X-Git-Url: https://git.subgeniuskitty.com/unix-history/.git/blobdiff_plain/f93197fc776222d2758eb666b8fbabb1bd6feb18..658f5fdcff636c9b1a09f81684b1128aca389721:/usr/src/sys/kern/vfs_lookup.c diff --git a/usr/src/sys/kern/vfs_lookup.c b/usr/src/sys/kern/vfs_lookup.c index 32a0be527d..60e571f568 100644 --- a/usr/src/sys/kern/vfs_lookup.c +++ b/usr/src/sys/kern/vfs_lookup.c @@ -1,20 +1,30 @@ -/* vfs_lookup.c 6.4 84/01/03 */ +/* + * Copyright (c) 1982, 1986 Regents of the University of California. + * All rights reserved. The Berkeley software License Agreement + * specifies the terms and conditions for redistribution. + * + * @(#)vfs_lookup.c 7.6 (Berkeley) %G% + */ -#include "../h/param.h" -#include "../h/systm.h" -#include "../h/inode.h" -#include "../h/fs.h" -#include "../h/mount.h" -#include "../h/dir.h" -#include "../h/user.h" -#include "../h/buf.h" -#include "../h/conf.h" -#include "../h/uio.h" -#include "../h/nami.h" -#include "../h/kernel.h" +#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" +#include "malloc.h" +#ifdef KTRACE +#include "proc.h" +#include "ktrace.h" +#endif struct buf *blkatoff(); -int dirchk = 0; +int dirchk = 1; /* * Structures associated with name cacheing. @@ -27,66 +37,55 @@ int dirchk = 0; #define NHASH(h, i, d) ((unsigned)((h) + (i) + 13 * (int)(d)) & ((NCHHASH)-1)) #endif -union nchash { - union nchash *nch_head[2]; - struct nch *nch_chain[2]; +union nchash { + union nchash *nch_head[2]; + struct namecache *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 { /* stats on usefulness */ - long ncs_goodhits; /* hits that we can reall use */ - long ncs_badhits; /* hits we must drop */ - long ncs_miss; /* misses */ - long ncs_long; /* long names that ignore cache */ - long ncs_pass2; /* names found with passes == 2 */ - long ncs_2passes; /* number of times we attempt it */ -} nchstats; +struct namecache *nchhead, **nchtail; /* LRU chain pointers */ +struct nchstats nchstats; /* cache effectiveness statistics */ /* - * Convert a pathname into a pointer to a locked inode, - * with side effects usable in creating and removing files. + * Convert a pathname into a pointer to a locked inode. * This is a very central and rather complicated routine. - * - * The func argument gives the routine which returns successive - * characters of the name to be translated. - * - * 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 + * this can result in a deadlock situation (see comments in code below). + * + * The flag argument is LOOKUP, CREATE, or DELETE depending on whether + * the name is to be looked up, created, or deleted. When CREATE or + * DELETE is specified, information usable in creating or deleteing a + * directory entry is also calculated. 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. 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. * - * The follow argument is 1 when symbolic links are to be followed + * The FOLLOW flag is set when symbolic links are to be followed * when they occur at the end of the name translation process. + * Symbolic links are always followed for all other pathname + * components other than the last. * - * Name caching works as follows: + * The segflg defines whether the name is to be copied from user + * space or kernel space. * - * 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. + * Name caching works as follows: * - * 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. + * 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. * - * 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. + * For simplicity (and economy of storage), names longer than + * a maximum length of NCHNAMLEN are not cached; they occur + * infrequently in any case, and are almost never of interest. * - * 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. + * 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. * * Overall outline of namei: * @@ -95,7 +94,7 @@ struct nchstats { /* stats on usefulness */ * dirloop: * check accessibility of directory * dirloop2: - * copy next component of name to u.u_dent + * 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 @@ -105,10 +104,11 @@ struct nchstats { /* stats on usefulness */ * 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 (CREATE 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 + * if not at end, add name to cache; if at end and 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 @@ -118,18 +118,18 @@ struct nchstats { /* stats on usefulness */ * but unlocked. */ struct inode * -namei(func, flag, follow) - int (*func)(), flag, follow; +namei(ndp) + 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 namecache *ncp; /* cache slot for entry */ register struct fs *fs; /* file system that directory is in */ register 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 */ + register caddr_t 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 */ @@ -139,44 +139,46 @@ namei(func, flag, follow) /* */ int numdirpasses; /* strategy for directory search */ int endsearch; /* offset to end directory search */ - int prevoff; /* u.u_offset of previous entry */ + 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 i; + int error, i; int lockparent; - int docache; + int docache; /* == 0 do not cache last component */ + int makeentry; /* != 0 if name to be added to cache */ 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 */ + off_t enduseful; /* pointer past last used dir slot */ - lockparent = flag & LOCKPARENT; - docache = (flag & NOCACHE) ^ NOCACHE; - flag &= ~(LOCKPARENT|NOCACHE); - if (flag == DELETE) + lockparent = ndp->ni_nameiop & LOCKPARENT; + docache = (ndp->ni_nameiop & NOCACHE) ^ NOCACHE; + flag = ndp->ni_nameiop &~ (LOCKPARENT|NOCACHE|FOLLOW); + if (flag == DELETE || lockparent) docache = 0; /* * Get a buffer for the name to be translated, and copy the * name into the buffer. */ - nbp = geteblk(MAXPATHLEN); - for (cp = nbp->b_un.b_addr; *cp = (*func)(); ) { - if ((*cp&0377) == ('/'|0200) || (*cp&0200) && flag != LOOKUP) { - u.u_error = EPERM; - goto bad; - } - cp++; - if (cp >= nbp->b_un.b_addr + MAXPATHLEN) { - u.u_error = ENOENT; - goto bad; - } - } - if (u.u_error) + MALLOC(nbp, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK); + if (ndp->ni_segflg == UIO_SYSSPACE) + error = copystr(ndp->ni_dirp, nbp, MAXPATHLEN, (u_int *)0); + else + error = copyinstr(ndp->ni_dirp, nbp, MAXPATHLEN, (u_int *)0); + if (error) { + u.u_error = error; goto bad; + } +#ifdef KTRACE + if (KTRPOINT(u.u_procp, KTR_NAMEI)) + ktrnamei(u.u_procp->p_tracep, nbp); +#endif /* * Get starting directory. */ - cp = nbp->b_un.b_addr; + cp = nbp; if (*cp == '/') { while (*cp == '/') cp++; @@ -185,9 +187,10 @@ namei(func, flag, follow) } else dp = u.u_cdir; fs = dp->i_fs; - ilock(dp); + ILOCK(dp); dp->i_count++; - u.u_pdir = (struct inode *)0xc0000000; /* illegal */ + ndp->ni_pdir = (struct inode *)0xc0000000; /* illegal */ + ndp->ni_endoff = 0; /* * We come to dirloop to search a new directory. @@ -207,34 +210,73 @@ dirloop: dirloop2: /* - * Copy next component of name to u.u_dent. + * 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; + u.u_error = ENAMETOOLONG; goto bad; } - u.u_dent.d_name[i++] = *cp; + if (*cp & 0200) + if ((*cp&0377) == ('/'|0200) || flag != DELETE) { + u.u_error = EINVAL; + goto bad; + } + ndp->ni_dent.d_name[i++] = *cp; hash += (unsigned char)*cp * i; } - u.u_dent.d_namlen = i; - u.u_dent.d_name[i] = 0; + 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] == '.'); + makeentry = 1; + if (*cp == '\0' && docache == 0) + makeentry = 0; /* * Check for degenerate name (e.g. / or "") * which is a way of talking about a directory, * e.g. like "/." or ".". */ - if (u.u_dent.d_name[0] == 0) { + if (ndp->ni_dent.d_name[0] == '\0') { if (flag != LOOKUP || lockparent) { u.u_error = EISDIR; goto bad; } - brelse(nbp); + FREE(nbp, M_NAMEI); 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) { + for (;;) { + if (dp == u.u_rdir || dp == rootdir) { + ndp->ni_dent.d_ino = dp->i_number; + pdp = dp; + dp->i_count++; + goto haveino; + } + if (dp->i_number != ROOTINO) + break; + for (i = 1; i < NMOUNT; i++) { + if (mount[i].m_fs != NULL && + mount[i].m_dev == dp->i_dev) { + iput(dp); + dp = mount[i].m_inodp; + ILOCK(dp); + dp->i_count++; + fs = dp->i_fs; + break; + } + } + } + } + /* * We now have a segment name to search for, and a directory to search. * @@ -245,92 +287,103 @@ dirloop2: * holding long names (which would either waste space, or * add greatly to the complexity). */ - if (u.u_dent.d_namlen > NCHNAMLEN) { + if (ndp->ni_dent.d_namlen > NCHNAMLEN) { nchstats.ncs_long++; - docache = 0; + makeentry = 0; } else { nhp = &nchash[NHASH(hash, dp->i_number, dp->i_dev)]; - for (ncp = nhp->nch_forw; ncp != (struct nch *)nhp; + for (ncp = nhp->nch_forw; ncp != (struct namecache *)nhp; ncp = ncp->nc_forw) { if (ncp->nc_ino == dp->i_number && ncp->nc_dev == dp->i_dev && - ncp->nc_nlen == u.u_dent.d_namlen && - !bcmp(ncp->nc_name, u.u_dent.d_name, ncp->nc_nlen)) + ncp->nc_nlen == ndp->ni_dent.d_namlen && + !bcmp(ncp->nc_name, ndp->ni_dent.d_name, + (unsigned)ncp->nc_nlen)) break; } - - if (ncp == (struct nch *)nhp) { + if (ncp == (struct namecache *)nhp) { nchstats.ncs_miss++; ncp = NULL; } else { - if (*cp == '/' || docache) { - - nchstats.ncs_goodhits++; - - /* - * move this slot to end of LRU - * chain, if not already there - */ + if (ncp->nc_id != ncp->nc_ip->i_id) + nchstats.ncs_falsehits++; + else if (!makeentry) + nchstats.ncs_badhits++; + else { + /* + * move this slot to end of LRU + * chain, if not already there + */ if (ncp->nc_nxt) { - /* remove from LRU chain */ + /* remove from LRU chain */ *ncp->nc_prev = ncp->nc_nxt; ncp->nc_nxt->nc_prev = ncp->nc_prev; - /* and replace at end of it */ + /* 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 (!isdotdot || dp != u.u_rdir) + dp = ncp->nc_ip; if (dp == NULL) - panic("nami: null cache ino"); - if (pdp != dp) { - ilock(dp); - dp->i_count++; - iunlock(pdp); - } else + panic("namei: null cache ino"); + if (pdp == dp) dp->i_count++; + else if (isdotdot) { + IUNLOCK(pdp); + igrab(dp); + } else { + igrab(dp); + IUNLOCK(pdp); + } - u.u_dent.d_ino = dp->i_number; - /* u_dent.d_reclen is garbage ... */ - - goto haveino; + /* + * 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 segment and we are renaming or deleting - * or otherwise don't want cache entry to exist + * Last component and we are renaming or deleting, + * the cache entry is invalid, or otherwise don't + * want cache entry to exist. */ - - nchstats.ncs_badhits++; - - /* remove from LRU chain */ + /* 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 hash chain */ - remque(ncp); - - /* release ref on the inode */ - irele(ncp->nc_ip); - ncp->nc_ip = NULL; - - /* insert at head of LRU list (first to grab) */ + remque(ncp); /* remove from hash chain */ + /* 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 */ + /* and make a dummy hash chain */ ncp->nc_forw = ncp; ncp->nc_back = ncp; - ncp = NULL; } } @@ -345,7 +398,7 @@ dirloop2: if (flag == CREATE && *cp == 0) { slotstatus = NONE; slotfreespace = 0; - slotneeded = DIRSIZ(&u.u_dent); + slotneeded = DIRSIZ(&ndp->ni_dent); } /* * If this is the same directory that this process @@ -359,17 +412,15 @@ dirloop2: */ if (flag != LOOKUP || dp->i_number != u.u_ncache.nc_inumber || dp->i_dev != u.u_ncache.nc_dev) { - u.u_offset = 0; + ndp->ni_offset = 0; numdirpasses = 1; } else { - if ((dp->i_flag & ICHG) || dp->i_ctime >= u.u_ncache.nc_time) { - u.u_ncache.nc_prevoffset &= ~(DIRBLKSIZ - 1); - u.u_ncache.nc_time = time.tv_sec; - } - u.u_offset = u.u_ncache.nc_prevoffset; - entryoffsetinblock = blkoff(fs, u.u_offset); + if (u.u_ncache.nc_prevoffset > dp->i_size) + u.u_ncache.nc_prevoffset = 0; + ndp->ni_offset = u.u_ncache.nc_prevoffset; + entryoffsetinblock = blkoff(fs, ndp->ni_offset); if (entryoffsetinblock != 0) { - bp = blkatoff(dp, u.u_offset, (char **)0); + bp = blkatoff(dp, ndp->ni_offset, (char **)0); if (bp == 0) goto bad; } @@ -377,53 +428,48 @@ dirloop2: nchstats.ncs_2passes++; } endsearch = roundup(dp->i_size, DIRBLKSIZ); + enduseful = 0; +#ifdef NAMEI_DIAGNOSTIC + printf("{%s}:\n", ndp->ni_dent.d_name); +#endif searchloop: - while (u.u_offset < endsearch) { + while (ndp->ni_offset < endsearch) { /* * If offset is on a block boundary, * read the next