vnode.c

Solaris操作系统下的过滤驱动程序, C源码程序.

/*
 * Copyright (c) 1997-2003 Erez Zadok
 * Copyright (c) 2001-2003 Stony Brook University
 * Copyright (c) 1997-2000 Columbia University
 *
 * For specific licensing information, see the COPYING file distributed with
 * this package, or get one from ftp://ftp.filesystems.org/pub/fist/COPYING.
 *
 * This Copyright notice must be kept intact and distributed with all
 * fistgen sources INCLUDING sources generated by fistgen.
 */
/*
 * Copyright (c) 1992 by Sun Microsystems, Inc.
 */

#ifdef HAVE_CONFIG_H
# include <config.h>
#endif /* HAVE_CONFIG_H */
#ifdef FISTGEN
# include "fist_wrapfs.h"
#endif /* FISTGEN */
#include <fist.h>
#include <wrapfs.h>

#if 0
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/errno.h>
#include <sys/vnode.h>
#include <sys/vfs.h>
#include <sys/uio.h>
#include <sys/cred.h>
#include <sys/pathname.h>
#include <sys/dirent.h>
#include <sys/debug.h>
#include <sys/sysmacros.h>
#include <sys/tiuser.h>
#include <sys/cmn_err.h>
#include <sys/stat.h>
#include <sys/mode.h>
#include <rpc/types.h>
#include <rpc/auth.h>
#include <rpc/clnt.h>
#include <sys/fs_subr.h>
#include <sys/mman.h>
#include <sys/vm.h>
#include <vm/as.h>
#include <vm/pvn.h>
#include <vm/seg_vn.h>
#include <vm/seg_kp.h>
#include <vm/seg_map.h>
#include <vm/page.h>

#include <fist.h>
#include <wrapfs.h>
#endif

/*
 *  Vnode ops for fist_wrapfs
 */
static int wrapfs_open(vnode_t **, int, cred_t *);
static int wrapfs_close(vnode_t *, int, int, offset_t, cred_t *);
static int wrapfs_getattr(vnode_t *, vattr_t *, int, cred_t *);
static int wrapfs_access(vnode_t *, int, int, cred_t *);
static int wrapfs_lookup(vnode_t *, char *, vnode_t **,
			     pathname_t *, int, vnode_t *, cred_t *);
static int wrapfs_remove(vnode_t *, char *, cred_t *);
static int wrapfs_rename(vnode_t *, char *, vnode_t *, char *, cred_t *);
static int wrapfs_mkdir(vnode_t *, char *, vattr_t *, vnode_t **, cred_t *);
static int wrapfs_rmdir(vnode_t *, char *, vnode_t *, cred_t *);
static int wrapfs_readdir(vnode_t *, uio_t *, cred_t *, int *);
static int wrapfs_symlink(vnode_t *, char *, vattr_t *, char *, cred_t *);
static int wrapfs_fsync(vnode_t *, int, cred_t *);
static void wrapfs_inactive(vnode_t *, cred_t *);
static void wrapfs_rwlock(vnode_t *, int);
static void wrapfs_rwunlock(vnode_t * vp, int);
static int wrapfs_seek(vnode_t * vp, offset_t, offset_t *);
static int wrapfs_cmp(vnode_t *, vnode_t *);
static int wrapfs_read(vnode_t *, uio_t *, int, cred_t *);
static int wrapfs_write(vnode_t *, uio_t *, int, cred_t *);
static int wrapfs_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *);
static int wrapfs_setfl(vnode_t *, int, int, cred_t *);
static int wrapfs_setattr(vnode_t *, vattr_t *, int, cred_t *);
static int wrapfs_create(vnode_t *, char *, vattr_t *, vcexcl_t, int,
			     vnode_t **, cred_t *, int);
static int wrapfs_link(vnode_t *, vnode_t *, char *, cred_t *);
static int wrapfs_readlink(vnode_t *, uio_t *, cred_t *);
static int wrapfs_fid(vnode_t *, fid_t *);
static int wrapfs_frlock(vnode_t *, int, struct flock64 *, int, offset_t, cred_t *);
static int wrapfs_space(vnode_t *, int, struct flock64 *, int, offset_t, cred_t *);
static int wrapfs_realvp(vnode_t *, vnode_t **);
extern int wrapfs_getpage(vnode_t *, offset_t, size_t, uint_t *, page_t **,
		       size_t, struct seg *, caddr_t, enum seg_rw, cred_t *);
extern int wrapfs_getapage(vnode_t *, u_int, u_int, u_int *, page_t **,
		       u_int, struct seg *, caddr_t, enum seg_rw, cred_t *);
extern int wrapfs_putpage(vnode_t *, offset_t, size_t, int, cred_t *);
extern int wrapfs_putapage(struct vnode *vp, page_t * pp, u_offset_t * offp, size_t * lenp, int flags, struct cred *cr);
extern int wrapfs_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t,
			  uchar_t, uchar_t, uint_t, cred_t *);
extern int wrapfs_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
			     uchar_t, uchar_t, uint_t, cred_t *);
extern int wrapfs_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
			     uint_t, uint_t, uint_t, cred_t *);

