >= iext[i].start_block && bn < iext[i].start_block + iext[i].len) {
        uint addr = iext[i].start_blkaddr + (bn - iext[i].start_block);
        brelse(bp);
        return addr;
      }
    }
    brelse(bp);
  }

  if (!alloc)
    return 0;

  // --- Allocate a new physical block ---
  uint addr = balloc(ip->dev);

  // Try to extend the last inline extent if blocks are contiguous.
  if (n > 0) {
    struct fsv1_extent *last = &ext[n - 1];
    if (bn == last->start_block + last->len &&
      addr == last->start_blkaddr + last->len) {
      last->len++;
    return addr;
      }
  }

  // Fits in a new inline slot?
  if (n < (int)MAX_INLINE_EXTENTS) {
    ext[n].start_block   = bn;
    ext[n].start_blkaddr = addr;
    ext[n].len           = 1;
    return addr;
  }

  // Spill to indirect extent block.
  if (ip->addrs[NDIRECT] == 0)
    ip->addrs[NDIRECT] = balloc(ip->dev);

  struct buf          *bp   = bread(ip->dev, ip->addrs[NDIRECT]);
  struct fsv1_extent *iext = (struct fsv1_extent *)bp->data;
  int nind = BSIZE / sizeof(struct fsv1_extent);

  for (int i = 0; i < nind; i++) {
    if (iext[i].len == 0) {
      // Try contiguity with previous indirect extent.
      if (i > 0 &&
        bn   == iext[i-1].start_block   + iext[i-1].len &&
        addr == iext[i-1].start_blkaddr + iext[i-1].len) {
        iext[i-1].len++;
        } else {
          iext[i].start_block   = bn;
          iext[i].start_blkaddr = addr;
          iext[i].len           = 1;
        }
        log_write(bp);
      brelse(bp);
      return addr;
    }
  }
  brelse(bp);
  panic("bmap_extent: out of extent entries");
}

// Free all data blocks referenced by a T_EXTENT inode.
static void
itrunc_extent(struct inode *ip)
{
  struct fsv1_extent *ext = (struct fsv1_extent *)ip->addrs;

  // Free inline extents.
  for (int n = 0; n < (int)MAX_INLINE_EXTENTS; n++) {
    if (ext[n].len == 0)
      break;
    for (uint b = 0; b < ext[n].len; b++)
      bfree(ip->dev, ext[n].start_blkaddr + b);
    ext[n].start_block   = 0;
    ext[n].start_blkaddr = 0;
    ext[n].len           = 0;
  }

  // Free indirect extent block and its extents.
  if (ip->addrs[NDIRECT]) {
    struct buf          *bp   = bread(ip->dev, ip->addrs[NDIRECT]);
    struct fsv1_extent *iext = (struct fsv1_extent *)bp->data;
    int nind = BSIZE / sizeof(struct fsv1_extent);

    for (int i = 0; i < nind; i++) {
      if (iext[i].len == 0)
        break;
      for (uint b = 0; b < iext[i].len; b++)
        bfree(ip->dev, iext[i].start_blkaddr + b);
    }
    brelse(bp);
    bfree(ip->dev, ip->addrs[NDIRECT]);
    ip->addrs[NDIRECT] = 0;
  }
}

// ===========================================================================
// bmap: logical block → physical block  (dispatch by inode type)
// ===========================================================================

static uint
bmap(struct inode *ip, uint bn)
{
  uint addr, *a;
  struct buf *bp;

  if (ip->type == T_EXTENT)
    return bmap_extent(ip, bn, 1);

  // Classic direct / singly-indirect / doubly-indirect / triply-indirect.

  if (bn < NDIRECT) {
    if ((addr = ip->addrs[bn]) == 0)
      ip->addrs[bn] = addr = balloc(ip->dev);
    return addr;
  }
  bn -= NDIRECT;

  // Singly indirect
  if (bn < NINDIRECT) {
    if ((addr = ip->addrs[NDIRECT]) == 0)
      ip->addrs[NDIRECT] = addr = balloc(ip->dev);
    bp = bread(ip->dev, addr);
    a  = (uint *)bp->data;
    if ((addr = a[bn]) == 0) {
      a[bn] = addr = balloc(ip->dev);
      log_write(bp);
    }
    brelse(bp);
    return addr;
  }
  bn -= NINDIRECT;

  // Doubly indirect
  if (bn < NINDIRECT * NINDIRECT) {
    if ((addr = ip->addrs[NDIRECT + 1]) == 0)
      ip->addrs[NDIRECT + 1] = addr = balloc(ip->dev);

    bp = bread(ip->dev, addr);
    a  = (uint *)bp->data;
    if ((addr = a[bn / NINDIRECT]) == 0) {
      a[bn / NINDIRECT] = addr = balloc(ip->dev);
      log_write(bp);
    }
    brelse(bp);

    bp = bread(ip->dev, addr);
    a  = (uint *)bp->data;
    if ((addr = a[bn % NINDIRECT]) == 0) {
      a[bn % NINDIRECT] = addr = balloc(ip->dev);
      log_write(bp);
    }
    brelse(bp);
    return addr;
  }
  bn -= NINDIRECT * NINDIRECT;