directory block. * Release previous if it exists. */ - if (blkoff(fs, u.u_offset) == 0) { + if (blkoff(fs, ndp->ni_offset) == 0) { if (bp != NULL) brelse(bp); - bp = blkatoff(dp, u.u_offset, (char **)0); + bp = blkatoff(dp, ndp->ni_offset, (char **)0); if (bp == 0) goto bad; entryoffsetinblock = 0; } - /* * If still looking for a slot, and at a DIRBLKSIZE - * boundary, have to start looking for free space - * again. + * boundary, have to start looking for free space again. */ if (slotstatus == NONE && (entryoffsetinblock&(DIRBLKSIZ-1)) == 0) { slotoffset = -1; slotfreespace = 0; } - /* - * Get pointer to next entry, and do consistency checking: - * record length must be multiple of 4 - * record length must not be zero - * entry must fit in rest of this DIRBLKSIZ block - * record must be large enough to contain name - * When dirchk is set we also check: - * name is not longer than MAXNAMLEN - * name must be as long as advertised, and null terminated - * Checking last two conditions is done only when dirchk is - * set, to save time. + * Get pointer to next entry. + * Full validation checks are slow, so we only check + * enough to insure forward progress through the + * directory. Complete checks can be run by patching + * "dirchk" to be true. */ ep = (struct direct *)(bp->b_un.b_addr + entryoffsetinblock); - i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)); - if ((ep->d_reclen & 0x3) || ep->d_reclen == 0 || - ep->d_reclen > i || DIRSIZ(ep) > ep->d_reclen || - dirchk && (ep->d_namlen > MAXNAMLEN || dirbadname(ep))) { - dirbad(dp, "mangled entry"); - u.u_offset += i; + if (ep->d_reclen == 0 || + dirchk && dirbadentry(ep, entryoffsetinblock)) { + dirbad(dp, ndp->ni_offset, "mangled entry"); + i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)); + ndp->ni_offset += i; entryoffsetinblock += i; continue; } @@ -442,17 +488,16 @@ searchloop: if (size > 0) { if (size >= slotneeded) { slotstatus = FOUND; - slotoffset = u.u_offset; + slotoffset = ndp->ni_offset; slotsize = ep->d_reclen; } else if (slotstatus == NONE) { slotfreespace += size; if (slotoffset == -1) - slotoffset = u.u_offset; + slotoffset = ndp->ni_offset; if (slotfreespace >= slotneeded) { slotstatus = COMPACT; - slotsize = - u.u_offset+ep->d_reclen - - slotoffset; + slotsize = ndp->ni_offset + + ep->d_reclen - slotoffset; } } } @@ -462,14 +507,23 @@ searchloop: * Check for a name match. */ if (ep->d_ino) { - if (ep->d_namlen == u.u_dent.d_namlen && - !bcmp(u.u_dent.d_name, ep->d_name, ep->d_namlen)) +#ifdef NAMEI_DIAGNOSTIC + printf("{%s} ", ep->d_name); +#endif + if (ep->d_namlen == ndp->ni_dent.d_namlen && + !bcmp(ndp->ni_dent.d_name, ep->d_name, + (unsigned)ep->d_namlen)) goto found; } - prevoff = u.u_offset; - u.u_offset += ep->d_reclen; + prevoff = ndp->ni_offset; + ndp->ni_offset += ep->d_reclen; entryoffsetinblock += ep->d_reclen; + if (ep->d_ino) + enduseful = ndp->ni_offset; } +#ifdef NAMEI_DIAGNOSTIC + printf("\nnotfound\n"); +#endif /* notfound: */ /* * If we started in the middle of the directory and failed @@ -477,7 +531,7 @@ searchloop: */ if (numdirpasses == 2) { numdirpasses--; - u.u_offset = 0; + ndp->ni_offset = 0; endsearch = u.u_ncache.nc_prevoffset; goto searchloop; } @@ -496,36 +550,43 @@ searchloop: /* * Return an indication of where the new directory * entry should be put. If we didn't find a slot, - * then set u.u_count to 0 indicating that the - * new slot belongs at the end of the directory. - * If we found a slot, then the new entry can be - * put in the range [u.u_offset..u.u_offset+u.u_count) + * then set ndp->ni_count to 0 indicating that the new + * slot belongs at the end of the directory. If we found + * a slot, then the new entry can be put in the range + * [ndp->ni_offset .. ndp->ni_offset + ndp->ni_count) */ if (slotstatus == NONE) { - u.u_offset = roundup(dp->i_size, DIRBLKSIZ); - u.u_count = 0; + ndp->ni_offset = roundup(dp->i_size, DIRBLKSIZ); + ndp->ni_count = 0; + enduseful = ndp->ni_offset; } else { - u.u_offset = slotoffset; - u.u_count = slotsize; + ndp->ni_offset = slotoffset; + ndp->ni_count = slotsize; + if (enduseful < slotoffset + slotsize) + enduseful = slotoffset + slotsize; } + ndp->ni_endoff = roundup(enduseful, DIRBLKSIZ); dp->i_flag |= IUPD|ICHG; if (bp) brelse(bp); - brelse(nbp); + FREE(nbp, M_NAMEI); /* * 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 u.u_pdir. + * directory inode in ndp->ni_pdir. */ - u.u_pdir = dp; + ndp->ni_pdir = dp; return (NULL); } u.u_error = ENOENT; goto bad; found: +#ifdef NAMEI_DIAGNOSTIC + printf("\nfound\n"); +#endif if (numdirpasses == 2) nchstats.ncs_pass2++; /* @@ -533,7 +594,7 @@ found: * of this entry. */ if (entryoffsetinblock + DIRSIZ(ep) > dp->i_size) { - dirbad(dp, "i_size too small"); + dirbad(dp, ndp->ni_offset, "i_size too small"); dp->i_size = entryoffsetinblock + DIRSIZ(ep); dp->i_flag |= IUPD|ICHG; } @@ -544,16 +605,16 @@ found: * in the cache as to where the entry was found. */ if (*cp == '\0' && flag == LOOKUP) { - u.u_ncache.nc_prevoffset = u.u_offset; + u.u_ncache.nc_prevoffset = ndp->ni_offset &~ (DIRBLKSIZ - 1); 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 u.u_dent, + * Save directory entry's inode number and reclen in ndp->ni_dent, * and release directory buffer. */ - bcopy((caddr_t)ep, (caddr_t)&u.u_dent, (u_int)DIRSIZ(ep)); + ndp->ni_dent.d_ino = ep->d_ino; + ndp->ni_dent.d_reclen = ep->d_reclen; brelse(bp); bp = NULL; @@ -561,7 +622,7 @@ found: * 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 u.u_pdir), otherwise we go + * the directory (in ndp->ni_pdir), otherwise we go * on and lock the inode, being careful with ".". */ if (flag == DELETE && *cp == 0) { @@ -570,72 +631,46 @@ found: */ if (access(dp, IWRITE)) goto bad; - u.u_pdir = dp; /* for dirremove() */ + ndp->ni_pdir = dp; /* for dirremove() */ /* - * Return pointer to current entry in u.u_offset, + * Return pointer to current entry in ndp->ni_offset, * and distance past previous entry (if there - * is a previous entry in this block) in u.u_count. - * Save directory inode pointer in u.u_pdir for dirremove(). + * is a previous entry in this block) in ndp->ni_count. + * Save directory inode pointer in ndp->ni_pdir for dirremove(). */ - if ((u.u_offset&(DIRBLKSIZ-1)) == 0) - u.u_count = 0; + if ((ndp->ni_offset&(DIRBLKSIZ-1)) == 0) + ndp->ni_count = 0; else - u.u_count = u.u_offset - prevoff; + ndp->ni_count = ndp->ni_offset - prevoff; if (lockparent) { - if (dp->i_number == u.u_dent.d_ino) + if (dp->i_number == ndp->ni_dent.d_ino) dp->i_count++; else { - dp = iget(dp->i_dev, fs, u.u_dent.d_ino); + dp = iget(dp->i_dev, fs, ndp->ni_dent.d_ino); if (dp == NULL) { - iput(u.u_pdir); + iput(ndp->ni_pdir); goto bad; } /* - * If directory is setuid, then user must own + * 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 ((u.u_pdir->i_mode & ISUID) && + if ((ndp->ni_pdir->i_mode & ISVTX) && u.u_uid != 0 && - u.u_uid != u.u_pdir->i_uid && + u.u_uid != ndp->ni_pdir->i_uid && dp->i_uid != u.u_uid) { - iput(u.u_pdir); + iput(ndp->ni_pdir); u.u_error = EPERM; goto bad; } } } - brelse(nbp); + FREE(nbp, M_NAMEI); 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. - */ - isdotdot = 0; - if (bcmp(u.u_dent.d_name, "..", 3) == 0) { - isdotdot++; - if (dp == u.u_rdir) - u.u_dent.d_ino = dp->i_number; - else if (u.u_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; - } - } - } - /* * If rewriting (rename), return the inode and the * information required to rewrite the present directory @@ -645,21 +680,21 @@ found: if ((flag == CREATE && lockparent) && *cp == 0) { if (access(dp, IWRITE)) goto bad; - u.u_pdir = dp; /* for dirrewrite() */ + ndp->ni_pdir = dp; /* for dirrewrite() */ /* * Careful about locking second inode. * This can only occur if the target is ".". */ - if (dp->i_number == u.u_dent.d_ino) { + if (dp->i_number == ndp->ni_dent.d_ino) { u.u_error = EISDIR; /* XXX */ goto bad; } - dp = iget(dp->i_dev, fs, u.u_dent.d_ino); + dp = iget(dp->i_dev, fs, ndp->ni_dent.d_ino); if (dp == NULL) { - iput(u.u_pdir); + iput(ndp->ni_pdir); goto bad; } - brelse(nbp); + FREE(nbp, M_NAMEI); return (dp); } @@ -685,65 +720,52 @@ found: */ pdp = dp; if (isdotdot) { - iunlock(pdp); /* race to