dynamic allocation of vnodes
[unix-history] / usr / src / sys / kern / vfs_subr.c
index 80a41d1..4720466 100644 (file)
@@ -14,7 +14,7 @@
  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
  *
  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
  *
- *     @(#)vfs_subr.c  7.6 (Berkeley) %G%
+ *     @(#)vfs_subr.c  7.38 (Berkeley) %G%
  */
 
 /*
  */
 
 /*
 #include "mount.h"
 #include "time.h"
 #include "vnode.h"
 #include "mount.h"
 #include "time.h"
 #include "vnode.h"
+#include "specdev.h"
 #include "namei.h"
 #include "ucred.h"
 #include "errno.h"
 #include "namei.h"
 #include "ucred.h"
 #include "errno.h"
-
-/*
- * Add a new mount point to the list of mounted filesystems.
- * Lock the filesystem so that namei will not cross into the
- * the tree below the covered vnode.
- */
-vfs_add(mountedvp, mp, flags)
-       register struct vnode *mountedvp;
-       register struct mount *mp;
-       int flags;
-{
-       register int error;
-
-       error = vfs_lock(mp);
-       if (error)
-               return (error);
-       if (mountedvp == (struct vnode *)0) {
-               /*
-                * We are mounting the root filesystem.
-                */
-               rootfs = mp;
-               mp->m_next = mp;
-               mp->m_prev = mp;
-       } else {
-               if (mountedvp->v_mountedhere != (struct mount *)0) {
-                       vfs_unlock(mp);
-                       return(EBUSY);
-               }
-               /*
-                * Put the new filesystem on the mount list after root.
-                */
-               mp->m_next = rootfs->m_next;
-               mp->m_prev = rootfs;
-               rootfs->m_next = mp;
-               mp->m_next->m_prev = mp;
-               mountedvp->v_mountedhere = mp;
-       }
-       mp->m_vnodecovered = mountedvp;
-       if (flags & M_RDONLY) {
-               mp->m_flag |= M_RDONLY;
-       } else {
-               mp->m_flag &= ~M_RDONLY;
-       }
-       if (flags & M_NOSUID) {
-               mp->m_flag |= M_NOSUID;
-       } else {
-               mp->m_flag &= ~M_NOSUID;
-       }
-       return (0);
-}
+#include "malloc.h"
 
 /*
  * Remove a mount point from the list of mounted filesystems.
 
 /*
  * Remove a mount point from the list of mounted filesystems.
@@ -141,8 +93,8 @@ getvfs(fsid)
 
        mp = rootfs;
        do {
 
        mp = rootfs;
        do {
-               if (mp->m_fsid.val[0] == fsid->val[0] &&
-                   mp->m_fsid.val[1] == fsid->val[1]) {
+               if (mp->m_stat.f_fsid.val[0] == fsid->val[0] &&
+                   mp->m_stat.f_fsid.val[1] == fsid->val[1]) {
                        return (mp);
                }
                mp = mp->m_next;
                        return (mp);
                }
                mp = mp->m_next;
@@ -160,8 +112,8 @@ void vattr_null(vap)
        vap->va_type = VNON;
        vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
                vap->va_fsid = vap->va_fileid = vap->va_size =
        vap->va_type = VNON;
        vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
                vap->va_fsid = vap->va_fileid = vap->va_size =
-               vap->va_size1 = vap->va_blocksize = vap->va_rdev =
-               vap->va_bytes = vap->va_bytes1 =
+               vap->va_size_rsv = vap->va_blocksize = vap->va_rdev =
+               vap->va_bytes = vap->va_bytes_rsv =
                vap->va_atime.tv_sec = vap->va_atime.tv_usec =
                vap->va_mtime.tv_sec = vap->va_mtime.tv_usec =
                vap->va_ctime.tv_sec = vap->va_ctime.tv_usec =
                vap->va_atime.tv_sec = vap->va_atime.tv_usec =
                vap->va_mtime.tv_sec = vap->va_mtime.tv_usec =
                vap->va_ctime.tv_sec = vap->va_ctime.tv_usec =
@@ -180,7 +132,7 @@ ndinit(ndp)
        ndp->ni_iovcnt = 1;
        ndp->ni_base = (caddr_t)&ndp->ni_dent;
        ndp->ni_rw = UIO_WRITE;
        ndp->ni_iovcnt = 1;
        ndp->ni_base = (caddr_t)&ndp->ni_dent;
        ndp->ni_rw = UIO_WRITE;
-       ndp->ni_segflg = UIO_SYSSPACE;
+       ndp->ni_uioseg = UIO_SYSSPACE;
 }
 
 /*
 }
 
 /*
@@ -212,3 +164,653 @@ ndrele(ndp)
                vrele(ndp->ni_rdir);
        crfree(ndp->ni_cred);
 }
                vrele(ndp->ni_rdir);
        crfree(ndp->ni_cred);
 }
+
+/*
+ * Routines having to do with the management of the vnode table.