  // Triply indirect
  if (bn < NINDIRECT * NINDIRECT * NINDIRECT) {
    if ((addr = ip->addrs[NDIRECT + 2]) == 0)
      ip->addrs[NDIRECT + 2] = addr = balloc(ip->dev);

    bp = bread(ip->dev, addr);
    a  = (uint *)bp->data;
    uint idx1 = bn / (NINDIRECT * NINDIRECT);
    if ((addr = a[idx1]) == 0) {
      a[idx1] = addr = balloc(ip->dev);
      log_write(bp);
    }
    brelse(bp);

    bp = bread(ip->dev, addr);
    a  = (uint *)bp->data;
    uint idx2 = (bn / NINDIRECT) % NINDIRECT;
    if ((addr = a[idx2]) == 0) {
      a[idx2] = addr = balloc(ip->dev);
      log_write(bp);
    }
    brelse(bp);

    bp = bread(ip->dev, addr);
    a  = (uint *)bp->data;
    uint idx3 = bn % NINDIRECT;
    if ((addr = a[idx3]) == 0) {
      a[idx3] = addr = balloc(ip->dev);
      log_write(bp);
    }
    brelse(bp);
    return addr;
  }

  panic("bmap: out of range");
}

// ===========================================================================
// itrunc: free all data blocks
// ===========================================================================

void
v1_itrunc_all(struct inode *ip)
{
  int i, j;
  struct buf *bp;
  uint *a;

  if (ip->type == T_EXTENT) {
    itrunc_extent(ip);
    ip->size  = 0;
    ip->ctime = now_sec();
    iupdate(ip);
    return;
  }

  // Direct blocks
  for (i = 0; i < NDIRECT; i++) {
    if (ip->addrs[i]) {
      bfree(ip->dev, ip->addrs[i]);
      ip->addrs[i] = 0;
    }
  }

  // Singly indirect
  if (ip->addrs[NDIRECT]) {
    bp = bread(ip->dev, ip->addrs[NDIRECT]);
    a  = (uint *)bp->data;
    for (j = 0; j < NINDIRECT; j++)
      if (a[j]) bfree(ip->dev, a[j]);
    brelse(bp);
    bfree(ip->dev, ip->addrs[NDIRECT]);
    ip->addrs[NDIRECT] = 0;
  }

  // Doubly indirect
  if (ip->addrs[NDIRECT + 1]) {
    bp = bread(ip->dev, ip->addrs[NDIRECT + 1]);
    a  = (uint *)bp->data;
    for (j = 0; j < NINDIRECT; j++) {
      if (a[j]) {
        struct buf *bp2 = bread(ip->dev, a[j]);
        uint *a2 = (uint *)bp2->data;
        for (int k = 0; k < NINDIRECT; k++)
          if (a2[k]) bfree(ip->dev, a2[k]);
        brelse(bp2);
        bfree(ip->dev, a[j]);
      }
    }
    brelse(bp);
    bfree(ip->dev, ip->addrs[NDIRECT + 1]);
    ip->addrs[NDIRECT + 1] = 0;
  }

  // Triply indirect
  if (ip->addrs[NDIRECT + 2]) {
    bp = bread(ip->dev, ip->addrs[NDIRECT + 2]);
    a  = (uint *)bp->data;
    for (j = 0; j < NINDIRECT; j++) {
      if (!a[j]) continue;
      struct buf *bp2 = bread(ip->dev, a[j]);
      uint *a2 = (uint *)bp2->data;
      for (int k = 0; k < NINDIRECT; k++) {
        if (!a2[k]) continue;
        struct buf *bp3 = bread(ip->dev, a2[k]);
        uint *a3 = (uint *)bp3->data;
        for (int l = 0; l < NINDIRECT; l++)
          if (a3[l]) bfree(ip->dev, a3[l]);
        brelse(bp3);
        bfree(ip->dev, a2[k]);
      }
      brelse(bp2);
      bfree(ip->dev, a[j]);
    }
    brelse(bp);
    bfree(ip->dev, ip->addrs[NDIRECT + 2]);
    ip->addrs[NDIRECT + 2] = 0;
  }

  ip->size  = 0;
  ip->ctime = now_sec();
  iupdate(ip);
}

// ===========================================================================
// itrunc_to: shrink inode to newsize, freeing blocks beyond it.
// ip must be locked.  newsize must be <= ip->size.
// Triply-indirect partial truncate is not implemented; those blocks are
// only fully freed when newsize falls below their range start.
// ===========================================================================

void
v1_itrunc_to(struct inode *ip, uint64 newsize)
{
  int i, j;
  struct buf *bp;
  uint *a;

  if (newsize >= ip->size)
    return;

  if (ip->type == T_EXTENT) {
    if (newsize == 0)
      itrunc_extent(ip);
    ip->size  = newsize;
    ip->ctime = now_sec();
    iupdate(ip);
    return;
  }

  // first block index that is no longer needed
  uint first_free = (newsize == 0) ? 0 : (newsize - 1) / BSIZE + 1;

  // Direct blocks
  for (i = first_free; i < NDIRECT; i++) {
    if (ip->addrs[i]) {
      bfree(ip->dev, ip->addrs[i]);
      ip->addrs[i] = 0;
    }
  }

  // Singly indirect (logical blocks NDIRECT .. NDIRECT+NINDIRECT-1)
  if (ip->addrs[NDIRECT]) {
    if (first_free <= NDIRECT) {
      bp = bread(ip->dev, ip->addrs[NDIRECT]);
      a  = (uint *)bp->data;
      for (j = 0; j < NINDIRECT; j++)
        if (a[j]) bfree(ip->dev, a[j]);
      brelse(bp);
      bfree(ip->dev, ip->addrs[NDIRECT]);
      ip->addrs[NDIRECT] = 0;
    } else {
      uint local = first_free - NDIRECT;
      if (local < NINDIRECT) {
        bp = bread(ip->dev, ip->addrs[NDIRECT]);
        a  = (uint *)bp->data;
        int mod = 0;
        for (j = local; j < NINDIRECT; j++)
          if (a[j]) { bfree(ip->dev, a[j]); a[j] = 0; mod = 1; }
        if (mod) log_write(bp);
        brelse(bp);
      }
    }
  }

