X-Git-Url: https://git.subgeniuskitty.com/unix-history/.git/blobdiff_plain/15797ecffd4ff72f0ad83460b11109059f8d2bf7..0c36425bcf45f094765bfa0bacd3e9a8403c3664:/usr/src/sys/ufs/ffs/ufs_lookup.c diff --git a/usr/src/sys/ufs/ffs/ufs_lookup.c b/usr/src/sys/ufs/ffs/ufs_lookup.c index 2d4dc4349e..36ab7c41dd 100644 --- a/usr/src/sys/ufs/ffs/ufs_lookup.c +++ b/usr/src/sys/ufs/ffs/ufs_lookup.c @@ -1,361 +1,156 @@ -/* ufs_lookup.c 6.21 85/02/22 */ +/* + * Copyright (c) 1989 The Regents of the University of California. + * All rights reserved. + * + * Redistribution and use in source and binary forms are permitted + * provided that the above copyright notice and this paragraph are + * duplicated in all such forms and that any documentation, + * advertising materials, and other materials related to such + * distribution and use acknowledge that the software was developed + * by the University of California, Berkeley. The name of the + * University may not be used to endorse or promote products derived + * from this software without specific prior written permission. + * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. + * + * @(#)ufs_lookup.c 7.19 (Berkeley) %G% + */ #include "param.h" -#include "systm.h" -#include "inode.h" -#include "fs.h" -#include "mount.h" -#include "dir.h" #include "user.h" #include "buf.h" -#include "conf.h" -#include "uio.h" -#include "kernel.h" - -struct buf *blkatoff(); -struct buf *freenamebuf; -int dirchk = 0; - -/* - * Structures associated with name cacheing. - */ -#define NCHHASH 32 /* size of hash table */ - -#if ((NCHHASH)&((NCHHASH)-1)) != 0 -#define NHASH(h, i, d) ((unsigned)((h) + (i) + 13 * (int)(d)) % (NCHHASH)) -#else -#define NHASH(h, i, d) ((unsigned)((h) + (i) + 13 * (int)(d)) & ((NCHHASH)-1)) -#endif +#include "file.h" +#include "vnode.h" +#include "../ufs/inode.h" +#include "../ufs/fs.h" -union nchash { - union nchash *nch_head[2]; - struct nch *nch_chain[2]; -} nchash[NCHHASH]; -#define nch_forw nch_chain[0] -#define nch_back nch_chain[1] - -struct nch *nchhead, **nchtail; /* LRU chain pointers */ -struct nchstats nchstats; /* cache effectiveness statistics */ +struct nchstats nchstats; +int dirchk = 1; /* - * Convert a pathname into a pointer to a locked inode, - * with side effects usable in creating and removing files. + * Convert a component of a pathname into a pointer to a locked inode. * This is a very central and rather complicated routine. - * - * The segflg defines whether the name is to be copied from user - * space or kernel space. - * - * The flag argument is (LOOKUP, CREATE, DELETE) depending on whether - * the name is to be (looked up, created, deleted). If flag has - * LOCKPARENT or'ed into it and the target of the pathname exists, - * namei returns both the target and its parent directory locked. * If the file system is not maintained in a strict tree hierarchy, - * this can result in a deadlock situation. When creating and - * LOCKPARENT is specified, the target may not be ".". When deleting - * and LOCKPARENT is specified, the target may be ".", but the caller - * must check to insure it does an irele and iput instead of two iputs. + * this can result in a deadlock situation (see comments in code below). * - * The FOLLOW flag is set when symbolic links are to be followed - * when they occur at the end of the name translation process. + * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on + * whether the name is to be looked up, created, renamed, or deleted. + * When CREATE, RENAME, or DELETE is specified, information usable in + * creating, renaming, or deleting a directory entry may be calculated. + * If flag has LOCKPARENT or'ed into it and the target of the pathname + * exists, lookup returns both the target and its parent directory locked. + * When creating or renaming and LOCKPARENT is specified, the target may + * not be ".". When deleting and LOCKPARENT is specified, the target may + * be "."., but the caller must check to ensure it does an vrele and iput + * instead of two iputs. * - * Name caching works as follows: + * Overall outline of ufs_lookup: * - * names found by directory scans are retained in a cache - * for future reference. It is managed LRU, so frequently - * used names will hang around. Cache is indexed by hash value - * obtained from (ino,dev,name) where ino & dev refer to the - * directory containing name. - * - * For simplicity (and economy of storage), names longer than - * some (small) maximum length are not cached, they occur - * infrequently in any case, and are almost never of interest. - * - * Upon reaching the last segment of a path, if the reference - * is for DELETE, or NOCACHE is set (rewrite), and the - * name is located in the cache, it will be dropped. - * - * We must be sure never to enter