get the inode */ - dp = iget(dp->i_dev, fs, u.u_dent.d_ino); + IUNLOCK(pdp); /* race to get the inode */ + dp = iget(dp->i_dev, fs, ndp->ni_dent.d_ino); if (dp == NULL) goto bad2; - } else if (dp->i_number == u.u_dent.d_ino) { + } else if (dp->i_number == ndp->ni_dent.d_ino) { dp->i_count++; /* we want ourself, ie "." */ } else { - dp = iget(dp->i_dev, fs, u.u_dent.d_ino); - iunlock(pdp); + 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. + * Insert name into cache if appropriate. */ - if (bcmp(u.u_dent.d_name, ".", 2) != 0 && !isdotdot && docache) { + if (makeentry) { if (ncp != NULL) - panic("nami: duplicating cache"); - - /* - * free the cache slot at head of lru chain - */ + panic("namei: duplicating cache"); + /* + * Free the cache slot at head of lru chain. + */ if (ncp = nchhead) { - /* remove from lru chain */ + /* 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); - - /* drop hold on inode (if we had one) */ - if (ncp->nc_ip) - irele(ncp->nc_ip); - - /* grab the inode we just found */ + remque(ncp); /* remove from old hash chain */ + /* grab the inode we just found */ ncp->nc_ip = dp; - dp->i_count++; - - /* fill in cache info */ + /* 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_nlen = u.u_dent.d_namlen; - bcopy(u.u_dent.d_name, ncp->nc_name, ncp->nc_nlen); - - /* link at end of lru chain */ + ncp->nc_id = dp->i_id; /* identifier */ + ncp->nc_nlen = ndp->ni_dent.d_namlen; + bcopy(ndp->ni_dent.d_name, ncp->nc_name, + (unsigned)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 */ + /* and insert on hash chain */ insque(ncp, nhp); } } @@ -754,21 +776,25 @@ haveino: /* * Check for symbolic link */ - if ((dp->i_mode & IFMT) == IFLNK && (follow || *cp == '/')) { + 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) { + if (dp->i_size + pathlen >= MAXPATHLEN - 1) { + u.u_error = ENAMETOOLONG; + goto bad2; + } + if (++nlink > MAXSYMLINKS) { u.u_error = ELOOP; goto bad2; } - ovbcopy(cp, nbp->b_un.b_addr + dp->i_size, pathlen); + ovbcopy(cp, nbp + dp->i_size, pathlen); u.u_error = - rdwri(UIO_READ, dp, nbp->b_un.b_addr, (int)dp->i_size, - 0, 1, (int *)0); + rdwri(UIO_READ, dp, nbp, (int)dp->i_size, + (off_t)0, 1, (int *)0); if (u.u_error) goto bad2; - cp = nbp->b_un.b_addr; + cp = nbp; iput(dp); if (*cp == '/') { irele(pdp); @@ -776,11 +802,11 @@ haveino: cp++; if ((dp = u.u_rdir) == NULL) dp = rootdir; - ilock(dp); + ILOCK(dp); dp->i_count++; } else { dp = pdp; - ilock(dp); + ILOCK(dp); } fs = dp->i_fs; goto dirloop; @@ -796,9 +822,9 @@ haveino: irele(pdp); goto dirloop; } - brelse(nbp); + FREE(nbp, M_NAMEI); if (lockparent) - u.u_pdir = pdp; + ndp->ni_pdir = pdp; else irele(pdp); return (dp); @@ -809,27 +835,41 @@ bad: brelse(bp); if (dp) iput(dp); - brelse(nbp); + FREE(nbp, M_NAMEI); return (NULL); } -dirbad(ip, how) +dirbad(ip, offset, how) struct inode *ip; + off_t offset; char *how; { printf("%s: bad dir ino %d at offset %d: %s\n", - ip->i_fs->fs_fsmnt, ip->i_number, u.u_offset, how); + ip->i_fs->fs_fsmnt, ip->i_number, offset, how); } -dirbadname(ep) +/* + * Do consistency checking on a directory entry: + * record length must be multiple of 4 + * entry must fit in rest of its DIRBLKSIZ block + * record must be large enough to contain entry + * name is not longer than MAXNAMLEN + * name must be as long as advertised, and null terminated + */ +dirbadentry(ep, entryoffsetinblock) register struct direct *ep; + int entryoffsetinblock; { register int i; + if ((ep->d_reclen & 0x3) != 0 || + ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) || + ep->d_reclen < DIRSIZ(ep) || ep->d_namlen > MAXNAMLEN) + return (1); for (i = 0; i < ep->d_namlen; i++) - if (ep->d_name[i] == 0) + if (ep->d_name[i] == '\0') return (1); return (ep->d_name[i]); } @@ -837,43 +877,48 @@ dirbadname(ep) /* * 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 u.u_pdir is + * 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 - * namei. Remaining parameters (u.u_offset, u.u_count) indicate + * namei. Remaining parameters (ndp->ni_offset, ndp->ni_count) indicate * how the space for the new entry is to be gotten. */ -direnter(ip) +direnter(ip, ndp) struct inode *ip; + register struct nameidata *ndp; { register struct direct *ep, *nep; + register struct inode *dp = ndp->ni_pdir; struct buf *bp; int loc, spacefree, error = 0; u_int dsize; int newentrysize; char *dirbuf; - u.u_dent.d_ino = ip->i_number; - u.u_segflg = 1; - newentrysize = DIRSIZ(&u.u_dent); - if (u.u_count == 0) { + ndp->ni_dent.d_ino = ip->i_number; + newentrysize = DIRSIZ(&ndp->ni_dent); + if (ndp->ni_count == 0) { /* - * If u.u_count is 0, then namei could find no space in the - * directory. In this case u.u_offset will be on a directory + * If ndp->ni_count is 0, then namei could find no space in the + * directory. In this case ndp->ni_offset will be on a directory * block boundary and we will write the new entry into a fresh * block. */ - if (u.u_offset&(DIRBLKSIZ-1)) + if (ndp->ni_offset&(DIRBLKSIZ-1)) panic("wdir: newblk"); - u.u_dent.d_reclen = DIRBLKSIZ; - error = rdwri(UIO_WRITE, u.u_pdir, (caddr_t)&u.u_dent, - newentrysize, u.u_offset, 1, (int *)0); - iput(u.u_pdir); + ndp->ni_dent.d_reclen = DIRBLKSIZ; + error = rdwri(UIO_WRITE, dp, (caddr_t)&ndp->ni_dent, + newentrysize, ndp->ni_offset, 1, (int *)0); + if (DIRBLKSIZ > dp->i_fs->fs_fsize) + panic("wdir: blksize"); /* XXX - should grow w/bmap() */ + else + dp->i_size = roundup(dp->i_size, DIRBLKSIZ); + iput(dp); return (error); } /* - * If u.u_count is non-zero, then namei found space for the - * new entry in the range u.u_offset to u.u_offset+u.u_count. + * If ndp->ni_count is non-zero, then namei found space for the new + * entry in the range ndp->ni_offset to ndp->ni_offset + ndp->ni_count. * in the directory. To use this space, we may have to compact * the entries located there, by copying them together towards * the beginning of the block, leaving the free space in @@ -884,30 +929,30 @@ direnter(ip) * Increase size of directory if entry eats into new space. * This should never push the size past a new multiple of * DIRBLKSIZE. + * + * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN. */ - if (u.u_offset + u.u_count > u.u_pdir->i_size) - u.u_pdir->i_size = u.u_offset + u.u_count; - + 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. */ - bp = blkatoff(u.u_pdir, u.u_offset, (char **)&dirbuf); + bp = blkatoff(dp, ndp->ni_offset, (char **)&dirbuf); if (bp == 0) { - iput(u.u_pdir); + iput(dp); return (u.u_error); } - /* * Find space for the new entry. In the simple case, the * entry at offset base will have the space. If it does * not, then namei arranged that compacting the region - * u.u_offset to u.u_offset+u.u_count would yield the space. + * ndp->ni_offset to ndp->ni_offset+ndp->ni_count would yield the space. */ ep = (struct direct *)dirbuf; dsize = DIRSIZ(ep); spacefree = ep->d_reclen - dsize; - for (loc = ep->d_reclen; loc < u.u_count; ) { + for (loc = ep->d_reclen; loc < ndp->ni_count; ) { nep = (struct direct *)(dirbuf + loc); if (ep->d_ino) { /* trim the existing slot */ @@ -929,26 +974,28 @@ direnter(ip) if (ep->d_ino == 0) { if (spacefree + dsize < newentrysize) panic("wdir: compact1"); - u.u_dent.d_reclen = spacefree + dsize; + ndp->ni_dent.d_reclen = spacefree + dsize; } else { if (spacefree < newentrysize) panic("wdir: compact2"); - u.u_dent.d_reclen = spacefree; + ndp->ni_dent.d_reclen = spacefree; ep->d_reclen = dsize; ep = (struct direct *)((char *)ep + dsize); } - bcopy((caddr_t)&u.u_dent, (caddr_t)ep, (u_int)newentrysize); + bcopy((caddr_t)&ndp->ni_dent, (caddr_t)ep, (u_int)newentrysize); bwrite(bp); - u.u_pdir->i_flag |= IUPD|ICHG; - iput(u.u_pdir); + dp->i_flag |= IUPD|ICHG; + if (ndp->ni_endoff && ndp->ni_endoff < dp->i_size) + itrunc(dp, (u_long)ndp->ni_endoff); + 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 - * u_offset contains the offset into the directory of the - * entry to be eliminated. The u_count field contains the + * 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 * is 0, the first entry is being deleted, so we need only * zero the inode number to mark the entry as free. If the @@ -956,27 +1003,28 @@ direnter(ip) * the space of the now empty