  // Doubly indirect (logical blocks NDIRECT+NINDIRECT .. +NINDIRECT^2-1)
  if (ip->addrs[NDIRECT + 1]) {
    uint di_start = NDIRECT + NINDIRECT;
    if (first_free <= di_start) {
      bp = bread(ip->dev, ip->addrs[NDIRECT + 1]);
      a  = (uint *)bp->data;
      for (j = 0; j < NINDIRECT; j++) {
        if (!a[j]) continue;
        struct buf *bp2 = bread(ip->dev, a[j]);
        uint *a2 = (uint *)bp2->data;
        for (int k = 0; k < NINDIRECT; k++)
          if (a2[k]) bfree(ip->dev, a2[k]);
        brelse(bp2);
        bfree(ip->dev, a[j]);
      }
      brelse(bp);
      bfree(ip->dev, ip->addrs[NDIRECT + 1]);
      ip->addrs[NDIRECT + 1] = 0;
    } else {
      uint off = first_free - di_start;
      if (off < (uint)(NINDIRECT * NINDIRECT)) {
        uint idx1 = off / NINDIRECT;
        uint idx2 = off % NINDIRECT;
        bp = bread(ip->dev, ip->addrs[NDIRECT + 1]);
        a  = (uint *)bp->data;
        int bp_mod = 0;
        for (j = idx1; j < NINDIRECT; j++) {
          if (!a[j]) continue;
          struct buf *bp2 = bread(ip->dev, a[j]);
          uint *a2 = (uint *)bp2->data;
          int bp2_mod = 0;
          int k_start = (j == (int)idx1) ? (int)idx2 : 0;
          for (int k = k_start; k < NINDIRECT; k++)
            if (a2[k]) { bfree(ip->dev, a2[k]); a2[k] = 0; bp2_mod = 1; }
          if (bp2_mod) log_write(bp2);
          brelse(bp2);
          if (k_start == 0) {
            bfree(ip->dev, a[j]);
            a[j] = 0;
            bp_mod = 1;
          }
        }
        if (bp_mod) log_write(bp);
        brelse(bp);
      }
    }
  }

  // Triply indirect: only fully freed when first_free is at or before its range
  if (ip->addrs[NDIRECT + 2]) {
    uint ti_start = NDIRECT + NINDIRECT + NINDIRECT * NINDIRECT;
    if (first_free <= ti_start) {
      bp = bread(ip->dev, ip->addrs[NDIRECT + 2]);
      a  = (uint *)bp->data;
      for (j = 0; j < NINDIRECT; j++) {
        if (!a[j]) continue;
        struct buf *bp2 = bread(ip->dev, a[j]);
        uint *a2 = (uint *)bp2->data;
        for (int k = 0; k < NINDIRECT; k++) {
          if (!a2[k]) continue;
          struct buf *bp3 = bread(ip->dev, a2[k]);
          uint *a3 = (uint *)bp3->data;
          for (int l = 0; l < NINDIRECT; l++)
            if (a3[l]) bfree(ip->dev, a3[l]);
          brelse(bp3);
          bfree(ip->dev, a2[k]);
        }
        brelse(bp2);
        bfree(ip->dev, a[j]);
      }
      brelse(bp);
      bfree(ip->dev, ip->addrs[NDIRECT + 2]);
      ip->addrs[NDIRECT + 2] = 0;
    }
  }

  ip->size  = newsize;
  ip->mtime = now_sec();
  ip->ctime = now_sec();
  iupdate(ip);
}

// ===========================================================================
// stati
// ===========================================================================

void
stati(struct inode *ip, struct stat *st)
{
  struct stat kst;
  memset(&kst, 0, sizeof(kst));

  kst.dev       = ip->dev;
  kst.ino       = ip->inum;
  kst.type      = ip->type;
  kst.nlink     = ip->nlink;
  kst.size      = ip->size;
  kst.rwmode    = ip->rwmode;
  kst.supermode = ip->supermode;
  kst.showmode  = ip->showmode;
  kst.atime     = ip->atime;
  kst.mtime     = ip->mtime;
  kst.ctime     = ip->ctime;

  for (int i = 0; i < NDIRECT + 3; i++)
    kst.addrs[i] = ip->addrs[i];

  memmove(st, &kst, sizeof(kst));
}

// ===========================================================================
// readi / writei
// ===========================================================================

int
v1_read(struct inode *ip, int user_dst, uint64 dst, uint64 off, uint n)
{
  int tot = 0;
  struct buf *bp;

  if (off >= ip->size)
    return 0;
  if (off + n > ip->size)
    n = ip->size - off;

  while (tot < (int)n) {
    uint addr = bmap(ip, off / BSIZE);
    if (addr == 0)
      break;

    bp = bread(ip->dev, addr);
    if (bp == 0)
      break;

    int m = min((int)n - tot, (int)(BSIZE - off % BSIZE));

    if (user_dst) {
      if (copyout(myproc()->pgdir, dst, bp->data + (off % BSIZE), m) < 0) {
        brelse(bp);
        break;
      }
    } else {
      memmove((void *)dst, bp->data + (off % BSIZE), m);
    }

    brelse(bp);
    tot += m;
    off += m;
    dst += m;
  }

  // Update access time.  Caller is responsible for calling iupdate() if
  // the change needs to be persisted immediately (avoid calling it in hot
  // paths like readdir).
  ip->atime = now_sec();