+ */
+struct vnode *vfreeh, **vfreet;
+extern struct vnodeops dead_vnodeops, spec_vnodeops;
+extern void vclean();
+long numvnodes;
+
+/*
+ * Initialize the vnode structures and initialize each file system type.
+ */
+vfsinit()
+{
+       struct vfsops **vfsp;
+
+       /*
+        * Initialize the vnode name cache
+        */
+       nchinit();
+       /*
+        * Initialize each file system type.
+        */
+       for (vfsp = &vfssw[0]; vfsp <= &vfssw[MOUNT_MAXTYPE]; vfsp++) {
+               if (*vfsp == NULL)
+                       continue;
+               (*(*vfsp)->vfs_init)();
+       }
+}
+
+/*
+ * Return the next vnode from the free list.
+ */
+getnewvnode(tag, mp, vops, vpp)
+       enum vtagtype tag;
+       struct mount *mp;
+       struct vnodeops *vops;
+       struct vnode **vpp;
+{
+       register struct vnode *vp, *vq;
+
+       if (numvnodes < desiredvnodes) {
+               vp = (struct vnode *)malloc(sizeof *vp, M_VNODE, M_WAITOK);
+               bzero((char *)vp, sizeof *vp);
+               numvnodes++;
+       } else {
+               if ((vp = vfreeh) == NULL) {
+                       tablefull("vnode");
+                       *vpp = 0;
+                       return (ENFILE);
+               }
+               if (vp->v_usecount)
+                       panic("free vnode isn't");
+               if (vq = vp->v_freef)
+                       vq->v_freeb = &vfreeh;
+               else
+                       vfreet = &vfreeh;
+               vfreeh = vq;
+               vp->v_freef = NULL;
+               vp->v_freeb = NULL;
+               if (vp->v_type != VBAD)
+                       vgone(vp);
+               vp->v_flag = 0;
+               vp->v_shlockc = 0;
+               vp->v_exlockc = 0;
+               vp->v_lastr = 0;
+               vp->v_socket = 0;
+       }
+       vp->v_type = VNON;
+       cache_purge(vp);
+       vp->v_tag = tag;
+       vp->v_op = vops;
+       insmntque(vp, mp);
+       VREF(vp);
+       *vpp = vp;
+       return (0);
+}
+
+/*
+ * Move a vnode from one mount queue to another.
+ */
+insmntque(vp, mp)
+       register struct vnode *vp;
+       register struct mount *mp;
+{
+       struct vnode *vq;
+
+       /*
+        * Delete from old mount point vnode list, if on one.
+        */
+       if (vp->v_mountb) {
+               if (vq = vp->v_mountf)
+                       vq->v_mountb = vp->v_mountb;
+               *vp->v_mountb = vq;
+       }
+       /*
+        * Insert into list of vnodes for the new mount point, if available.
