| 1 | /* |
| 2 | * Copyright (c) 1994 Jan-Simon Pendry |
| 3 | * Copyright (c) 1994 |
| 4 | * The Regents of the University of California. All rights reserved. |
| 5 | * |
| 6 | * This code is derived from software contributed to Berkeley by |
| 7 | * Jan-Simon Pendry. |
| 8 | * |
| 9 | * %sccs.include.redist.c% |
| 10 | * |
| 11 | * @(#)union_subr.c 1.5 (Berkeley) %G% |
| 12 | */ |
| 13 | |
| 14 | #include <sys/param.h> |
| 15 | #include <sys/systm.h> |
| 16 | #include <sys/time.h> |
| 17 | #include <sys/kernel.h> |
| 18 | #include <sys/vnode.h> |
| 19 | #include <sys/namei.h> |
| 20 | #include <sys/malloc.h> |
| 21 | #include <sys/file.h> |
| 22 | #include "union.h" /*<miscfs/union/union.h>*/ |
| 23 | |
| 24 | #ifdef DIAGNOSTIC |
| 25 | #include <sys/proc.h> |
| 26 | #endif |
| 27 | |
| 28 | static struct union_node *unhead; |
| 29 | static int unvplock; |
| 30 | |
| 31 | int |
| 32 | union_init() |
| 33 | { |
| 34 | |
| 35 | unhead = 0; |
| 36 | unvplock = 0; |
| 37 | } |
| 38 | |
| 39 | /* |
| 40 | * allocate a union_node/vnode pair. the vnode is |
| 41 | * referenced and locked. the new vnode is returned |
| 42 | * via (vpp). (mp) is the mountpoint of the union filesystem, |
| 43 | * (dvp) is the parent directory where the upper layer object |
| 44 | * should exist (but doesn't) and (cnp) is the componentname |
| 45 | * information which is partially copied to allow the upper |
| 46 | * layer object to be created at a later time. (uppervp) |
| 47 | * and (lowervp) reference the upper and lower layer objects |
| 48 | * being mapped. either, but not both, can be nil. |
| 49 | * |
| 50 | * all union_nodes are maintained on a singly-linked |
| 51 | * list. new nodes are only allocated when they cannot |
| 52 | * be found on this list. entries on the list are |
| 53 | * removed when the vfs reclaim entry is called. |
| 54 | * |
| 55 | * a single lock is kept for the entire list. this is |
| 56 | * needed because the getnewvnode() function can block |
| 57 | * waiting for a vnode to become free, in which case there |
| 58 | * may be more than one process trying to get the same |
| 59 | * vnode. this lock is only taken if we are going to |
| 60 | * call getnewvnode, since the kernel itself is single-threaded. |
| 61 | * |
| 62 | * if an entry is found on the list, then call vget() to |
| 63 | * take a reference. this is done because there may be |
| 64 | * zero references to it and so it needs to removed from |
| 65 | * the vnode free list. |
| 66 | */ |
| 67 | int |
| 68 | union_allocvp(vpp, mp, undvp, dvp, cnp, uppervp, lowervp) |
| 69 | struct vnode **vpp; |
| 70 | struct mount *mp; |
| 71 | struct vnode *undvp; |
| 72 | struct vnode *dvp; /* may be null */ |
| 73 | struct componentname *cnp; /* may be null */ |
| 74 | struct vnode *uppervp; /* may be null */ |