static int wrapfs_poll(vnode_t *, short, int, short *, pollhead_t **);
static int wrapfs_dump(vnode_t *, caddr_t, int, int);
static int wrapfs_pathconf(vnode_t *, int, u_long *, cred_t *);
static int wrapfs_pageio(vnode_t *, page_t *, u_offset_t, size_t, int, cred_t *);
static int wrapfs_dumpctl(vnode_t *, int);
static void wrapfs_dispose(vnode_t *, page_t *, int, int, cred_t *);
static int wrapfs_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *);
static int wrapfs_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *);

vnodeops_t wrapfs_vnodeops =
{
  wrapfs_open,			/* open */
  wrapfs_close,			/* close */
  wrapfs_read,			/* read */
  wrapfs_write,			/* write */
  wrapfs_ioctl,			/* ioctl */
  wrapfs_setfl,			/* setfl */
  wrapfs_getattr,		/* getattr */
  wrapfs_setattr,		/* setattr */
  wrapfs_access,		/* access */
  wrapfs_lookup,		/* lookup */
  wrapfs_create,		/* create */
  wrapfs_remove,		/* remove */
  wrapfs_link,			/* link */
  wrapfs_rename,		/* rename */
  wrapfs_mkdir,			/* mkdir */
  wrapfs_rmdir,			/* rmdir */
  wrapfs_readdir,		/* readdir */
  wrapfs_symlink,		/* symlink */
  wrapfs_readlink,		/* readlink */
  wrapfs_fsync,			/* fsync */
  wrapfs_inactive,		/* inactive */
  wrapfs_fid,			/* fid */
  wrapfs_rwlock,		/* rwlock */
  wrapfs_rwunlock,		/* rwunlock */
  wrapfs_seek,			/* seek */
  wrapfs_cmp,			/* cmp */
  wrapfs_frlock,		/* frlock */
  wrapfs_space,			/* space */
  wrapfs_realvp,		/* realvp */
  wrapfs_getpage,		/* getpage */
  wrapfs_putpage,		/* putpage */
  wrapfs_map,			/* map */
  wrapfs_addmap,		/* addmap */
  wrapfs_delmap,		/* delmap */
  wrapfs_poll,			/* poll */
  wrapfs_dump,			/* dump */
  wrapfs_pathconf,		/* pathconf */
  wrapfs_pageio,		/* pageio */
  wrapfs_dumpctl,		/* dumpctl */
  wrapfs_dispose,		/* dispose */
  wrapfs_setsecattr,		/* setsecattr */
  wrapfs_getsecattr		/* getsecattr */
};


/* interpose on an old vnode and return new one */
vnode_t *
wrapfs_interpose(vnode_t * hidden_vp, vfs_t * this_vfsp)
{
  vnode_t *fw_vp;
  fist_wrapfs_node_t *fwnp;
  struct fist_wrapfs_info *fwip = NULL;

  mutex_enter(&vfstofwi(this_vfsp)->fwi_ht_lock);

  /* look if entry exists in HT */
  fw_vp = fist_ht_find_vp(hidden_vp, this_vfsp);
  if (fw_vp) {
    fist_dprint(6, "WRAPFS_INTERPOSE found vp in HT!\n");
    VN_HOLD(fw_vp);		/* maybe VN_RELE(hidden_vp) */
    VN_RELE(hidden_vp);
    goto out;
  }
  /* allocate new vnode */
  fw_vp = kmem_zalloc(sizeof(vnode_t), KM_SLEEP);
  /* XXX: need VN_INIT2 that will reuse lock var of interposed vnode */
  VN_INIT(fw_vp, this_vfsp, hidden_vp->v_type, (dev_t) NULL);

  /* allocate and fill in fist_wrapfs_node_t */
  fwnp = (fist_wrapfs_node_t *)
    kmem_zalloc(sizeof(fist_wrapfs_node_t), KM_SLEEP);
  if (!fwnp) {
    kmem_free(fw_vp, sizeof(vnode_t));
    fw_vp = NULL;
    goto out;
  }

  /* init fwnp */
  fwnp->fwn_mapcnt = 0;		/* no mapped pages */
  mutex_init(&fwnp->fwn_lock, "fist_wrapfs private data lock",
	     MUTEX_DEFAULT, NULL);
  /* store new vnode */
  fwnp->fwn_vnodep = hidden_vp;
  fw_vp->v_data = (caddr_t) fwnp;

