2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
10 /* I/O operations for the Unix open by name (namei) interface. */
12 #include <afsconfig.h>
13 #include <afs/param.h>
25 #include <afs/assert.h>
28 #include <sys/param.h>
31 #include <sys/lockf.h>
33 #if defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
36 #include <afs/afsutil.h>
39 #include <afs/afsint.h>
43 #include "viceinode.h"
45 #include "partition.h"
46 #include <afs/errors.h>
48 /*@+fcnmacros +macrofcndecl@*/
51 extern off64_t afs_lseek(int FD, off64_t O, int F);
52 #endif /*S_SPLINT_S */
53 #define afs_lseek(FD, O, F) lseek64(FD, (off64_t)(O), F)
54 #define afs_stat stat64
55 #define afs_fstat fstat64
56 #define afs_open open64
57 #define afs_fopen fopen64
58 #else /* !O_LARGEFILE */
60 extern off_t afs_lseek(int FD, off_t O, int F);
61 #endif /*S_SPLINT_S */
62 #define afs_lseek(FD, O, F) lseek(FD, (off_t)(O), F)
64 #define afs_fstat fstat
66 #define afs_fopen fopen
67 #endif /* !O_LARGEFILE */
68 /*@=fcnmacros =macrofcndecl@*/
70 /*@printflike@*/ extern void Log(const char *format, ...);
72 extern char *volutil_PartitionName_r(int volid, char *buf, int buflen);
77 namei_iread(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
86 if (FDH_SEEK(fdP, offset, SEEK_SET) < 0) {
91 nBytes = FDH_READ(fdP, buf, size);
97 namei_iwrite(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
106 if (FDH_SEEK(fdP, offset, SEEK_SET) < 0) {
107 FDH_REALLYCLOSE(fdP);
110 nBytes = FDH_WRITE(fdP, buf, size);
117 /* Inode number format:
118 * low 26 bits - vnode number - all 1's if volume special file.
120 * next 3 bits spare (0's)
121 * high 32 bits - uniquifier (regular) or type if spare
123 #define NAMEI_VNODEMASK 0x003ffffff
124 #define NAMEI_TAGMASK 0x7
125 #define NAMEI_TAGSHIFT 26
126 #define NAMEI_UNIQMASK 0xffffffff
127 #define NAMEI_UNIQSHIFT 32
128 #define NAMEI_INODESPECIAL ((Inode)NAMEI_VNODEMASK)
129 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
131 /* dir1 is the high 8 bits of the 26 bit vnode */
132 #define VNO_DIR1(vno) ((vno >> 14) & 0xff)
133 /* dir2 is the next 9 bits */
134 #define VNO_DIR2(vno) ((vno >> 9) & 0x1ff)
135 /* "name" is the low 9 bits of the vnode, the 3 bit tag and the uniq */
137 #define NAMEI_SPECDIR "special"
138 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
140 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
141 * group, not counting temporary (move) volumes.
142 * This is the number of separate files, all having
143 * the same vnode number, which can occur in a volume
154 static int namei_GetLinkCount2(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite);
156 static int GetFreeTag(IHandle_t * ih, int vno);
158 /* namei_HandleToInodeDir
160 * Construct the path name of the directory holding the inode data.
161 * Format: /<vicepx>/INODEDIR
164 #define PNAME_BLEN 64
166 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
168 char *tmp = name->n_base;
170 memset(name, '\0', sizeof(*name));
172 (void)volutil_PartitionName_r(ih->ih_dev, tmp, NAMEI_LCOMP_LEN);
173 tmp += VICE_PREFIX_SIZE;
174 tmp += ih->ih_dev > 25 ? 2 : 1;
177 (void)strcpy(tmp, INODEDIR);
178 (void)strcpy(name->n_path, name->n_base);
181 #define addtoname(N, C) \
183 strcat((N)->n_path, "/"); strcat((N)->n_path, C); \
188 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
192 namei_HandleToInodeDir(name, ih);
193 (void)int32_to_flipbase64(tmp, (int64_t) (ih->ih_vid & 0xff));
194 (void)strcpy(name->n_voldir1, tmp);
195 addtoname(name, name->n_voldir1);
196 (void)int32_to_flipbase64(tmp, (int64_t) ih->ih_vid);
197 (void)strcpy(name->n_voldir2, tmp);
198 addtoname(name, name->n_voldir2);
201 /* namei_HandleToName
203 * Constructs a file name for the fully qualified handle.