| 75 | struct vnode *lowervp; /* may be null */ |
| 76 | { |
| 77 | int error; |
| 78 | struct union_node *un; |
| 79 | struct union_node **pp; |
| 80 | struct vnode *xlowervp = 0; |
| 81 | |
| 82 | if (uppervp == 0 && lowervp == 0) |
| 83 | panic("union: unidentifiable allocation"); |
| 84 | |
| 85 | if (uppervp && lowervp && (uppervp->v_type != lowervp->v_type)) { |
| 86 | xlowervp = lowervp; |
| 87 | lowervp = 0; |
| 88 | } |
| 89 | |
| 90 | loop: |
| 91 | for (un = unhead; un != 0; un = un->un_next) { |
| 92 | if ((un->un_lowervp == lowervp || |
| 93 | un->un_lowervp == 0) && |
| 94 | (un->un_uppervp == uppervp || |
| 95 | un->un_uppervp == 0) && |
| 96 | (UNIONTOV(un)->v_mount == mp)) { |
| 97 | if (vget(UNIONTOV(un), 0)) |
| 98 | goto loop; |
| 99 | if (UNIONTOV(un) != undvp) |
| 100 | VOP_LOCK(UNIONTOV(un)); |
| 101 | if (uppervp != un->un_uppervp) { |
| 102 | if (un->un_uppervp) |
| 103 | vrele(un->un_uppervp); |
| 104 | un->un_uppervp = uppervp; |
| 105 | } |
| 106 | if (lowervp != un->un_lowervp) { |
| 107 | if (un->un_lowervp) |
| 108 | vrele(un->un_lowervp); |
| 109 | un->un_lowervp = lowervp; |
| 110 | } |
| 111 | *vpp = UNIONTOV(un); |
| 112 | return (0); |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | /* |
| 117 | * otherwise lock the vp list while we call getnewvnode |
| 118 | * since that can block. |
| 119 | */ |
| 120 | if (unvplock & UN_LOCKED) { |
| 121 | unvplock |= UN_WANT; |
| 122 | sleep((caddr_t) &unvplock, PINOD); |
| 123 | goto loop; |
| 124 | } |
| 125 | unvplock |= UN_LOCKED; |
| 126 | |
| 127 | error = getnewvnode(VT_UNION, mp, union_vnodeop_p, vpp); |
| 128 | if (error) |
| 129 | goto out; |
| 130 | |
| 131 | MALLOC((*vpp)->v_data, void *, sizeof(struct union_node), |
| 132 | M_TEMP, M_WAITOK); |
| 133 | |
| 134 | if (uppervp) |
| 135 | (*vpp)->v_type = uppervp->v_type; |
| 136 | else |
| 137 | (*vpp)->v_type = lowervp->v_type; |
| 138 | un = VTOUNION(*vpp); |
| 139 | un->un_vnode = *vpp; |
| 140 | un->un_next = 0; |
| 141 | un->un_uppervp = uppervp; |
| 142 | un->un_lowervp = lowervp; |
| 143 | un->un_flags = 0; |
| 144 | if (uppervp == 0 && cnp) { |
| 145 | un->un_path = malloc(cnp->cn_namelen+1, M_TEMP, M_WAITOK); |
| 146 | bcopy(cnp->cn_nameptr, un->un_path, cnp->cn_namelen); |
| 147 | un->un_path[cnp->cn_namelen] = '\0'; |
| 148 | VREF(dvp); |
| 149 | un->un_dirvp = dvp; |
| 150 | } else { |
| 151 | un->un_path = 0; |
| 152 | un->un_dirvp = 0; |
| 153 | } |
| 154 | |
| 155 | /* add to union vnode list */ |
| 156 | for (pp = &unhead; *pp; pp = &(*pp)->un_next) |
| 157 | continue; |
| 158 | *pp = un; |
| 159 | |
| 160 | un->un_flags |= UN_LOCKED; |
| 161 | |
| 162 | #ifdef DIAGNOSTIC |
| 163 | un->un_pid = curproc->p_pid; |
| 164 | #endif |
| 165 | |
| 166 | if (xlowervp) |
| 167 | vrele(xlowervp); |
| 168 | |