  /* operations are fist_wrapfs vnode ops */
  fw_vp->v_op = &wrapfs_vnodeops;

  /* set rest of fields to NULL */
  fw_vp->v_vfsmountedhere = NULL;
  fw_vp->v_filocks = NULL;

  /* don't do this one for now */
  /* fw_vp->v_cv = NULL; */

  /* increment interposed vnodes counter */
  fwip = vfstofwi(this_vfsp);
  fwip->fwi_num_vnodes++;

  /* insert into HT */
  fist_ht_insert_vp(hidden_vp, fw_vp);
  /* VN_RELE(hidden_vp); */

#if 0
  /* hold our vnode */
  VN_HOLD(fw_vp);
#endif

out:
  mutex_exit(&vfstofwi(this_vfsp)->fwi_ht_lock);

  print_location();

  /* finally  return vnode to this newly created one */
  return (fw_vp);
}


static int
wrapfs_open(
	    vnode_t ** vpp,
	    int flag,
	    cred_t * cr
)
{
  int error = EPERM;
  vnode_t *hidden_vp, *new_vp;

  fist_dprint(4, "wrapfs_open vpp %x, flag 0x%x\n", vpp, flag);
  if (flag & FAPPEND) {
    fist_dprint(6, "***WARNING*** file is opened in append-only mode!!!\n");
    flag &= ~FAPPEND;		/* turn off FAPPEND flag */
  }

  new_vp = hidden_vp = vntofwn(*vpp)->fwn_vnodep;
  /* we NEED read access */
  flag |= FREAD;

  /* pass operation to hidden filesystem, and return status */
  error = VOP_OPEN(&hidden_vp, flag, cr);

  /* check if a new vnode was returned (cloned) */
  if (!error && new_vp != hidden_vp) {
    fist_dprint(6, "WRAPFS_OPEN1: hidden_vp->v_count %d\n",
		hidden_vp->v_count);
    /*
     * yes: need to allocate a new fist_wrapfs vnode,
     * initialize it, store interposed vnode in it, and
     * finally return the fist_wrapfs vnode back.
     */
    *vpp = wrapfs_interpose(hidden_vp, (*vpp)->v_vfsp);
    fist_dprint(6, "WRAPFS_OPEN2: vpp->v_count %d, hidden_vp->v_count %d\n",
		(*vpp)->v_count, hidden_vp->v_count);
  }
  print_location();
  return (error);
}


static int
wrapfs_close(
	     vnode_t * vp,
	     int flag,
	     int count,
	     offset_t offset,
	     cred_t * cr
)
{
  int error = EPERM;
  vnode_t *hidden_vp;

  fist_dprint(4, "wrapfs_close vp %x\n", vp);

  hidden_vp = vntofwn(vp)->fwn_vnodep;

  fist_dprint(6, "WRAPFS_CLOSE1: vp->v_count %d, hidden_vp->v_count %d\n",
	      vp->v_count, hidden_vp->v_count);

  /* pass operation to hidden filesystem, and return status */
  error = VOP_CLOSE(hidden_vp, flag, count, offset, cr);

  fist_dprint(6, "WRAPFS_CLOSE2: vp->v_count %d, hidden_vp->v_count %d\n",
	      vp->v_count, hidden_vp->v_count);
  print_location();
  return (error);
}


static int
wrapfs_read(
	    vnode_t * vp,
	    uio_t * uiop,
	    int ioflag,
	    cred_t * cr
)
{
  int error = EPERM;
  vnode_t *hidden_vp;
  uio_t temp_uio;
  iovec_t *temp_iovec;
  caddr_t current_base;
  int i, bytes_read;
  int num_pages, resid;
  long long start_loffset, end_loffset;
  long long cleartext_start_loffset, cleartext_end_loffset, current_loffset;

  fist_dprint(4, "wrapfs_read vp %x\n", vp);

#ifdef FIST_DEBUG
  fist_print_uios("wrapfs_read", uiop);
#endif /* FIST_DEBUG */

  cleartext_start_loffset = uiop->uio_loffset;
  cleartext_end_loffset = uiop->uio_loffset + uiop->uio_resid;
  start_loffset = cleartext_start_loffset & ~(PAGESIZE - 1);
  end_loffset = cleartext_end_loffset & ~(PAGESIZE - 1);
  /*
   * if not multiple of PAGESIZE, then the above formula loses one page.
   * adjust for it
   */
  if (cleartext_end_loffset > end_loffset)
    end_loffset += PAGESIZE;
  resid = end_loffset - start_loffset;
  num_pages = resid >> PAGESHIFT;