  return tot;
}

int
v1_write(struct inode *ip, int user_src, uint64 src, uint64 off, uint n)
{
  if (off > ip->size || off + n < off)
    return -1;
  if ((uint64)off + n > (uint64)MAXFILE * BSIZE)
    return -1;

  uint tot, m;
  struct buf *bp;

  for (tot = 0; tot < n; tot += m, off += m, src += m) {
    bp = bread(ip->dev, bmap(ip, off / BSIZE));
    m  = min(n - tot, BSIZE - off % BSIZE);

    if (user_src) {
      if (copyin(myproc()->pgdir, (addr_t)src, bp->data + (off % BSIZE), (uint64)m) < 0) {
        brelse(bp);
        break;
      }
    } else {
      memmove(bp->data + (off % BSIZE), (void *)src, m);
    }

    log_write(bp);
    brelse(bp);
  }

  if (off > ip->size)
    ip->size = off;

  ip->mtime = now_sec();
  ip->ctime = now_sec();
  iupdate(ip);
  return tot;
}

// ===========================================================================
// Directory HTree
// ===========================================================================

int
namecmp(const char *s, const char *t)
{
  return strncmp(s, t, DIRSIZ);
}

// BKDR hash → bucket index in [16, 16+HASHRANGE).
unsigned int
BKDRHash(char *str)
{
  unsigned int seed = 131;
  unsigned int hash = 0;
  while (*str)
    hash = hash * seed + (*str++);
  return (hash % HASHRANGE) + 16;
}

// ---------------------------------------------------------------------------
// tree_expand: grow directory inode to cover byte offset `want`.
// Allocates backing blocks via bmap and zero-fills any partial-block tail.
// ---------------------------------------------------------------------------

uint
tree_expand(struct inode *dp, uint want)
{
  if (want < dp->size)
    return want;

  // Ensure all blocks up to the one containing `want` exist.
  uint end_block = want / BSIZE;
  for (uint b = dp->size / BSIZE; b <= end_block; b++)
    bmap(dp, b);

  // Zero-fill the old partial tail of the current last block.
  if (dp->size > 0) {
    uint inner = dp->size % BSIZE;
    if (inner != 0) {
      uint fill_end = (dp->size / BSIZE == want / BSIZE)
                    ? (want % BSIZE) + 1
                    : BSIZE;
      struct buf *bp = bread(dp->dev, bmap(dp, (dp->size - 1) / BSIZE));
      memset(bp->data + inner, 0, fill_end - inner);
      log_write(bp);
      brelse(bp);
    }
  }

  dp->size  = want + 1;
  dp->mtime = now_sec();
  dp->ctime = now_sec();
  iupdate(dp);
  return want;
}

// ---------------------------------------------------------------------------
// tree_search: hash-table directory lookup / slot-allocation.
//
// `add == 0` → find entry matching `name`  (de->inum = 0 if not found).
// `add == 1` → find a free slot            (expands the dir if needed).
//
// Primary bucket:  [hash*16*sizeof(dirent), (hash+1)*16*sizeof(dirent))
// Overflow bucket: [OVERFLOW_DIR_START, OVERFLOW_DIR_END)
// ---------------------------------------------------------------------------

void
tree_search(struct inode *dp, char *name, uint *poff, struct dirent *de, uint add)
{
  unsigned int hash_value = BKDRHash(name);

  uint tree_node_off      = (uint)(hash_value * DIRENT_BUCKET_BYTES);
  uint tree_next_node_off = (uint)((hash_value + 1) * DIRENT_BUCKET_BYTES);

  if (tree_node_off > MAXFSIZE) {
    cprintf("tree_search: name=%s hash=%u off=%u MAXFSIZE=%u\n",
            name, hash_value, tree_node_off, MAXFSIZE);
    panic("tree_search: bucket offset exceeds MAXFSIZE");
  }

  // Primary bucket
  for (; tree_node_off < tree_next_node_off; tree_node_off += sizeof(struct dirent)) {
    if (tree_node_off >= dp->size) {
      if (add)
        tree_expand(dp, tree_next_node_off - 1);
      else {
        de->inum = 0;
        return;
      }
    }

    if (readi(dp, 0, (uint64)de, tree_node_off, sizeof(struct dirent)) != sizeof(struct dirent)) {
      cprintf("tree_search: readi failed: name=%s off=%u size=%u\n",
              name, tree_node_off, dp->size);
      panic("tree_search: readi");
    }

    if (!add) {
      if (namecmp(de->name, name) == 0) {
        if (poff) *poff = tree_node_off;
        return;
      }
    } else {
      if (de->inum == 0) {
        if (poff) *poff = tree_node_off;
        return;
      }
    }
  }

  // Overflow bucket
  tree_node_off      = OVERFLOW_DIR_START;
  tree_next_node_off = OVERFLOW_DIR_END;

  for (; tree_node_off < tree_next_node_off; tree_node_off += sizeof(struct dirent)) {
    if (tree_node_off >= dp->size) {
      if (add)
        tree_expand(dp, tree_next_node_off - 1);
      else {
        de->inum = 0;
        return;
      }
    }

    if (readi(dp, 0, (uint64)de, tree_node_off, sizeof(struct dirent)) != sizeof(struct dirent)) {
      cprintf("tree_search: readi failed (overflow): name=%s off=%u size=%u\n",
              name, tree_node_off, dp->size);
      panic("tree_search: readi overflow");
    }

    if (!add) {
      if (namecmp(de->name, name) == 0) {
        if (poff) *poff = tree_node_off;
        return;
      }
    } else {
      if (de->inum == 0) {
        if (poff) *poff = tree_node_off;
        return;
      }
    }
  }

  de->inum = 0;
}

// ===========================================================================
// dirlookup / dirlink / dirlookup_old / build_dir_tree
// ===========================================================================

struct inode *
v1_dirlookup(struct inode *dp, char *name, uint *poff)
{
  uint off, inum;
  struct dirent de;

  if (dp->type != T_DIR)
    panic("dirlookup: not a directory");