the name ".." into the cache - * because of the extremely kludgey way that rename() alters - * ".." in a situation like - * mv a/x b/x - * where x is a directory, and x/.. is the ".." in question. - * - * Overall outline of namei: - * - * copy in name - * get starting directory - * dirloop: * check accessibility of directory - * dirloop2: - * copy next component of name to ndp->ni_dent - * handle degenerate case where name is null string * look for name in cache, if found, then if at end of path - * and deleting or creating, drop it, else to haveino + * and deleting or creating, drop it, else return name * search for name in directory, to found or notfound * notfound: - * if creating, return locked directory, leaving info on avail. slots + * if creating, return locked directory, leaving info on available slots * else return error * found: * if at end of path and deleting, return information to allow delete - * if at end of path and rewriting (create and LOCKPARENT), lock target + * if at end of path and rewriting (RENAME and LOCKPARENT), lock target * inode and return info to allow rewrite - * if .. and on mounted filesys, look in mount table for parent - * if not at end, if neither creating nor deleting, add name to cache - * haveino: - * if symbolic link, massage name in buffer and continue at dirloop - * if more components of name, do next level at dirloop - * return the answer as locked inode + * if not at end, add name to cache; if at end and neither creating + * nor deleting, add name to cache * - * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode, - * but unlocked. + * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked. */ -struct inode * -namei(ndp) +ufs_lookup(vp, ndp) + struct vnode *vp; register struct nameidata *ndp; { - register char *cp; /* pointer into pathname argument */ -/* these variables refer to things which must be freed or unlocked */ - register struct inode *dp = 0; /* the directory we are searching */ - register struct nch *ncp; /* cache slot for entry */ + register struct vnode *vdp; /* vnode copy of dp */ + register struct inode *dp; /* the directory we are searching */ register struct fs *fs; /* file system that directory is in */ - register struct buf *bp = 0; /* a buffer of directory entries */ + struct buf *bp = 0; /* a buffer of directory entries */ register struct direct *ep; /* the current directory entry */ int entryoffsetinblock; /* offset of ep in bp's buffer */ - register struct buf *nbp; /* buffer storing path name argument */ -/* these variables hold information about the search for a slot */ enum {NONE, COMPACT, FOUND} slotstatus; int slotoffset = -1; /* offset of area with free space */ int slotsize; /* size of area at slotoffset */ int slotfreespace; /* amount of space free in slot */ int slotneeded; /* size of the entry we're seeking */ -/* */ int numdirpasses; /* strategy for directory search */ int endsearch; /* offset to end directory search */ int prevoff; /* ndp->ni_offset of previous entry */ - int nlink = 0; /* number of symbolic links taken */ struct inode *pdp; /* saved dp during symlink work */ - int error, i; - int lockparent; - int docache; - unsigned hash; /* value of name hash for entry */ - union nchash *nhp; /* cache chain head for entry */ - int isdotdot; /* != 0 if current name is ".." */ - int flag; /* op ie, LOOKUP, CREATE, or DELETE */ + struct inode *tdp; /* returned by iget */ off_t enduseful; /* pointer past last used dir slot */ - - lockparent = ndp->ni_nameiop & LOCKPARENT; - docache = (ndp->ni_nameiop & NOCACHE) ^ NOCACHE; - flag = ndp->ni_nameiop &~ (LOCKPARENT|NOCACHE|FOLLOW); - if (flag == DELETE) - docache = 0; - /* - * Get a buffer for the name to be translated, and copy the - * name into the buffer. - */ - nbp = freenamebuf; - if (nbp == NULL) - nbp = geteblk(MAXPATHLEN); - else - freenamebuf = nbp->av_forw; - if (ndp->ni_segflg == UIO_SYSSPACE) - error = copystr(ndp->ni_dirp, nbp->b_un.b_addr, MAXPATHLEN, - (u_int *)0); - else - error = copyinstr(ndp->ni_dirp, nbp->b_un.b_addr, MAXPATHLEN, - (u_int *)0); - if (error) { - u.u_error = error; - goto bad; - } - - /* - * Get starting directory. - */ - cp = nbp->b_un.b_addr; - if (*cp == '/') { - while (*cp == '/') - cp++; - if ((dp = u.u_rdir) == NULL) - dp = rootdir; - } else - dp = u.u_cdir; + int flag; /* LOOKUP, CREATE, RENAME, or DELETE */ + int lockparent; /* 1 => lockparent flag is set */ + int wantparent; /* 1 => wantparent or lockparent flag */ + int error; + + ndp->ni_dvp = vp; + ndp->ni_vp = NULL; + dp = VTOI(vp); fs = dp->i_fs; - ILOCK(dp); - dp->i_count++; - ndp->ni_pdir = (struct inode *)0xc0000000; /* illegal */ - ndp->ni_endoff = 0; + lockparent = ndp->ni_nameiop & LOCKPARENT; + flag = ndp->ni_nameiop & OPFLAG; + wantparent = ndp->ni_nameiop & (LOCKPARENT|WANTPARENT); - /* - * We come to dirloop to search a new directory. - * The directory must be locked so that it can be - * iput, and fs must be already set to dp->i_fs. - */ -dirloop: /* * Check accessiblity of directory. */ - if ((dp->i_mode&IFMT) != IFDIR) { - u.u_error = ENOTDIR; - goto bad; - } - if (access(dp, IEXEC)) - goto bad; - -dirloop2: - /* - * Copy next component of name to ndp->ni_dent. - */ - hash = 0; - for (i = 0; *cp != 0 && *cp != '/'; cp++) { - if (i >= MAXNAMLEN) { - u.u_error = ENOENT; - goto bad; - } - if ((*cp&0377) == ('/'|0200) || (*cp&0200) && flag != DELETE) { - u.u_error = EPERM; - goto bad; - } - ndp->ni_dent.d_name[i++] = *cp; - hash += (unsigned char)*cp * i; - } - ndp->ni_dent.d_namlen = i; - ndp->ni_dent.d_name[i] = '\0'; - isdotdot = (i == 2 && - ndp->ni_dent.d_name[0] == '.' && ndp->ni_dent.d_name[1] == '.'); - - /* - * Check for degenerate name (e.g. / or "") - * which is a way of talking about a directory, - * e.g. like "/." or ".". - */ - if (ndp->ni_dent.d_name[0] == '\0') { - if (flag != LOOKUP || lockparent) { - u.u_error = EISDIR; - goto bad; - } - nbp->av_forw = freenamebuf; - freenamebuf = nbp; - return (dp); - } + if ((dp->i_mode&IFMT) != IFDIR) + return (ENOTDIR); + if (error = iaccess(dp, IEXEC, ndp->ni_cred)) + return (error); /* * We now have a segment name to search for, and a directory to search. * * Before tediously performing a linear scan of the directory, * check the name cache to see if the directory/name pair - * we are looking for is known already. We don't do this - * if the segment name is long, simply so the cache can avoid - * holding long names (which would either waste space, or - * add greatly to the complexity). + * we are looking for is known already. */ - if (ndp->ni_dent.d_namlen > NCHNAMLEN) { - nchstats.ncs_long++; - docache = 0; - } else { - nhp = &nchash[NHASH(hash, dp->i_number, dp->i_dev)]; - for (ncp = nhp->nch_forw; ncp != (struct nch *)nhp; - ncp = ncp->nc_forw) { - if (ncp->nc_ino == dp->i_number && - ncp->nc_dev == dp->i_dev && - ncp->nc_nlen == ndp->ni_dent.d_namlen && - !bcmp(ncp->nc_name, ndp->ni_dent.d_name, - ncp->nc_nlen)) - break; - } + if (error = cache_lookup(ndp)) { + int vpid; /* capability number of vnode */ - if (ncp == (struct nch *)nhp) { - nchstats.ncs_miss++; - ncp = NULL; + if (error == ENOENT) + return (error); + if (vp == ndp->ni_rdir && ndp->ni_isdotdot) + panic("ufs_lookup: .. through root"); + /* + * Get the next vnode in the path. + * See comment below starting `Step through' for + * an explaination of the locking protocol. + */ + pdp = dp; + dp = VTOI(ndp->ni_vp); + vdp = ndp->ni_vp; + vpid = vdp->v_id; + if (pdp == dp) { + VREF(vdp); + error = 0; + } else if (ndp->ni_isdotdot) { + IUNLOCK(pdp); + error = vget(vdp); } else { - if (ncp->nc_id != ncp->nc_ip->i_id) { - nchstats.ncs_falsehits++; - } else if (*cp == '\0' && !docache) { - nchstats.ncs_badhits++; - } else { - - /* - * move this slot to end of LRU - * chain, if not already there - */ - if (ncp->nc_nxt) { - /* remove from LRU chain */ - *ncp->nc_prev = ncp->nc_nxt; - ncp->nc_nxt->nc_prev = ncp->nc_prev; - - /* and replace at end of it */ - ncp->nc_nxt = NULL; - ncp->nc_prev = nchtail; - *nchtail = ncp; - nchtail = &ncp->nc_nxt; - } - - /* - * Get the next inode in the path. - * See comment above other `IUNLOCK' code for - * an explaination of the locking protocol. - */ - pdp = dp; - dp = ncp->nc_ip; - if (dp == NULL) - panic("nami: null cache ino"); - if (pdp == dp) - dp->i_count++; - else { - if (isdotdot) { - IUNLOCK(pdp); - igrab(dp); - } else { - igrab(dp); - IUNLOCK(pdp); - } - } - - /* - * Verify that the inode that we got - * did not change while we were waiting - * for it to be locked. - */ - if (ncp->nc_id != ncp->nc_ip->i_id) { - iput(dp); - ILOCK(pdp); - dp = pdp; - nchstats.ncs_falsehits++; - } else { - ndp->ni_dent.d_ino = dp->i_number; - /* ni_dent.d_reclen is garbage ... */ - nchstats.ncs_goodhits++; - goto haveino; - } - } - - /* - * Last component and we are renaming or deleting, - * the cache entry is invalid, or otherwise don't - * want cache entry to exist. - */ - - /* remove from LRU chain */ - *ncp->nc_prev = ncp->nc_nxt; - if (ncp->nc_nxt) - ncp->nc_nxt->nc_prev = ncp->nc_prev; + error = vget(vdp); + IUNLOCK(pdp); + } + /* + * Check that the capability number did not change + * while we were waiting for the lock. + */ + if (!error) { + if (vpid == vdp->v_id) + return (0); else - nchtail = ncp->nc_prev; - - /* remove from hash chain */ - remque(ncp); - - /* insert at head of LRU list (first to grab) */ - ncp->nc_nxt = nchhead; - ncp->nc_prev = &nchhead; - nchhead->nc_prev = &ncp->nc_nxt; - nchhead = ncp; - - /* and make a dummy hash chain */ - ncp->nc_forw = ncp; - ncp->nc_back = ncp; - - ncp = NULL; + iput(dp); } + ILOCK(pdp); + dp = pdp; + ndp->ni_vp = NULL; } /* @@ -365,39 +160,33 @@ dirloop2: * case it doesn't already exist. */ slotstatus = FOUND; - if (flag == CREATE && *cp == 0) { + if ((flag == CREATE || flag == RENAME) && *ndp->ni_next == 0) { slotstatus = NONE; slotfreespace = 0; slotneeded = DIRSIZ(&ndp->ni_dent); } + /* - * If this is the same directory that this process - * previously searched, pick up where we last left off. + * If there is cached information on a previous search of + * this directory, pick up where we last left off. * We cache only lookups as these are the most common * and have the greatest payoff. Caching CREATE has little * benefit as it usually must search the entire directory * to determine that the entry does not exist. Caching the - * location of the last DELETE has not reduced profiling time - * and hence has been removed in the interest of simplicity. + * location of the last DELETE or RENAME has not reduced + * profiling time and hence has been removed in the interest + * of simplicity. */ - if (flag != LOOKUP || dp->i_number != u.u_ncache.nc_inumber || - dp->i_dev != u.u_ncache.nc_dev) { + if (flag != LOOKUP || dp->i_diroff == 0 || dp->i_diroff > dp->i_size) { ndp->ni_offset = 0; numdirpasses = 1; } else { - if ((dp->i_flag & ICHG) || dp->i_ctime >= u.u_ncache.nc_time) { - if (u.u_ncache.nc_prevoffset > dp->i_size) - u.u_ncache.nc_prevoffset = 0; - else - u.u_ncache.nc_prevoffset &= ~(DIRBLKSIZ - 1); - u.u_ncache.nc_time = time.tv_sec; - } - ndp->ni_offset = u.u_ncache.nc_prevoffset; + ndp->ni_offset = dp->i_diroff; entryoffsetinblock = blkoff(fs, ndp->ni_offset); if (entryoffsetinblock != 0) { - bp = blkatoff(dp, ndp->ni_offset, (char **)0); - if (bp == 0) - goto bad; + error = blkatoff(dp, ndp->ni_offset, (char **)0, &bp); + if (error) + return (error); } numdirpasses = 2; nchstats.ncs_2passes++; @@ -415,22 +204,20 @@ searchloop: if (blkoff(fs, ndp->ni_offset) == 0) { if (bp != NULL) brelse(bp); - bp = blkatoff(dp, ndp->ni_offset, (char **)0); - if (bp == 0) - goto bad; + error = blkatoff(dp, ndp->ni_offset, (char **)0, &bp); + if (error) + return (error); entryoffsetinblock = 0; } - /* * If still looking for a slot, and at a DIRBLKSIZE * boundary, have to start looking for free space again. */ if (slotstatus == NONE && - (entryoffsetinblock&(DIRBLKSIZ-1)) == 0) { + (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) { slotoffset = -1; slotfreespace = 0; } - /* * Get pointer to next entry. * Full validation checks are slow, so we only check @@ -439,8 +226,10 @@ searchloop: * "dirchk" to be true. */ ep = (struct direct *)(bp->b_un.b_addr + entryoffsetinblock); - if (ep->d_reclen <= 0 || + if (ep->d_reclen == 0 || dirchk && dirbadentry(ep, entryoffsetinblock)) { + int i; + dirbad(dp, ndp->ni_offset, "mangled entry"); i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)); ndp->ni_offset += i; @@ -482,9 +271,18 @@ searchloop: */ if (ep->d_ino) { if (ep->d_namlen == ndp->ni_dent.d_namlen && - !bcmp(ndp->ni_dent.d_name, ep->d_name, - ep->d_namlen)) + !bcmp(ndp->ni_ptr, ep->d_name, + (unsigned)ep->d_namlen)) { + /* + * Save directory entry's inode number and + * reclen in ndp->ni_dent, and release + * directory buffer. + */ + ndp->ni_dent.d_ino = ep->d_ino; + ndp->ni_dent.d_reclen = ep->d_reclen; + brelse(bp); goto found; + } } prevoff = ndp->ni_offset; ndp->ni_offset += ep->d_reclen; @@ -500,21 +298,24 @@ searchloop: if (numdirpasses == 2) { numdirpasses--; ndp->ni_offset = 0; - endsearch = u.u_ncache.nc_prevoffset; + endsearch = dp->i_diroff; goto searchloop; } + if (bp != NULL) + brelse(bp); /* * If creating, and at end of pathname and current * directory has not been removed, then can consider * allowing file to be created. */ - if (flag == CREATE && *cp == 0 && dp->i_nlink != 0) { + if ((flag == CREATE || flag == RENAME) && + *ndp->ni_next == 0 && dp->i_nlink != 0) { /* * Access for write is interpreted as allowing * creation of files in the directory. */ - if (access(dp, IWRITE)) - goto bad; + if (error = iaccess(dp, IWRITE, ndp->ni_cred)) + return (error); /* * Return an indication of where the new directory * entry should be put. If we didn't find a slot, @@ -535,23 +336,27 @@ searchloop: } ndp->ni_endoff = roundup(enduseful, DIRBLKSIZ); dp->i_flag |= IUPD|ICHG; - if (bp) - brelse(bp); - nbp->av_forw = freenamebuf; - freenamebuf = nbp; /* * We return with the directory locked, so that * the parameters we set up above will still be * valid if we actually decide to do a direnter(). - * We return NULL to indicate that the entry doesn't - * currently exist, leaving a pointer to the (locked) - * directory inode in ndp->ni_pdir. + * We return ni_vp == NULL to indicate that the entry + * does not currently exist; we leave