| 169 | out: |
| 170 | unvplock &= ~UN_LOCKED; |
| 171 | |
| 172 | if (unvplock & UN_WANT) { |
| 173 | unvplock &= ~UN_WANT; |
| 174 | wakeup((caddr_t) &unvplock); |
| 175 | } |
| 176 | |
| 177 | return (error); |
| 178 | } |
| 179 | |
| 180 | int |
| 181 | union_freevp(vp) |
| 182 | struct vnode *vp; |
| 183 | { |
| 184 | struct union_node **unpp; |
| 185 | struct union_node *un = VTOUNION(vp); |
| 186 | |
| 187 | for (unpp = &unhead; *unpp != 0; unpp = &(*unpp)->un_next) { |
| 188 | if (*unpp == un) { |
| 189 | *unpp = un->un_next; |
| 190 | break; |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | if (un->un_path) |
| 195 | FREE(un->un_path, M_TEMP); |
| 196 | |
| 197 | FREE(vp->v_data, M_TEMP); |
| 198 | vp->v_data = 0; |
| 199 | return (0); |
| 200 | } |
| 201 | |
| 202 | /* |
| 203 | * copyfile. copy the vnode (fvp) to the vnode (tvp) |
| 204 | * using a sequence of reads and writes. both (fvp) |
| 205 | * and (tvp) are locked on entry and exit. |
| 206 | */ |
| 207 | int |
| 208 | union_copyfile(p, cred, fvp, tvp) |
| 209 | struct proc *p; |
| 210 | struct ucred *cred; |
| 211 | struct vnode *fvp; |
| 212 | struct vnode *tvp; |
| 213 | { |
| 214 | char *buf; |
| 215 | struct uio uio; |
| 216 | struct iovec iov; |
| 217 | int error = 0; |
| 218 | |
| 219 | /* |
| 220 | * strategy: |
| 221 | * allocate a buffer of size MAXBSIZE. |
| 222 | * loop doing reads and writes, keeping track |
| 223 | * of the current uio offset. |
| 224 | * give up at the first sign of trouble. |
| 225 | */ |
| 226 | |
| 227 | uio.uio_procp = p; |
| 228 | uio.uio_segflg = UIO_SYSSPACE; |
| 229 | uio.uio_offset = 0; |
| 230 | |
| 231 | VOP_UNLOCK(fvp); /* XXX */ |
| 232 | LEASE_CHECK(fvp, p, cred, LEASE_READ); |
| 233 | VOP_LOCK(fvp); /* XXX */ |
| 234 | VOP_UNLOCK(tvp); /* XXX */ |
| 235 | LEASE_CHECK(tvp, p, cred, LEASE_WRITE); |
| 236 | VOP_LOCK(tvp); /* XXX */ |
| 237 | |
| 238 | buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK); |
| 239 | |
| 240 | /* ugly loop follows... */ |
| 241 | do { |
| 242 | off_t offset = uio.uio_offset; |
| 243 | |
| 244 | uio.uio_iov = &iov; |
| 245 | uio.uio_iovcnt = 1; |
| 246 | iov.iov_base = buf; |
| 247 | iov.iov_len = MAXBSIZE; |
| 248 | uio.uio_resid = iov.iov_len; |
| 249 | uio.uio_rw = UIO_READ; |
| 250 | error = VOP_READ(fvp, &uio, 0, cred); |
| 251 | |
| 252 | if (error == 0) { |
| 253 | uio.uio_iov = &iov; |
| 254 | uio.uio_iovcnt = 1; |
| 255 | iov.iov_base = buf; |
| 256 | iov.iov_len = MAXBSIZE - uio.uio_resid; |
| 257 | uio.uio_offset = offset; |
| 258 | uio.uio_rw = UIO_WRITE; |
| 259 | uio.uio_resid = iov.iov_len; |
| 260 | |
| 261 | if (uio.uio_resid == 0) |
| 262 | break; |
| 263 | |
| 264 | do { |
| 265 | error = VOP_WRITE(tvp, &uio, 0, cred); |
| 266 | } while ((uio.uio_resid > 0) && (error == 0)); |
| 267 | } |
| 268 | |