  // "." and ".." are stored at fixed offsets 0 and sizeof(dirent).
  if (namecmp(name, ".") == 0 || namecmp(name, "..") == 0) {
    for (off = 0; off < 2 * sizeof(de); off += sizeof(de)) {
      if (readi(dp, 0, (uint64)&de, off, sizeof(de)) != sizeof(de))
        panic("dirlookup: read");
      if (de.inum == 0)
        continue;
      if (namecmp(name, de.name) == 0) {
        if (poff) *poff = off;
        inum = de.inum;
        return iget(dp->dev, inum);
      }
    }
    return 0;
  }

  // The format stores DIRSIZ bytes of a name. A longer one cannot be here,
  // and comparing only its first DIRSIZ would find a different file.
  if (strlen(name) > DIRSIZ)
    return 0;
  tree_search(dp, name, poff, &de, 0);
  if (de.inum)
    return iget(dp->dev, de.inum);
  return 0;
}

int
v1_dirlink(struct inode *dp, char *name, uint64 inum)
{
  uint off = 0;
  struct dirent de;
  struct inode *ip;

  if (strlen(name) > DIRSIZ)
    return -1;                  // see v1_dirlookup
  if ((ip = dirlookup(dp, name, 0)) != 0) {
    iput(ip);
    return -1; // already exists
  }

  tree_search(dp, name, &off, &de, 1);

  strncpy(de.name, name, DIRSIZ);
  de.inum = inum;
  if (writei(dp, 0, (uint64)&de, off, sizeof(de)) != sizeof(de))
    panic("dirlink: writei");

  return 0;
}

// Remove the entry `name`, which dirlookup reported lives at byte offset `off`.
// v1 marks a slot free by zeroing it in place; the hash bucket keeps its size.
static int
v1_dirunlink(struct inode *dp, char *name, uint off)
{
  struct dirent de;
  (void)name;
  memset(&de, 0, sizeof(de));
  if (writei(dp, 0, (uint64)&de, off, sizeof(de)) != sizeof(de))
    return -1;
  return 0;
}

// 1 if the directory holds nothing but "." and "..".
static int
v1_dirempty(struct inode *dp)
{
  struct dirent de;
  for (uint64 off = 2 * sizeof(de); off < dp->size; off += sizeof(de)) {
    if (readi(dp, 0, (uint64)&de, off, sizeof(de)) != sizeof(de))
      return -1;
    if (de.inum != 0)
      return 0;
  }
  return 1;
}

static void
v1_ifree(struct inode *ip)
{
  itrunc(ip);
  ifree(ip->dev, ip->inum);
}

static void
v1_truncate(struct inode *ip, uint64 newsize)
{
  // Truncating to zero has its own path: it releases the indirect blocks
  // wholesale rather than walking back from an offset, and it is by far the
  // common case (every O_TRUNC open).
  if (newsize == 0)
    v1_itrunc_all(ip);
  else
    v1_itrunc_to(ip, newsize);
}

// Read directory entries in getdents64 layout.
//
// Directories are hash-indexed (see tree_search/tree_expand), so dp->size can
// be large and mostly empty even for a handful of real entries -- around 115 KB
// for the ~55 files in /bin. Reading one 16-byte dirent at a time meant a
// bmap()+bread() per slot, nearly all of them empty, which was slow enough to
// look like a hang. A whole block at a time fixes that.
static int
v1_getdents(struct inode *dp, char *out, int cap, uint64 *poff)
{
  struct dirent block[BSIZE / sizeof(struct dirent)];
  int written = 0;

  while (*poff < dp->size) {
    uint64 blockoff = *poff - (*poff % BSIZE);
    int nread = readi(dp, 0, (uint64)block, blockoff, BSIZE);
    if (nread <= 0)
      break;
    int nslots = nread / sizeof(struct dirent);
    int slot   = (int)((*poff - blockoff) / sizeof(struct dirent));

    for (; slot < nslots; slot++) {
      uint64 entry_off = blockoff + (uint64)(slot + 1) * sizeof(struct dirent);
      struct dirent *xde = &block[slot];
      if (xde->inum == 0) { *poff = entry_off; continue; }

      int namelen = 0;
      while (namelen < DIRSIZ && xde->name[namelen]) namelen++;
      int reclen = (8 + 8 + 2 + 1 + namelen + 1 + 7) & ~7;
      if (written + reclen > cap)
        return written;               /* buffer full; resume from *poff */
      *poff = entry_off;

      uint8 dtype = 0;                /* DT_UNKNOWN */
      if (xde->inum == dp->inum) {
        // "." and any other self-referential entry: iget() would hand back this
        // same, already-locked inode, and ilock() on it would self-deadlock on
        // the sleeplock the caller is still holding. We already know it is a
        // directory, so skip the relock entirely.
        dtype = 4;                    /* DT_DIR */
      } else {
        struct inode *eip = iget(dp->dev, xde->inum);
        if (eip) {
          ilock(eip);
          if (eip->type == T_DIR)      dtype = 4;   /* DT_DIR */
          else if (eip->type == T_DEV) dtype = 2;   /* DT_CHR */
          else                         dtype = 8;   /* DT_REG */
          iunlockput(eip);
        }
      }

      char *kptr = out + written;
      uint64 *p64 = (uint64 *)kptr;
      p64[0] = xde->inum;
      p64[1] = *poff;
      *(uint16 *)(kptr + 16) = (uint16)reclen;
      *(uint8  *)(kptr + 18) = dtype;
      memmove(kptr + 19, xde->name, namelen);
      kptr[19 + namelen] = '\0';
      written += reclen;
    }
  }
  return written;
}