+        */
+       vp->v_mount = mp;
+       if (mp == NULL) {
+               vp->v_mountf = NULL;
+               vp->v_mountb = NULL;
+               return;
+       }
+       if (mp->m_mounth) {
+               vp->v_mountf = mp->m_mounth;
+               vp->v_mountb = &mp->m_mounth;
+               mp->m_mounth->v_mountb = &vp->v_mountf;
+               mp->m_mounth = vp;
+       } else {
+               mp->m_mounth = vp;
+               vp->v_mountb = &mp->m_mounth;
+               vp->v_mountf = NULL;
+       }
+}
+
+/*
+ * Create a vnode for a block device.
+ * Used for root filesystem, argdev, and swap areas.
+ * Also used for memory file system special devices.
+ */
+bdevvp(dev, vpp)
+       dev_t dev;
+       struct vnode **vpp;
+{
+       register struct vnode *vp;
+       struct vnode *nvp;
+       int error;
+
+       error = getnewvnode(VT_NON, (struct mount *)0, &spec_vnodeops, &nvp);
+       if (error) {
+               *vpp = 0;
+               return (error);
+       }
+       vp = nvp;
+       vp->v_type = VBLK;
+       if (nvp = checkalias(vp, dev, (struct mount *)0)) {
+               vput(vp);
+               vp = nvp;
+       }
+       *vpp = vp;
+       return (0);
+}
+
+/*
+ * Check to see if the new vnode represents a special device
+ * for which we already have a vnode (either because of
+ * bdevvp() or because of a different vnode representing
+ * the same block device). If such an alias exists, deallocate
+ * the existing contents and return the aliased vnode. The
+ * caller is responsible for filling it with its new contents.
+ */
+struct vnode *
+checkalias(nvp, nvp_rdev, mp)
+       register struct vnode *nvp;
+       dev_t nvp_rdev;
+       struct mount *mp;
+{
+       register struct vnode *vp;
+       struct vnode **vpp;
+
+       if (nvp->v_type != VBLK && nvp->v_type != VCHR)
+               return ((struct vnode *)0);
+
+       vpp = &speclisth[SPECHASH(nvp_rdev)];
+loop:
+       for (vp = *vpp; vp; vp = vp->v_specnext) {
+               if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
+                       continue;
+               /*
+                * Alias, but not in use, so flush it out.
+                */
+               if (vp->v_usecount == 0) {
+                       vgone(vp);
+                       goto loop;
+               }
+               if (vget(vp))
+                       goto loop;
+               break;
+       }
+       if (vp == NULL || vp->v_tag != VT_NON) {
+               MALLOC(nvp->v_specinfo, struct specinfo *,
+                       sizeof(struct specinfo), M_VNODE, M_WAITOK);
+               nvp->v_rdev = nvp_rdev;
+               nvp->v_hashchain = vpp;
+               nvp->v_specnext = *vpp;
+               *vpp = nvp;
+               if (vp != NULL) {
+                       nvp->v_flag |= VALIASED;
+                       vp->v_flag |= VALIASED;
+                       vput(vp);
+               }
+               return ((struct vnode *)0);
+       }
+       VOP_UNLOCK(vp);
+       vclean(vp, 0);
+       vp->v_op = nvp->v_op;
+       vp->v_tag = nvp->v_tag;
+       nvp->v_type = VNON;
+       insmntque(vp, mp);
+       return (vp);
+}
+
+/*
+ * Grab a particular vnode from the free list, increment its
+ * reference count and lock it. The vnode lock bit is set the
+ * vnode is being eliminated in vgone. The process is awakened
+ * when the transition is completed, and an error returned to
+ * indicate that the vnode is no longer usable (possibly having
+ * been changed to a new file system type).
+ */
+vget(vp)
+       register struct vnode *vp;
+{
+       register struct vnode *vq;
+
+       if (vp->v_flag & VXLOCK) {
+               vp->v_flag |= VXWANT;
+               sleep((caddr_t)vp, PINOD);
+               return (1);
+       }
+       if (vp->v_usecount == 0) {
+               if (vq = vp->v_freef)
+                       vq->v_freeb = vp->v_freeb;
+               else
+                       vfreet = vp->v_freeb;
+               *vp->v_freeb = vq;
+               vp->v_freef = NULL;
+               vp->v_freeb = NULL;
+       }
+       VREF(vp);
+       VOP_LOCK(vp);
+       return (0);
+}
+
+/*
+ * Vnode reference, just increment the count
+ */
+void vref(vp)
+       struct vnode *vp;
+{
+
+       vp->v_usecount++;
+}
+
+/*
+ * vput(), just unlock and vrele()
+ */
+void vput(vp)
+       register struct vnode *vp;
+{
+       VOP_UNLOCK(vp);
+       vrele(vp);
+}
+
+/*
+ * Vnode release.