| 269 | } while (error == 0); |
| 270 | |
| 271 | free(buf, M_TEMP); |
| 272 | return (error); |
| 273 | } |
| 274 | |
| 275 | /* |
| 276 | * union_vn_create: creates and opens a new shadow file |
| 277 | * on the upper union layer. this function is similar |
| 278 | * in spirit to calling vn_open but it avoids calling namei(). |
| 279 | * the problem with calling namei is that a) it locks too many |
| 280 | * things, and b) it doesn't start at the "right" directory, |
| 281 | * whereas relookup is told where to start. |
| 282 | */ |
| 283 | int |
| 284 | union_vn_create(vpp, un, cmode, p) |
| 285 | struct vnode **vpp; |
| 286 | struct union_node *un; |
| 287 | int cmode; |
| 288 | struct proc *p; |
| 289 | { |
| 290 | struct vnode *vp; |
| 291 | struct ucred *cred = p->p_ucred; |
| 292 | struct vattr vat; |
| 293 | struct vattr *vap = &vat; |
| 294 | int fmode = FFLAGS(O_WRONLY|O_CREAT|O_TRUNC|O_EXCL); |
| 295 | int error; |
| 296 | int hash; |
| 297 | char *cp; |
| 298 | struct componentname cn; |
| 299 | |
| 300 | *vpp = NULLVP; |
| 301 | |
| 302 | cn.cn_namelen = strlen(un->un_path); |
| 303 | cn.cn_pnbuf = (caddr_t) malloc(cn.cn_namelen, M_NAMEI, M_WAITOK); |
| 304 | bcopy(un->un_path, cn.cn_pnbuf, cn.cn_namelen+1); |
| 305 | cn.cn_nameiop = CREATE; |
| 306 | cn.cn_flags = (LOCKLEAF|LOCKPARENT|HASBUF|SAVENAME|ISLASTCN); |
| 307 | cn.cn_proc = p; |
| 308 | cn.cn_cred = p->p_ucred; |
| 309 | cn.cn_nameptr = cn.cn_pnbuf; |
| 310 | for (hash = 0, cp = cn.cn_nameptr; *cp != 0 && *cp != '/'; cp++) |
| 311 | hash += (unsigned char)*cp; |
| 312 | cn.cn_hash = hash; |
| 313 | cn.cn_consume = 0; |
| 314 | |
| 315 | if (error = relookup(un->un_dirvp, &vp, &cn)) |
| 316 | return (error); |
| 317 | if (vp == NULLVP) { |
| 318 | VATTR_NULL(vap); |
| 319 | vap->va_type = VREG; |
| 320 | vap->va_mode = cmode; |
| 321 | LEASE_CHECK(un->un_dirvp, p, cred, LEASE_WRITE); |
| 322 | if (error = VOP_CREATE(un->un_dirvp, &vp, |
| 323 | &cn, vap)) |
| 324 | return (error); |
| 325 | } else { |
| 326 | VOP_ABORTOP(un->un_dirvp, &cn); |
| 327 | if (un->un_dirvp == vp) |
| 328 | vrele(un->un_dirvp); |
| 329 | else |
| 330 | vput(vp); |
| 331 | error = EEXIST; |
| 332 | goto bad; |
| 333 | } |
| 334 | |
| 335 | if (vp->v_type != VREG) { |
| 336 | error = EOPNOTSUPP; |
| 337 | goto bad; |
| 338 | } |
| 339 | |
| 340 | VOP_UNLOCK(vp); /* XXX */ |
| 341 | LEASE_CHECK(vp, p, cred, LEASE_WRITE); |
| 342 | VOP_LOCK(vp); /* XXX */ |
| 343 | VATTR_NULL(vap); |
| 344 | vap->va_size = 0; |
| 345 | if (error = VOP_SETATTR(vp, vap, cred, p)) |
| 346 | goto bad; |
| 347 | |
| 348 | if (error = VOP_OPEN(vp, fmode, cred, p)) |
| 349 | goto bad; |
| 350 | |
| 351 | vp->v_writecount++; |
| 352 | *vpp = vp; |
| 353 | return (0); |
| 354 | bad: |
| 355 | vput(vp); |
| 356 | return (error); |
| 357 | } |