// Linear-scan fallback (for readdir iteration over the whole directory).
struct inode *
dirlookup_old(struct inode *dp, char *name, uint *poff)
{
  uint off, inum;
  struct dirent de;

  if (dp->type != T_DIR)
    panic("dirlookup_old: not a directory");

  for (off = 0; off < dp->size; off += sizeof(de)) {
    if (readi(dp, 0, (uint64)&de, off, sizeof(de)) != sizeof(de))
      panic("dirlookup_old: read");
    if (de.inum == 0)
      continue;
    if (namecmp(name, de.name) == 0) {
      if (poff) *poff = off;
      inum = de.inum;
      return iget(dp->dev, inum);
    }
  }
  return 0;
}

void
build_dir_tree(struct inode *dp, uint64 pinum)
{
  dp->size = 2 * sizeof(struct dirent);
  iupdate(dp);

  struct dirent de;

  // Write "." entry at offset 0.
  if (readi(dp, 0, (uint64)&de, 0, sizeof(de)) != sizeof(de))
    panic("build_dir_tree: readi .");
  strncpy(de.name, ".", DIRSIZ);
  de.inum = dp->inum;
  if (writei(dp, 0, (uint64)&de, 0, sizeof(de)) != sizeof(de))
    panic("build_dir_tree: writei .");

  // Write ".." entry at offset sizeof(dirent).
  if (readi(dp, 0, (uint64)&de, sizeof(de), sizeof(de)) != sizeof(de))
    panic("build_dir_tree: readi ..");
  strncpy(de.name, "..", DIRSIZ);
  de.inum = pinum;
  if (writei(dp, 0, (uint64)&de, sizeof(de), sizeof(de)) != sizeof(de))
    panic("build_dir_tree: writei ..");
}

// ===========================================================================
// Path lookup
// ===========================================================================

// Copy the next component of `path` into `name` (VFS_NAME_MAX + 1 bytes) and
// return what follows it, or 0 at the end. A component longer than
// VFS_NAME_MAX sets *toolong and ends the walk.
static char *
skipelem(char *path, char *name, int *toolong)
{
  char *s;
  int  len;

  while (*path == '/')
    path++;
  if (*path == 0)
    return 0;
  s = path;
  while (*path != '/' && *path != 0)
    path++;
  len = path - s;
  if (len > VFS_NAME_MAX) {
    *toolong = 1;
    return 0;
  }
  memmove(name, s, len);
  name[len] = 0;
  while (*path == '/')
    path++;
  return path;
}

// ===========================================================================
// Mount table
//
// A mount is a substitution made during path resolution: when the walk lands on
// a directory that is a mount point, it continues from the mounted filesystem's
// root instead. Nothing on either disk records it, so unmounting is forgetting.
//
// The reverse direction matters as much as the forward one. ".." at the root of
// a mounted filesystem has to leave it, and the mounted filesystem has no idea
// it is mounted anywhere -- SchFS would answer with its own root, which is how
// a subtree becomes a place you can walk into and never out of.
// ===========================================================================

#define NMOUNT 32

static struct {
  struct spinlock lock;
  struct {
    int    used;
    uint   dev;      // device holding the mount point directory
    uint64 inum;     // the mount point directory itself
    uint   fsdev;    // device of the filesystem mounted on it
    uint64 root;     // that filesystem's root inode
    // What userland called it. Resolution never needs these -- it works in
    // (dev, inum) -- but /proc/mounts does, and a path cannot be recovered from
    // an inode without walking the whole tree backwards. So the strings are
    // kept at mount time, when the caller still has them.
    char   path[MAXPATH];
    char   source[MAXPATH];
    char   type[16];
  } m[NMOUNT];
} mtable;

// Copy a string into a fixed field, always NUL-terminating.
static void
mstrcpy(char *dst, const char *src, int cap)
{
  int i = 0;
  if (src)
    for (; i < cap - 1 && src[i]; i++)
      dst[i] = src[i];
  dst[i] = '\0';
}

void
mountinit(void)
{
  initlock(&mtable.lock, "mtable");
}

// Attach `fsdev`:`root` over the directory `dev`:`inum`.
int
mount_add(uint dev, uint64 inum, uint fsdev, uint64 root,
          const char *path, const char *source, const char *type)
{
  // The lookup that found the mount point already crossed any mount on it,
  // so mounting devtmpfs on /dev a second time lands on devtmpfs's own root.
  // That stacked it on itself and listed it twice in /proc/mounts.
  if (dev == fsdev && inum == root)
    return -1;
  acquire(&mtable.lock);
  for (int i = 0; i < NMOUNT; i++)
    if (mtable.m[i].used && mtable.m[i].dev == dev &&
        mtable.m[i].inum == inum) {
      release(&mtable.lock);
      return -1;                        /* already mounted there */
    }
  for (int i = 0; i < NMOUNT; i++)
    if (!mtable.m[i].used) {
      mtable.m[i].used  = 1;
      mtable.m[i].dev   = dev;
      mtable.m[i].inum  = inum;
      mtable.m[i].fsdev = fsdev;
      mtable.m[i].root  = root;
      mstrcpy(mtable.m[i].path,   path,   MAXPATH);
      mstrcpy(mtable.m[i].source, source, MAXPATH);
      mstrcpy(mtable.m[i].type,   type,   16);
      release(&mtable.lock);
      return 0;
    }
  release(&mtable.lock);
  return -1;                            /* table full */
}