+ * If count drops to zero, call inactive routine and return to freelist.
+ */
+void vrele(vp)
+       register struct vnode *vp;
+{
+
+       if (vp == NULL)
+               panic("vrele: null vp");
+       vp->v_usecount--;
+       if (vp->v_usecount < 0)
+               vprint("vrele: bad ref count", vp);
+       if (vp->v_usecount > 0)
+               return;
+       if (vfreeh == (struct vnode *)0) {
+               /*
+                * insert into empty list
+                */
+               vfreeh = vp;
+               vp->v_freeb = &vfreeh;
+       } else {
+               /*
+                * insert at tail of list
+                */
+               *vfreet = vp;
+               vp->v_freeb = vfreet;
+       }
+       vp->v_freef = NULL;
+       vfreet = &vp->v_freef;
+       VOP_INACTIVE(vp);
+}
+
+/*
+ * Page or buffer structure gets a reference.
+ */
+vhold(vp)
+       register struct vnode *vp;
+{
+
+       vp->v_holdcnt++;
+}
+
+/*
+ * Page or buffer structure frees a reference.
+ */
+holdrele(vp)
+       register struct vnode *vp;
+{
+
+       if (vp->v_holdcnt <= 0)
+               panic("holdrele: holdcnt");
+       vp->v_holdcnt--;
+}
+
+/*
+ * Remove any vnodes in the vnode table belonging to mount point mp.
+ *
+ * If MNT_NOFORCE is specified, there should not be any active ones,
+ * return error if any are found (nb: this is a user error, not a
+ * system error). If MNT_FORCE is specified, detach any active vnodes
+ * that are found.
+ */
+int busyprt = 0;       /* patch to print out busy vnodes */
+
+vflush(mp, skipvp, flags)
+       struct mount *mp;
+       struct vnode *skipvp;
+       int flags;
+{
+       register struct vnode *vp, *nvp;
+       int busy = 0;
+
+       for (vp = mp->m_mounth; vp; vp = nvp) {
+               nvp = vp->v_mountf;
+               /*
+                * Skip over a selected vnode.
+                * Used by ufs to skip over the quota structure inode.
+                */
+               if (vp == skipvp)
+                       continue;
+               /*
+                * With v_usecount == 0, all we need to do is clear
+                * out the vnode data structures and we are done.
+                */
+               if (vp->v_usecount == 0) {
+                       vgone(vp);
+                       continue;
+               }
+               /*
+                * For block or character devices, revert to an
+                * anonymous device. For all other files, just kill them.
+                */
+               if (flags & MNT_FORCE) {
+                       if (vp->v_type != VBLK && vp->v_type != VCHR) {
+                               vgone(vp);
+                       } else {
+                               vclean(vp, 0);
+                               vp->v_op = &spec_vnodeops;
+                               insmntque(vp, (struct mount *)0);
+                       }
+                       continue;
+               }
+               if (busyprt)
+                       vprint("vflush: busy vnode", vp);
+               busy++;
+       }
+       if (busy)
+               return (EBUSY);
+       return (0);
+}
+
+/*
+ * Disassociate the underlying file system from a vnode.
+ */
+void vclean(vp, doclose)
+       register struct vnode *vp;
+       long doclose;
+{
+       struct vnodeops *origops;
+       int active;
+
+       /*
+        * Check to see if the vnode is in use.
+        * If so we have to reference it before we clean it out
+        * so that its count cannot fall to zero and generate a
+        * race against ourselves to recycle it.
+        */
+       if (active = vp->v_usecount)
+               VREF(vp);
+       /*
+        * Prevent the vnode from being recycled or
+        * brought into use while we clean it out.