record by adding the record size * to the size of the previous entry. */ -dirremove() +dirremove(ndp) + register struct nameidata *ndp; { - register struct inode *dp = u.u_pdir; + register struct inode *dp = ndp->ni_pdir; register struct buf *bp; struct direct *ep; - if (u.u_count == 0) { + if (ndp->ni_count == 0) { /* * First entry in block: set d_ino to zero. */ - u.u_dent.d_ino = 0; - (void) rdwri(UIO_WRITE, dp, (caddr_t)&u.u_dent, - (int)DIRSIZ(&u.u_dent), u.u_offset, 1, (int *)0); + 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); } else { /* * Collapse new free space into previous entry. */ - bp = blkatoff(dp, (int)(u.u_offset - u.u_count), (char **)&ep); + bp = blkatoff(dp, ndp->ni_offset - ndp->ni_count, (char **)&ep); if (bp == 0) return (0); - ep->d_reclen += u.u_dent.d_reclen; + ep->d_reclen += ndp->ni_dent.d_reclen; bwrite(bp); dp->i_flag |= IUPD|ICHG; } @@ -988,13 +1036,14 @@ dirremove() * supplied. The parameters describing the directory entry are * set up by a call to namei. */ -dirrewrite(dp, ip) +dirrewrite(dp, ip, ndp) struct inode *dp, *ip; + struct nameidata *ndp; { - u.u_dent.d_ino = ip->i_number; - u.u_error = rdwri(UIO_WRITE, dp, (caddr_t)&u.u_dent, - (int)DIRSIZ(&u.u_dent), u.u_offset, 1, (int *)0); + 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); } @@ -1012,20 +1061,28 @@ blkatoff(ip, offset, res) { 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; + daddr_t bn; + bn = bmap(ip, lbn, B_READ, bsize); if (u.u_error) return (0); - bp = bread(ip->i_dev, bn, bsize); + if (bn == (daddr_t)-1) { + dirbad(ip, offset, "hole in dir"); + return (0); + } +#ifdef SECSIZE + bp = bread(ip->i_dev, fsbtodb(fs, bn), bsize, fs->fs_dbsize); +#else SECSIZE + bp = bread(ip->i_dev, fsbtodb(fs, bn), bsize); +#endif SECSIZE if (bp->b_flags & B_ERROR) { brelse(bp); return (0); } if (res) - *res = bp->b_un.b_addr + base; + *res = bp->b_un.b_addr + blkoff(fs, offset); return (bp); } @@ -1038,8 +1095,9 @@ blkatoff(ip, offset, res) * * NB: does not handle corrupted directories. */ -dirempty(ip) +dirempty(ip, parentino) register struct inode *ip; + ino_t parentino; { register off_t off; struct dirtemplate dbuf; @@ -1056,6 +1114,9 @@ dirempty(ip) */ 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; @@ -1069,7 +1130,9 @@ dirempty(ip) * 1 implies ".", 2 implies ".." if second * char is also "." */ - if (dp->d_namlen == 1 || dp->d_name[1] == '.') + if (dp->d_namlen == 1) + continue; + if (dp->d_name[1] == '.' && dp->d_ino == parentino) continue; return (0); } @@ -1106,7 +1169,8 @@ checkpath(source, target) if (error != 0) break; if (dirbuf.dotdot_namlen != 2 || - bcmp(dirbuf.dotdot_name, "..", 3) != 0) { + dirbuf.dotdot_name[0] != '.' || + dirbuf.dotdot_name[1] != '.') { error = ENOTDIR; break; } @@ -1138,23 +1202,19 @@ out: nchinit() { register union nchash *nchp; - register struct nch *ncp; + register struct namecache *ncp; nchhead = 0; nchtail = &nchhead; - - for (ncp = nch; ncp < &nch[nchsize]; ncp++) { + for (ncp = namecache; ncp < &namecache[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; @@ -1172,42 +1232,45 @@ nchinit() nchinval(dev) register dev_t dev; { - register struct nch *ncp, *nxtcp; + register struct namecache *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; - - /* remove the entry from its hash chain */ - remque(ncp); - /* and make a dummy one */ - ncp->nc_forw = ncp; + ncp->nc_ip = NULL; + remque(ncp); /* remove entry from its hash chain */ + ncp->nc_forw = ncp; /* and make a dummy one */ ncp->nc_back = ncp; - - /* delete this entry from LRU chain */ + /* delete this entry from LRU chain */ *ncp->nc_prev = nxtcp; if (nxtcp) nxtcp->nc_prev = ncp->nc_prev; else nchtail = ncp->nc_prev; - - /* free the inode we had */ - irele(ncp->nc_ip); - ncp->nc_ip = NULL; - - /* cause rescan of list, it may have altered */ + /* cause rescan of list, it may have altered */ nxtcp = nchhead; - /* put the now-free entry at head of LRU */ + /* 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 namecache *ncp; + + for (ncp = namecache; ncp < &namecache[nchsize]; ncp++) + ncp->nc_id = 0; +}