// Walk the mount table for /proc/mounts. Returns 0 and fills the caller's
// buffers for entry `i`, or -1 when there is no such entry.
//
// An index rather than a callback because the caller formats into a growing
// string buffer and must not hold this lock while doing it.
int
mount_get(int i, char *path, char *source, char *type, uint *fsdev)
{
  if (i < 0 || i >= NMOUNT)
    return -1;
  acquire(&mtable.lock);
  if (!mtable.m[i].used) {
    release(&mtable.lock);
    return -1;
  }
  mstrcpy(path,   mtable.m[i].path,   MAXPATH);
  mstrcpy(source, mtable.m[i].source, MAXPATH);
  mstrcpy(type,   mtable.m[i].type,   16);
  if (fsdev)
    *fsdev = mtable.m[i].fsdev;
  release(&mtable.lock);
  return 0;
}

// Detach by naming the *mounted* filesystem's root, which is what a path to
// the mount point resolves to once the mount exists. Naming the covered
// directory instead is not possible from userland: the whole point of the
// mount is that the path no longer reaches it.
int
mount_del_by_root(uint fsdev, uint64 root)
{
  acquire(&mtable.lock);
  for (int i = 0; i < NMOUNT; i++)
    if (mtable.m[i].used && mtable.m[i].fsdev == fsdev &&
        mtable.m[i].root == root) {
      mtable.m[i].used = 0;
      release(&mtable.lock);
      return 0;
    }
  release(&mtable.lock);
  return -1;
}

// If dev:inum is a mount point, hand back what is mounted on it.
static int
mount_down(uint dev, uint64 inum, uint *fsdev, uint64 *root)
{
  acquire(&mtable.lock);
  for (int i = 0; i < NMOUNT; i++)
    if (mtable.m[i].used && mtable.m[i].dev == dev &&
        mtable.m[i].inum == inum) {
      *fsdev = mtable.m[i].fsdev;
      *root  = mtable.m[i].root;
      release(&mtable.lock);
      return 1;
    }
  release(&mtable.lock);
  return 0;
}

// If dev:inum is a mounted filesystem's root, hand back the directory it
// covers, so ".." can climb out of it.
static int
mount_up(uint dev, uint64 inum, uint *pdev, uint64 *pinum)
{
  acquire(&mtable.lock);
  for (int i = 0; i < NMOUNT; i++)
    if (mtable.m[i].used && mtable.m[i].fsdev == dev &&
        mtable.m[i].root == inum) {
      *pdev  = mtable.m[i].dev;
      *pinum = mtable.m[i].inum;
      release(&mtable.lock);
      return 1;
    }
  release(&mtable.lock);
  return 0;
}

// Follow every mount stacked on this inode, releasing each one as we go. A
// loop rather than a single step because nothing stops a second filesystem
// being mounted on the first one's root.
static struct inode *
mount_cross(struct inode *ip)
{
  uint   fsdev;
  uint64 root;
  int    hops = 0;

  while (mount_down(ip->dev, ip->inum, &fsdev, &root)) {
    if (++hops > NMOUNT)
      break;                            /* mounts stacked in a cycle */
    struct inode *next = iget(fsdev, root);
    iput(ip);
    ip = next;
  }
  return ip;
}

// Resolves `path` starting from `start` (idup'd internally — caller keeps
// ownership of its own reference) rather than always `myproc()->cwd`. This is
// what makes a real (non-AT_FDCWD) dirfd argument to the *at() syscalls
// actually work: an absolute path still resolves from root either way.
//
// Symlinks are followed the way Linux follows them. One in the middle of the
// path always is; the last component only when `follow` says so, since
// lstat(), readlink(), unlink() and friends want the link itself. A relative
// target is resolved from the directory that holds the link, not from the
// caller's cwd, and more than MAXSYMLINKS of them in one lookup is ELOOP.
//
// The link is read with no directory locked: the walk only ever holds a
// reference to the directory it is in, so re-entering it costs nothing.
#define MAXSYMLINKS 40

static struct inode *
namex_from(struct inode *start, char *path, int nameiparent, char *name,
           int follow, int *err)
{
  struct inode *ip, *next;
  int links = 0, toolong = 0;

  if (err)
    *err = ENOENT;
  if (strnlen(path, VFS_PATH_MAX) >= VFS_PATH_MAX) {
    if (err)
      *err = ENAMETOOLONG;
    return 0;
  }

  // A private copy, because following a link rewrites what is left of it.
  // A page each rather than on the stack: PATH_MAX twice over is the whole of
  // a kernel stack's slack, and a page is exactly PATH_MAX.
  char *safe = kalloc(), *tgt = 0;
  if (!safe) {
    if (err)
      *err = ENOMEM;
    return 0;
  }
  safestrcpy(safe, path, VFS_PATH_MAX);
  char *spath = safe;