+        */
+       if (vp->v_flag & VXLOCK)
+               panic("vclean: deadlock");
+       vp->v_flag |= VXLOCK;
+       /*
+        * Even if the count is zero, the VOP_INACTIVE routine may still
+        * have the object locked while it cleans it out. The VOP_LOCK
+        * ensures that the VOP_INACTIVE routine is done with its work.
+        * For active vnodes, it ensures that no other activity can
+        * occur while the buffer list is being cleaned out.
+        */
+       VOP_LOCK(vp);
+       if (doclose)
+               vinvalbuf(vp, 1);
+       /*
+        * Prevent any further operations on the vnode from
+        * being passed through to the old file system.
+        */
+       origops = vp->v_op;
+       vp->v_op = &dead_vnodeops;
+       vp->v_tag = VT_NON;
+       /*
+        * If purging an active vnode, it must be unlocked, closed,
+        * and deactivated before being reclaimed.
+        */
+       (*(origops->vn_unlock))(vp);
+       if (active) {
+               if (doclose)
+                       (*(origops->vn_close))(vp, 0, NOCRED);
+               (*(origops->vn_inactive))(vp);
+       }
+       /*
+        * Reclaim the vnode.
+        */
+       if ((*(origops->vn_reclaim))(vp))
+               panic("vclean: cannot reclaim");
+       if (active)
+               vrele(vp);
+       /*
+        * Done with purge, notify sleepers in vget of the grim news.
+        */
+       vp->v_flag &= ~VXLOCK;
+       if (vp->v_flag & VXWANT) {
+               vp->v_flag &= ~VXWANT;
+               wakeup((caddr_t)vp);
+       }
+}
+
+/*
+ * Eliminate all activity associated with  the requested vnode
+ * and with all vnodes aliased to the requested vnode.
+ */
+void vgoneall(vp)
+       register struct vnode *vp;
+{
+       register struct vnode *vq;
+
+       if (vp->v_flag & VALIASED) {
+               /*
+                * If a vgone (or vclean) is already in progress,
+                * wait until it is done and return.
+                */
+               if (vp->v_flag & VXLOCK) {
+                       vp->v_flag |= VXWANT;
+                       sleep((caddr_t)vp, PINOD);
+                       return;
+               }
+               /*
+                * Ensure that vp will not be vgone'd while we
+                * are eliminating its aliases.
+                */
+               vp->v_flag |= VXLOCK;
+               while (vp->v_flag & VALIASED) {
+                       for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
+                               if (vq->v_rdev != vp->v_rdev ||
+                                   vq->v_type != vp->v_type || vp == vq)
+                                       continue;
+                               vgone(vq);
+                               break;
+                       }
+               }
+               /*
+                * Remove the lock so that vgone below will
+                * really eliminate the vnode after which time
+                * vgone will awaken any sleepers.
+                */
+               vp->v_flag &= ~VXLOCK;
+       }
+       vgone(vp);
+}
+
+/*
+ * Eliminate all activity associated with a vnode
+ * in preparation for reuse.
+ */
+void vgone(vp)
+       register struct vnode *vp;
+{
+       register struct vnode *vq;
+       struct vnode *vx;
+       long count;
+
+       /*
+        * If a vgone (or vclean) is already in progress,
+        * wait until it is done and return.
+        */
+       if (vp->v_flag & VXLOCK) {
+               vp->v_flag |= VXWANT;
+               sleep((caddr_t)vp, PINOD);
+               return;
+       }
+       /*
+        * Clean out the filesystem specific data.
+        */
+       vclean(vp, 1);
+       /*
+        * Delete from old mount point vnode list, if on one.
+        */
+       if (vp->v_mountb) {
+               if (vq = vp->v_mountf)
+                       vq->v_mountb = vp->v_mountb;
+               *vp->v_mountb = vq;
+               vp->v_mountf = NULL;
+               vp->v_mountb = NULL;
+       }
+       /*
+        * If special device, remove it from special device alias list.