  fist_dprint(6,
	      "READ: so=%d eo=%d cs=%d es=%d res=%d np=%d ps=%d\n",
	      (int) start_loffset,
	      (int) end_loffset,
	      (int) cleartext_start_loffset,
	      (int) cleartext_end_loffset,
	      resid,
	      num_pages,
	      PAGESIZE);

  temp_iovec = kmem_zalloc(num_pages * sizeof(iovec_t), KM_SLEEP);
  for (i = 0; i < num_pages; i++) {
    temp_iovec[i].iov_len = PAGESIZE;
    temp_iovec[i].iov_base = kmem_zalloc(PAGESIZE, KM_SLEEP);
    fist_dprint(6, "READ allocated %d address 0x%x\n",
		i, temp_iovec[i].iov_base);
  }

  temp_uio.uio_iov = temp_iovec;
  temp_uio.uio_iovcnt = num_pages;
  temp_uio.uio_loffset = start_loffset;
  temp_uio.uio_segflg = UIO_SYSSPACE;
  temp_uio.uio_fmode = uiop->uio_fmode;
  temp_uio.uio_llimit = uiop->uio_llimit;
  temp_uio.uio_resid = resid;

  hidden_vp = vntofwn(vp)->fwn_vnodep;

  /*
   * pass operation to hidden filesystem, and return status
   */

  error = VOP_READ(hidden_vp, &temp_uio, ioflag, cr);

  if (error) {
    fist_dprint(4, "VOP_READ in read returned error - not good\n");
    /* XXX to be checked */
    goto out_free;
  }

  current_loffset = start_loffset;
  for (i = 0; i < num_pages; i++) {
    bytes_read = PAGESIZE - temp_iovec[i].iov_len;
    if (bytes_read == 0)
      break;

    temp_iovec[i].iov_base -= bytes_read;
    current_base = temp_iovec[i].iov_base;

    /* decodes the page/block */
    wrapfs_decode_block(current_base, current_base, bytes_read, vp, vp->v_vfsp);

    /*
     * save the original size, for kmem_free.
     * no need for it w/ wrapfs; size is always PAGESIZE, hence this line
     * is commented out:
     *		temp_iovec[i].iov_len = uiop->uio_iov[i].iov_len;
     */
    /* treat first and last iovec separately, not all data in them is needed */
    if (current_loffset + bytes_read > cleartext_end_loffset) {
      bytes_read = cleartext_end_loffset - current_loffset;
    }
    if (i == 0) {
      bytes_read -= cleartext_start_loffset - start_loffset;
      current_loffset += cleartext_start_loffset - start_loffset;
      current_base += cleartext_start_loffset - start_loffset;
    }
    if ((error = uiomove(current_base, bytes_read, UIO_READ, uiop)))
      /*
       * XXX: we have to see the exact semantics of returning with an
       * EFAULT from read
       */
      break;
    current_loffset += bytes_read;
  }

out_free:
  for (i = 0; i < num_pages; i++) {
    fist_dprint(6, "READ freeing %d address 0x%x\n",
		i, temp_iovec[i].iov_base);
    kmem_free(temp_iovec[i].iov_base, PAGESIZE);
  }
  kmem_free(temp_iovec, num_pages * sizeof(iovec_t));

#ifdef FIST_DEBUG
  fist_print_uios("wrapfs_read (END)", uiop);
#endif /* FIST_DEBUG */

  print_location();
  return (error);
}


static int
wrapfs_write(
	     vnode_t * vp,
	     uio_t * uiop,
	     int ioflag,
	     cred_t * cr
)
{
  int error = EPERM;
  vnode_t *hidden_vp;
  vattr_t va;
  uio_t temp_uio;
  iovec_t *temp_iovec;
  iovec_t *free_iovec;		/* for freeing allocated memory */
  int i;
  caddr_t current_base;
  int resid, bytes_read, num_pages, first_page_bytes, real_first_page;
  long long start_loffset, end_loffset, real_start_loffset;
  long long cleartext_start_loffset, cleartext_end_loffset, current_loffset;
  int hidden_ioflag = (ioflag & ~FAPPEND);

  fist_dprint(4, "wrapfs_write vp %x, ioflag 0x5x\n", vp, ioflag);

#ifdef FIST_DEBUG
  fist_print_uios("wrapfs_write (START)", uiop);
#endif /* FIST_DEBUG */

  hidden_vp = vntofwn(vp)->fwn_vnodep;

  /* we don't want anybody to do updates while we write, so lock the vnode */
  mutex_enter(&vp->v_lock);

  /* get the attributes, length is necessary for correct updates */
  va.va_mask = AT_SIZE;
  if ((error = VOP_GETATTR(hidden_vp, &va, 0, cr))) {
    fist_dprint(4, "VOP_GETATTR returned error - not good\n");
    /* XXX to be checked */