  // Root is a SchFS object, not a v1 inode. ops_for(SCHFS_DEV) then routes
  // every operation on it through schfs_ops, which is the whole point: the v1
  // format code in this file is no longer reachable from `/`.
  ip = (*spath == '/') ? iget(SCHFS_DEV, schfs_kroot(SCHFS_DEV)) : idup(start);
  // The root itself can be a mount point, and a relative walk can start on one.
  ip = mount_cross(ip);

  while ((spath = skipelem(spath, name, &toolong)) != 0) {
    // ".." out of a mounted filesystem's root leaves it. The filesystem down
    // there does not know it is mounted anywhere and would answer with its own
    // root, which turns the mount into a subtree you can enter and never
    // leave. Step up to the covered directory first and resolve ".." there.
    if (name[0] == '.' && name[1] == '.' && name[2] == '\0') {
      uint   pdev;
      uint64 pinum;
      if (mount_up(ip->dev, ip->inum, &pdev, &pinum)) {
        struct inode *cover = iget(pdev, pinum);
        iput(ip);
        ip = cover;
      }
    }

    ilock(ip);
    if (ip->type != T_DIR) {
      iunlockput(ip);
      if (err)
        *err = ENOTDIR;
      goto fail;
    }
    if (nameiparent && *spath == '\0') {
      iunlock(ip);
      goto out;
    }
    if ((next = dirlookup(ip, name, 0)) == 0) {
      iunlockput(ip);
      goto fail;
    }
    iunlock(ip);
    next = mount_cross(next);

    ilock(next);
    if (next->type != T_SYMLINK || (*spath == '\0' && (!follow || nameiparent))) {
      iunlock(next);
      iput(ip);
      ip = next;
      continue;
    }

    // A link: put its target in front of what is left of the path and carry
    // on from the directory it was found in, or from root.
    if (!tgt && !(tgt = kalloc())) {
      iunlockput(next);
      iput(ip);
      if (err)
        *err = ENOMEM;
      goto fail;
    }
    int len = (++links > MAXSYMLINKS) ? -1
            : readi(next, 0, (uint64)tgt, 0, VFS_PATH_MAX - 1);
    iunlockput(next);
    int rest = strlen(spath);
    if (len <= 0 || len + 1 + rest >= VFS_PATH_MAX) {
      iput(ip);
      if (err)
        *err = len < 0 ? ELOOP : len == 0 ? ENOENT : ENAMETOOLONG;
      goto fail;
    }
    tgt[len] = '\0';
    if (rest) {
      tgt[len] = '/';
      memmove(tgt + len + 1, spath, rest + 1);
    }
    memmove(safe, tgt, strlen(tgt) + 1);
    spath = safe;
    if (*spath == '/') {
      iput(ip);
      ip = mount_cross(iget(SCHFS_DEV, schfs_kroot(SCHFS_DEV)));
    }
  }

  if (toolong) {
    iput(ip);
    if (err)
      *err = ENAMETOOLONG;
    goto fail;
  }
  if (nameiparent) {
    iput(ip);
    goto fail;
  }
out:
  kfree(safe);
  if (tgt)
    kfree(tgt);
  return ip;
fail:
  kfree(safe);
  if (tgt)
    kfree(tgt);
  return 0;
}

static struct inode *
namex(char *path, int nameiparent, char *name)
{
  return namex_from(myproc()->cwd, path, nameiparent, name, 0, 0);
}

struct inode *
namei(char *path)
{
  char name[VFS_NAME_MAX + 1];
  return namex(path, 0, name);
}

struct inode *
nameiparent(char *path, char *name)
{
  return namex(path, 1, name);
}

// dirfd-relative counterparts, for the *at() syscalls (openat/fstatat/
// unlinkat/...) when given a real directory fd instead of AT_FDCWD.
struct inode *
nameifrom(struct inode *start, char *path)
{
  char name[VFS_NAME_MAX + 1];
  return namex_from(start, path, 0, name, 0, 0);
}

struct inode *
nameiparentfrom(struct inode *start, char *path, char *name)
{
  return namex_from(start, path, 1, name, 0, 0);
}

// The lookup open(), stat(), chdir() and exec() want: a final symlink is
// followed too, and on failure `*err` says why (ENOENT, ENOTDIR, ELOOP,
// ENAMETOOLONG), so a loop is not reported as a missing file.
struct inode *
nameifollow(struct inode *start, char *path, int *err)
{
  char name[VFS_NAME_MAX + 1];
  return namex_from(start ? start : myproc()->cwd, path, 0, name, 1, err);
}

// ===========================================================================
// Public entry points
//
// Thin wrappers over ip->ops. Everything above the filesystem layer calls
// these and never learns which filesystem it is talking to; adding another one
// is a new fs_ops table and nothing else.
// ===========================================================================

void
iupdate(struct inode *ip)
{
  ip->ops->istore(ip);
}

int
readi(struct inode *ip, int user_dst, uint64 dst, uint64 off, uint n)
{
  return ip->ops->read(ip, user_dst, dst, off, n);
}

int
writei(struct inode *ip, int user_src, uint64 src, uint64 off, uint n)
{
  return ip->ops->write(ip, user_src, src, off, n);
}

void
itrunc(struct inode *ip)
{
  ip->ops->truncate(ip, 0);
}

void
itrunc_to(struct inode *ip, uint64 newsize)
{
  ip->ops->truncate(ip, newsize);
}

struct inode *
dirlookup(struct inode *dp, char *name, uint *poff)
{
  return dp->ops->dirlookup(dp, name, poff);
}

int
dirlink(struct inode *dp, char *name, uint64 inum)
{
  return dp->ops->dirlink(dp, name, inum);
}

int
dirunlink(struct inode *dp, char *name, uint off)
{
  return dp->ops->dirunlink(dp, name, off);
}

int
dirempty(struct inode *dp)
{
  return dp->ops->dirempty(dp);
}

int
getdents(struct inode *dp, char *out, int cap, uint64 *poff)
{
  return dp->ops->getdents(dp, out, cap, poff);
}

// The device decides the filesystem, so a caller that wants an object on a
// particular volume names that volume rather than a filesystem.
struct inode *
ialloc(uint dev, short type)
{
  return ops_for(dev)->ialloc(dev, type);
}