+        */
+       if (vp->v_type == VBLK || vp->v_type == VCHR) {
+               if (*vp->v_hashchain == vp) {
+                       *vp->v_hashchain = vp->v_specnext;
+               } else {
+                       for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
+                               if (vq->v_specnext != vp)
+                                       continue;
+                               vq->v_specnext = vp->v_specnext;
+                               break;
+                       }
+                       if (vq == NULL)
+                               panic("missing bdev");
+               }
+               if (vp->v_flag & VALIASED) {
+                       count = 0;
+                       for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
+                               if (vq->v_rdev != vp->v_rdev ||
+                                   vq->v_type != vp->v_type)
+                                       continue;
+                               count++;
+                               vx = vq;
+                       }
+                       if (count == 0)
+                               panic("missing alias");
+                       if (count == 1)
+                               vx->v_flag &= ~VALIASED;
+                       vp->v_flag &= ~VALIASED;
+               }
+               FREE(vp->v_specinfo, M_VNODE);
+               vp->v_specinfo = NULL;
+       }
+       /*
+        * If it is on the freelist, move it to the head of the list.
+        */
+       if (vp->v_freeb) {
+               if (vq = vp->v_freef)
+                       vq->v_freeb = vp->v_freeb;
+               else
+                       vfreet = vp->v_freeb;
+               *vp->v_freeb = vq;
+               vp->v_freef = vfreeh;
+               vp->v_freeb = &vfreeh;
+               vfreeh->v_freeb = &vp->v_freef;
+               vfreeh = vp;
+       }
+       vp->v_type = VBAD;
+}
+
+/*
+ * Lookup a vnode by device number.
+ */
+vfinddev(dev, type, vpp)
+       dev_t dev;
+       enum vtype type;
+       struct vnode **vpp;
+{
+       register struct vnode *vp;
+
+       for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
+               if (dev != vp->v_rdev || type != vp->v_type)
+                       continue;
+               *vpp = vp;
+               return (0);
+       }
+       return (1);
+}
+
+/*
+ * Calculate the total number of references to a special device.
+ */
+vcount(vp)
+       register struct vnode *vp;
+{
+       register struct vnode *vq;
+       int count;
+
+       if ((vp->v_flag & VALIASED) == 0)
+               return (vp->v_usecount);
+loop:
+       for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
+               if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
+                       continue;
+               /*
+                * Alias, but not in use, so flush it out.
+                */
+               if (vq->v_usecount == 0) {
+                       vgone(vq);
+                       goto loop;
+               }
+               count += vq->v_usecount;
+       }
+       return (count);
+}
+
+/*
+ * Print out a description of a vnode.
+ */
+static char *typename[] =
+   { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
+
+vprint(label, vp)
+       char *label;
+       register struct vnode *vp;
+{
+       char buf[64];
+
+       if (label != NULL)
+               printf("%s: ", label);
+       printf("type %s, usecount %d, refcount %d,", typename[vp->v_type],
+               vp->v_usecount, vp->v_holdcnt);
+       buf[0] = '\0';
+       if (vp->v_flag & VROOT)
+               strcat(buf, "|VROOT");
+       if (vp->v_flag & VTEXT)
+               strcat(buf, "|VTEXT");
+       if (vp->v_flag & VXLOCK)
+               strcat(buf, "|VXLOCK");
+       if (vp->v_flag & VXWANT)
+               strcat(buf, "|VXWANT");
+       if (vp->v_flag & VEXLOCK)
+               strcat(buf, "|VEXLOCK");
+       if (vp->v_flag & VSHLOCK)
+               strcat(buf, "|VSHLOCK");
+       if (vp->v_flag & VLWAIT)
+               strcat(buf, "|VLWAIT");
+       if (vp->v_flag & VALIASED)
+               strcat(buf, "|VALIASED");
+       if (vp->v_flag & VBWAIT)
+               strcat(buf, "|VBWAIT");
+       if (buf[0] != '\0')
+               printf(" flags (%s)", &buf[1]);
+       printf("\n\t");
+       VOP_PRINT(vp);
+}