a pointer to + * the (locked) directory inode in ndp->ni_dvp. + * + * NB - if the directory is unlocked, then this + * information cannot be used. */ - ndp->ni_pdir = dp; - return (NULL); + if (!lockparent) + IUNLOCK(dp); } - u.u_error = ENOENT; - goto bad; + /* + * Insert name into cache (as non-existent) if appropriate. + */ + if (ndp->ni_makeentry) + cache_enter(ndp); + return (ENOENT); + found: if (numdirpasses == 2) nchstats.ncs_pass2++; @@ -570,130 +375,85 @@ found: * If the final component of path name, save information * in the cache as to where the entry was found. */ - if (*cp == '\0' && flag == LOOKUP) { - u.u_ncache.nc_prevoffset = ndp->ni_offset; - u.u_ncache.nc_inumber = dp->i_number; - u.u_ncache.nc_dev = dp->i_dev; - u.u_ncache.nc_time = time.tv_sec; - } - /* - * Save directory entry in ndp->ni_dent, - * and release directory buffer. - */ - bcopy((caddr_t)ep, (caddr_t)&ndp->ni_dent, (u_int)DIRSIZ(ep)); - brelse(bp); - bp = NULL; + if (*ndp->ni_next == '\0' && flag == LOOKUP) + dp->i_diroff = ndp->ni_offset &~ (DIRBLKSIZ - 1); /* * If deleting, and at end of pathname, return * parameters which can be used to remove file. - * If the lockparent flag isn't set, we return only - * the directory (in ndp->ni_pdir), otherwise we go + * If the wantparent flag isn't set, we return only + * the directory (in ndp->ni_dvp), otherwise we go * on and lock the inode, being careful with ".". */ - if (flag == DELETE && *cp == 0) { + if (flag == DELETE && *ndp->ni_next == 0) { /* * Write access to directory required to delete files. */ - if (access(dp, IWRITE)) - goto bad; - ndp->ni_pdir = dp; /* for dirremove() */ + if (error = iaccess(dp, IWRITE, ndp->ni_cred)) + return (error); /* * Return pointer to current entry in ndp->ni_offset, * and distance past previous entry (if there * is a previous entry in this block) in ndp->ni_count. - * Save directory inode pointer in ndp->ni_pdir for dirremove(). + * Save directory inode pointer in ndp->ni_dvp for dirremove(). */ if ((ndp->ni_offset&(DIRBLKSIZ-1)) == 0) ndp->ni_count = 0; else ndp->ni_count = ndp->ni_offset - prevoff; - if (lockparent) { - if (dp->i_number == ndp->ni_dent.d_ino) - dp->i_count++; - else { - dp = iget(dp->i_dev, fs, ndp->ni_dent.d_ino); - if (dp == NULL) { - iput(ndp->ni_pdir); - goto bad; - } - /* - * If directory is "sticky", then user must own - * the directory, or the file in it, else he - * may not delete it (unless he's root). This - * implements append-only directories. - */ - if ((ndp->ni_pdir->i_mode & ISVTX) && - u.u_uid != 0 && - u.u_uid != ndp->ni_pdir->i_uid && - dp->i_uid != u.u_uid) { - iput(ndp->ni_pdir); - u.u_error = EPERM; - goto bad; - } - } - } - nbp->av_forw = freenamebuf; - freenamebuf = nbp; - return (dp); - } - - /* - * Special handling for ".." allowing chdir out of mounted - * file system: indirect .. in root inode to reevaluate - * in directory file system was mounted on. - */ - if (isdotdot) { - if (dp == u.u_rdir) - ndp->ni_dent.d_ino = dp->i_number; - else if (ndp->ni_dent.d_ino == ROOTINO && - dp->i_number == ROOTINO) { - for (i = 1; i < NMOUNT; i++) - if (mount[i].m_bufp != NULL && - mount[i].m_dev == dp->i_dev) { - iput(dp); - dp = mount[i].m_inodp; - ILOCK(dp); - dp->i_count++; - fs = dp->i_fs; - cp -= 2; /* back over .. */ - goto dirloop2; + vdp = ITOV(dp); + if (dp->i_number == ndp->ni_dent.d_ino) { + VREF(vdp); + } else { + if (error = iget(dp, ndp->ni_dent.d_ino, &tdp)) + return (error); + vdp = ITOV(tdp); + /* + * If directory is "sticky", then user must own + * the directory, or the file in it, else he + * may not delete it (unless he's root). This + * implements append-only directories. + */ + if ((dp->i_mode & ISVTX) && + ndp->ni_cred->cr_uid != 0 && + ndp->ni_cred->cr_uid != dp->i_uid && + tdp->i_uid != ndp->ni_cred->cr_uid) { + iput(tdp); + return (EPERM); } + if (!lockparent) + IUNLOCK(dp); } + ndp->ni_vp = vdp; + return (0); } /* - * If rewriting (rename), return the inode and the + * If rewriting (RENAME), return the inode and the * information required to rewrite the present directory * Must get inode of directory entry to verify it's a - * regular file, or empty directory. + * regular file, or empty directory. */ - if ((flag == CREATE && lockparent) && *cp == 0) { - if (access(dp, IWRITE)) - goto bad; - ndp->ni_pdir = dp; /* for dirrewrite() */ + if (flag == RENAME && wantparent && *ndp->ni_next == 0) { + if (error = iaccess(dp, IWRITE, ndp->ni_cred)) + return (error); /* - * Careful about locking second inode. - * This can only occur if the target is ".". + * Careful about locking second inode. + * This can only occur if the target is ".". */ - if (dp->i_number == ndp->ni_dent.d_ino) { - u.u_error = EISDIR; /* XXX */ - goto bad; - } - dp = iget(dp->i_dev, fs, ndp->ni_dent.d_ino); - if (dp == NULL) { - iput(ndp->ni_pdir); - goto bad; - } - nbp->av_forw = freenamebuf; - freenamebuf = nbp; - return (dp); + if (dp->i_number == ndp->ni_dent.d_ino) + return (EISDIR); + if (error = iget(dp, ndp->ni_dent.d_ino, &tdp)) + return (error); + ndp->ni_vp = ITOV(tdp); + if (!lockparent) + IUNLOCK(dp); + return (0); } /* - * Check for symbolic link, which may require us to massage the - * name before we continue translation. We do not `iput' the - * directory because we may need it again if the symbolic link + * Step through the translation in the name. We do not `iput' the + * directory because we may need it again if a symbolic link * is relative to the current directory. Instead we save it * unlocked as "pdp". We must get the target inode before unlocking * the directory to insure that the inode will not be removed @@ -711,132 +471,33 @@ found: * that point backwards in the directory structure. */ pdp = dp; - if (isdotdot) { + if (ndp->ni_isdotdot) { IUNLOCK(pdp); /* race to get the inode */ - dp = iget(dp->i_dev, fs, ndp->ni_dent.d_ino); - if (dp == NULL) - goto bad2; + if (error = iget(dp, ndp->ni_dent.d_ino, &tdp)) { + ILOCK(pdp); + return (error); + } + if (lockparent && *ndp->ni_next == '\0') + ILOCK(pdp); + ndp->ni_vp = ITOV(tdp); } else if (dp->i_number == ndp->ni_dent.d_ino) { - dp->i_count++; /* we want ourself, ie "." */ + vdp = ITOV(dp); + VREF(vdp); /* we want ourself, ie "." */ + ndp->ni_vp = vdp; } else { - dp = iget(dp->i_dev, fs, ndp->ni_dent.d_ino); - IUNLOCK(pdp); - if (dp == NULL) - goto bad2; - } - - /* - * insert name into cache (if we want it, and it isn't "." or "..") - * - * all other cases where making a cache entry would be wrong - * have already departed from the code sequence somewhere above. - */ - if (docache) { - if (ncp != NULL) - panic("nami: duplicating cache"); - - /* - * free the cache slot at head of lru chain - */ - if (ncp = nchhead) { - /* remove from lru chain */ - *ncp->nc_prev = ncp->nc_nxt; - if (ncp->nc_nxt) - ncp->nc_nxt->nc_prev = ncp->nc_prev; - else - nchtail = ncp->nc_prev; - - /* remove from old hash chain */ - remque(ncp); - - /* grab the inode we just found */ - ncp->nc_ip = dp; - - /* fill in cache info */ - ncp->nc_ino = pdp->i_number; /* parents inum */ - ncp->nc_dev = pdp->i_dev; /* & device */ - ncp->nc_idev = dp->i_dev; /* our device */ - ncp->nc_id = dp->i_id; /* identifier */ - ncp->nc_nlen = ndp->ni_dent.d_namlen; - bcopy(ndp->ni_dent.d_name, ncp->nc_name, ncp->nc_nlen); - - /* link at end of lru chain */ - ncp->nc_nxt = NULL; - ncp->nc_prev = nchtail; - *nchtail = ncp; - nchtail = &ncp->nc_nxt; - - /* and insert on hash chain */ - insque(ncp, nhp); - } - } - -haveino: - fs = dp->i_fs; - - /* - * Check for symbolic link - */ - if ((dp->i_mode & IFMT) == IFLNK && - ((ndp->ni_nameiop & FOLLOW) || *cp == '/')) { - u_int pathlen = strlen(cp) + 1; - - if (dp->i_size + pathlen >= MAXPATHLEN - 1 || - ++nlink > MAXSYMLINKS) { - u.u_error = ELOOP; - goto bad2; - } - ovbcopy(cp, nbp->b_un.b_addr + dp->i_size, pathlen); - u.u_error = - rdwri(UIO_READ, dp, nbp->b_un.b_addr, (int)dp->i_size, - 0, 1, (int *)0); - if (u.u_error) - goto bad2; - cp = nbp->b_un.b_addr; - iput(dp); - if (*cp == '/') { - irele(pdp); - while (*cp == '/') - cp++; - if ((dp = u.u_rdir) == NULL) - dp = rootdir; - ILOCK(dp); - dp->i_count++; - } else { - dp = pdp; - ILOCK(dp); - } - fs = dp->i_fs; - goto dirloop; + if (error = iget(dp, ndp->ni_dent.d_ino, &tdp)) + return (error); + if (!lockparent || *ndp->ni_next != '\0') + IUNLOCK(pdp); + ndp->ni_vp = ITOV(tdp); } /* - * Not a symbolic link. If more pathname, - * continue at next component, else return. + * Insert name into cache if appropriate. */ - if (*cp == '/') { - while (*cp == '/') - cp++; - irele(pdp); - goto dirloop; - } - nbp->av_forw = freenamebuf; - freenamebuf = nbp; - if (lockparent) - ndp->ni_pdir = pdp; - else - irele(pdp); - return (dp); -bad2: - irele(pdp); -bad: - if (bp) - brelse(bp); - if (dp) - iput(dp); - nbp->av_forw = freenamebuf; - freenamebuf = nbp; - return (NULL); + if (ndp->ni_makeentry) + cache_enter(ndp); + return (0); } @@ -848,12 +509,12 @@ dirbad(ip, offset, how) printf("%s: bad dir ino %d at offset %d: %s\n", ip->i_fs->fs_fsmnt, ip->i_number, offset, how); + panic("bad dir"); } /* * Do consistency checking on a directory entry: * record length must be multiple of 4 - * record length must not be non-negative * entry must fit in rest of its DIRBLKSIZ block * record must be large enough to contain entry * name is not longer than MAXNAMLEN @@ -865,7 +526,7 @@ dirbadentry(ep, entryoffsetinblock) { register int i; - if ((ep->d_reclen & 0x3) != 0 || ep->d_reclen <= 0 || + if ((ep->d_reclen & 0x3) != 0 || ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) || ep->d_reclen < DIRSIZ(ep) || ep->d_namlen > MAXNAMLEN) return (1); @@ -877,9 +538,9 @@ dirbadentry(ep, entryoffsetinblock) /* * Write a directory entry after a call to namei, using the parameters - * which it left in the u. area. The argument ip is the inode which - * the new directory entry will refer to. The u. area field ndp->ni_pdir is - * a pointer to the directory to be written, which was left locked by + * which it left in nameidata. The argument ip is the inode which the + * new directory entry will refer to. The nameidata field ndp->ni_dvp + * is a pointer to the directory to be written, which was left locked by * namei. Remaining parameters (ndp->ni_offset, ndp->ni_count) indicate * how the space for the new entry is to be gotten. */ @@ -888,7 +549,7 @@ direnter(ip, ndp) register struct nameidata *ndp; { register struct direct *ep, *nep; - register struct inode *dp = ndp->ni_pdir; + register struct inode *dp = VTOI(ndp->ni_dvp); struct buf *bp; int loc, spacefree, error = 0; u_int dsize; @@ -907,10 +568,13 @@ direnter(ip, ndp) if (ndp->ni_offset&(DIRBLKSIZ-1)) panic("wdir: newblk"); ndp->ni_dent.d_reclen = DIRBLKSIZ; - error = rdwri(UIO_WRITE, dp, (caddr_t)&ndp->ni_dent, - newentrysize, ndp->ni_offset, 1, (int *)0); + ndp->ni_count = newentrysize; + ndp->ni_resid = newentrysize; + ndp->ni_base = (caddr_t)&ndp->ni_dent; + error = + ufs_write(ndp->ni_dvp, &ndp->ni_uio, IO_SYNC, ndp->ni_cred); if (DIRBLKSIZ > dp->i_fs->fs_fsize) - panic("wdir: blksize"); /* XXX - should grow w/bmap() */ + panic("wdir: blksize"); /* XXX - should grow w/balloc */ else dp->i_size = roundup(dp->i_size, DIRBLKSIZ); iput(dp); @@ -935,17 +599,13 @@ direnter(ip, ndp) */ if (ndp->ni_offset + ndp->ni_count > dp->i_size) dp->i_size = ndp->ni_offset + ndp->ni_count; - /* - * Get the block containing the space for the new directory - * entry. Should return error by result instead of u.u_error. + * Get the block containing the space for the new directory entry. */ - bp = blkatoff(dp, ndp->ni_offset, (char **)&dirbuf); - if (bp == 0) { + if (error = blkatoff(dp, ndp->ni_offset, (char **)&dirbuf, &bp)) { iput(dp); - return (u.u_error); + return (error); } - /* * Find space for the new entry. In the simple case, the * entry at offset base will have the space. If it does @@ -963,7 +623,7 @@ direnter(ip, ndp) ep = (struct direct *)((char *)ep + dsize); } else { /* overwrite; nothing there; header is ours */ - spacefree += dsize; + spacefree += dsize; } dsize = DIRSIZ(nep); spacefree += nep->d_reclen - dsize; @@ -986,17 +646,17 @@ direnter(ip, ndp) ep = (struct direct *)((char *)ep + dsize); } bcopy((caddr_t)&ndp->ni_dent, (caddr_t)ep, (u_int)newentrysize); - bwrite(bp); + error = bwrite(bp); dp->i_flag |= IUPD|ICHG; if (ndp->ni_endoff && ndp->ni_endoff < dp->i_size) - itrunc(dp, ndp->ni_endoff); + error = itrunc(dp, (u_long)ndp->ni_endoff, IO_SYNC); iput(dp); return (error); } /* - * Remove a directory entry after a call to namei, using the - * parameters which it left in the u. area. The u. entry + * Remove a directory entry after a call to namei, using + * the parameters which it left in nameidata. The entry * ni_offset contains the offset into the directory of the * entry to be eliminated. The ni_count field contains the * size of the previous record in the directory. If this @@ -1009,30 +669,33 @@ direnter(ip, ndp) dirremove(ndp) register struct nameidata *ndp; { - register struct inode *dp = ndp->ni_pdir; - register struct buf *bp; + register struct inode *dp = VTOI(ndp->ni_dvp); struct direct *ep; + struct buf *bp; + int error; if (ndp->ni_count == 0) { /* * First entry in block: set d_ino to zero. */ ndp->ni_dent.d_ino = 0; - (void) rdwri(UIO_WRITE, dp, (caddr_t)&ndp->ni_dent, - (int)DIRSIZ(&ndp->ni_dent), ndp->ni_offset, 1, (int *)0); + ndp->ni_count = ndp->ni_resid = DIRSIZ(&ndp->ni_dent); + ndp->ni_base = (caddr_t)&ndp->ni_dent; + error = + ufs_write(ndp->ni_dvp, &ndp->ni_uio, IO_SYNC, ndp->ni_cred); } else { /* * Collapse new free space into previous entry. */ - bp = blkatoff(dp, (int)(ndp->ni_offset - ndp->ni_count), - (char **)&ep); - if (bp == 0) - return (0); + if (error = blkatoff(dp, ndp->ni_offset - ndp->ni_count, + (char **)&ep, &bp)) { + return (error); + } ep->d_reclen += ndp->ni_dent.d_reclen; - bwrite(bp); + error = bwrite(bp); dp->i_flag |= IUPD|ICHG; } - return (1); + return (error); } /* @@ -1046,9 +709,9 @@ dirrewrite(dp, ip, ndp) { ndp->ni_dent.d_ino = ip->i_number; - u.u_error = rdwri(UIO_WRITE, dp, (caddr_t)&ndp->ni_dent, - (int)DIRSIZ(&ndp->ni_dent), ndp->ni_offset, 1, (int *)0); - iput(dp); + ndp->ni_count = ndp->ni_resid = DIRSIZ(&ndp->ni_dent); + ndp->ni_base = (caddr_t)&ndp->ni_dent; + return (ufs_write(ITOV(dp), &ndp->ni_uio, IO_SYNC, ndp->ni_cred)); } /* @@ -1057,29 +720,28 @@ dirrewrite(dp, ip, ndp) * is non-zero, fill it in with a pointer to the * remaining space in the directory. */ -struct buf * -blkatoff(ip, offset, res) +blkatoff(ip, offset, res, bpp) struct inode *ip; off_t offset; char **res; + struct buf **bpp; { register struct fs *fs = ip->i_fs; daddr_t lbn = lblkno(fs, offset); - int base = blkoff(fs, offset); int bsize = blksize(fs, ip, lbn); - daddr_t bn = fsbtodb(fs, bmap(ip, lbn, B_WRITE, base, bsize)); - register struct buf *bp; + struct buf *bp; + daddr_t bn; + int error; - if (u.u_error) - return (0); - bp = bread(ip->i_dev, bn, bsize); - if (bp->b_flags & B_ERROR) { + *bpp = 0; + if (error = bread(ITOV(ip), lbn, bsize, NOCRED, &bp)) { brelse(bp); - return (0); + return (error); } if (res) - *res = bp->b_un.b_addr + base; - return (bp); + *res = bp->b_un.b_addr + blkoff(fs, offset); + *bpp = bp; + return (0); } /* @@ -1091,9 +753,10 @@ blkatoff(ip, offset, res) * * NB: does not handle corrupted directories. */ -dirempty(ip, parentino) +dirempty(ip, parentino, cred) register struct inode *ip; ino_t parentino; + struct ucred *cred; { register off_t off; struct dirtemplate dbuf; @@ -1102,16 +765,17 @@ dirempty(ip, parentino) #define MINDIRSIZ (sizeof (struct dirtemplate) / 2) for (off = 0; off < ip->i_size; off += dp->d_reclen) { - if (dp->d_reclen <= 0) - return (0); - error = rdwri(UIO_READ, ip, (caddr_t)dp, MINDIRSIZ, - off, 1, &count); + error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, + off, UIO_SYSSPACE, IO_NODELOCKED, cred, &count); /* * Since we read MINDIRSIZ, residual must * be 0 unless we're at end of file. */ if (error || count != 0) return (0); + /* avoid infinite loops */ + if (dp->d_reclen == 0) + return (0); /* skip empty entries */ if (dp->d_ino == 0) continue; @@ -1139,11 +803,12 @@ dirempty(ip, parentino) * Target is supplied locked, source is unlocked. * The target is always iput() before returning. */ -checkpath(source, target) +checkpath(source, target, cred) struct inode *source, *target; + struct ucred *cred; { struct dirtemplate dirbuf; - register struct inode *ip; + struct inode *ip; int error = 0; ip = target; @@ -1159,8 +824,9 @@ checkpath(source, target) error = ENOTDIR; break; } - error = rdwri(UIO_READ, ip, (caddr_t)&dirbuf, - sizeof (struct dirtemplate), (off_t)0, 1, (int *)0); + error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)&dirbuf, + sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE, + IO_NODELOCKED, cred, (int *)0); if (error != 0) break; if (dirbuf.dotdot_namlen != 2 || @@ -1176,11 +842,8 @@ checkpath(source, target) if (dirbuf.dotdot_ino == ROOTINO) break; iput(ip); - ip = iget(ip->i_dev, ip->i_fs, dirbuf.dotdot_ino); - if (ip == NULL) { - error = u.u_error; + if (error = iget(ip, dirbuf.dotdot_ino, &ip)) break; - } } out: @@ -1190,95 +853,3 @@ out: iput(ip); return (error); } - -/* - * Name cache initialization, from main() when we are booting - */ -nchinit() -{ - register union nchash *nchp; - register struct nch *ncp; - - nchhead = 0; - nchtail = &nchhead; - - for (ncp = nch; ncp < &nch[nchsize]; ncp++) { - ncp->nc_forw = ncp; /* hash chain */ - ncp->nc_back = ncp; - - ncp->nc_nxt = NULL; /* lru chain */ - *nchtail = ncp; - ncp->nc_prev = nchtail; - nchtail = &ncp->nc_nxt; - - /* all else is zero already */ - } - - for (nchp = nchash; nchp < &nchash[NCHHASH]; nchp++) { - nchp->nch_head[0] = nchp; - nchp->nch_head[1] = nchp; - } -} - -/* - * Cache flush, called when filesys is umounted to - * remove entries that would now be invalid - * - * The line "nxtcp = nchhead" near the end is to avoid potential problems - * if the cache lru chain is modified while we are dumping the - * inode. This makes the algorithm O(n^2), but do you think I care? - */ -nchinval(dev) - register dev_t dev; -{ - register struct nch *ncp, *nxtcp; - - for (ncp = nchhead; ncp; ncp = nxtcp) { - nxtcp = ncp->nc_nxt; - - if (ncp->nc_ip == NULL || - (ncp->nc_idev != dev && ncp->nc_dev != dev)) - continue; - - /* free the resources we had */ - ncp->nc_idev = NODEV; - ncp->nc_dev = NODEV; - ncp->nc_id = NULL; - ncp->nc_ino = 0; - ncp->nc_ip = NULL; - - - /* remove the entry from its hash chain */ - remque(ncp); - /* and make a dummy one */ - ncp->nc_forw = ncp; - ncp->nc_back = ncp; - - /* delete this entry from LRU chain */ - *ncp->nc_prev = nxtcp; - if (nxtcp) - nxtcp->nc_prev = ncp->nc_prev; - else - nchtail = ncp->nc_prev; - - /* cause rescan of list, it may have altered */ - nxtcp = nchhead; - /* put the now-free entry at head of LRU */ - ncp->nc_nxt = nxtcp; - ncp->nc_prev = &nchhead; - nxtcp->nc_prev = &ncp->nc_nxt; - nchhead = ncp; - } -} - -/* - * Name cache invalidation of all entries. - */ -cacheinvalall() -{ - register struct nch *ncp; - - for (ncp = nch; ncp < &nch[nchsize]; ncp++) { - ncp->nc_id = 0; - } -}