204 * Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special
207 namei_HandleToName(namei_t * name, IHandle_t * ih)
210 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
212 namei_HandleToVolDir(name, ih);
214 if (vno == NAMEI_VNODESPECIAL) {
215 (void)strcpy(name->n_dir1, NAMEI_SPECDIR);
216 addtoname(name, name->n_dir1);
217 name->n_dir2[0] = '\0';
219 (void)int32_to_flipbase64(str, VNO_DIR1(vno));
220 (void)strcpy(name->n_dir1, str);
221 addtoname(name, name->n_dir1);
222 (void)int32_to_flipbase64(str, VNO_DIR2(vno));
223 (void)strcpy(name->n_dir2, str);
224 addtoname(name, name->n_dir2);
226 (void)int64_to_flipbase64(str, (int64_t) ih->ih_ino);
227 (void)strcpy(name->n_inode, str);
228 addtoname(name, name->n_inode);
231 /* The following is a warning to tell sys-admins to not muck about in this
234 #define VICE_README "These files and directories are a part of the AFS \
235 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
236 ownership and permissions included.\n"
238 namei_ViceREADME(char *partition)
243 /* Create the inode directory if we're starting for the first time */
244 (void)afs_snprintf(filename, sizeof filename, "%s/%s", partition,
246 mkdir(filename, 0700);
248 (void)afs_snprintf(filename, sizeof filename, "%s/%s/README", partition,
250 fd = afs_open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0444);
252 (void)write(fd, VICE_README, strlen(VICE_README));
259 #define create_dir() \
261 if (mkdir(tmp, 0700)<0) { \
262 if (errno != EEXIST) \
270 #define create_nextdir(A) \
272 strcat(tmp, "/"); strcat(tmp, A); create_dir(); \
275 /* namei_CreateDataDirectories
277 * If creating the file failed because of ENOENT or ENOTDIR, try
278 * creating all the directories first.
281 namei_CreateDataDirectories(namei_t * name, int *created)
287 (void)strcpy(tmp, name->n_base);
290 create_nextdir(name->n_voldir1);
291 create_nextdir(name->n_voldir2);
292 create_nextdir(name->n_dir1);
293 if (name->n_dir2[0]) {
294 create_nextdir(name->n_dir2);
299 /* delTree(): Deletes an entire tree of directories (no files)
301 * root : Full path to the subtree. Should be big enough for PATH_MAX
302 * tree : the subtree to be deleted is rooted here. Specifies only the
303 * subtree beginning at tree (not the entire path). It should be
304 * a pointer into the "root" buffer.
306 * errp : errno of the first error encountered during the directory cleanup.
307 * *errp should have been initialized to 0.
310 * -1 : If errors were encountered during cleanup and error is set to
314 * If there are errors, we try to work around them and delete as many
315 * directories as possible. We don't attempt to remove directories that still
316 * have non-dir entries in them.
319 delTree(char *root, char *tree, int *errp)
327 /* delete the children first */
328 cp = strchr(tree, '/');
330 delTree(root, cp + 1, errp);
333 cp = tree + strlen(tree); /* move cp to the end of string tree */
335 /* now delete all entries in this dir */
336 if ((ds = opendir(root)) != (DIR *) NULL) {
338 while ((dirp = readdir(ds))) {
339 /* ignore . and .. */
340 if (!strcmp(dirp->d_name, ".") || !strcmp(dirp->d_name, ".."))
342 /* since root is big enough, we reuse the space to
343 * concatenate the dirname to the current tree
346 strcat(root, dirp->d_name);
347 if (afs_stat(root, &st) == 0 && S_ISDIR(st.st_mode)) {
348 /* delete this subtree */
349 delTree(root, cp + 1, errp);
351 *errp = *errp ? *errp : errno;
353 /* recover path to our cur tree by truncating it to
358 /* if (!errno) -- closedir not implicit if we got an error */
362 /* finally axe the current dir */
364 *errp = *errp ? *errp : errno;
366 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
368 #endif /* !AFS_PTHREAD_ENV */
373 /* if we encountered errors during cleanup, we return a -1 */
381 /* namei_RemoveDataDirectories
383 * Returns 0 on success.
384 * Returns -1 on error. Typically, callers ignore this error bcause we
385 * can continue running if the removes fail. The salvage process will
386 * finish tidying up for us. We only use the n_base and n_voldir1 entries
387 * and only do rmdir's.
391 namei_RemoveDataDirectories(namei_t * name)
393 char pbuf[MAXPATHLEN], *path = pbuf;
394 int prefixlen = strlen(name->n_base), err = 0;
396 strcpy(path, name->n_path);
398 /* move past the prefix */
399 path = path + prefixlen + 1; /* skip over the trailing / */
401 /* now delete all dirs upto path */
402 return delTree(pbuf, path, &err);
406 /* Create the file in the name space.
408 * Parameters stored as follows:
410 * p1 - volid - implied in containing directory.
411 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
412 * p3 - uniq -- bits 4-0 are in mode bits 4-0
413 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
415 * p1 - volid - creation time - dwHighDateTime
416 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
417 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
418 * p4 - parid - parent volume id - implied in containing directory.
420 * Return value is the inode number or (Inode)-1 if error.
421 * We "know" there is only one link table, so return EEXIST if there already
422 * is a link table. It's up to the calling code to test errno and increment
428 * This function is called by VCreateVolume to hide the implementation
429 * details of the inode numbers. This only allows for 7 volume special
430 * types, but if we get that far, this could should be dead by then.
433 namei_MakeSpecIno(int volid, int type)
436 ino = NAMEI_INODESPECIAL;
437 type &= NAMEI_TAGMASK;
438 ino |= ((Inode) type) << NAMEI_TAGSHIFT;
439 ino |= ((Inode) volid) << NAMEI_UNIQSHIFT;
443 /* SetOGM - set owner group and mode bits from parm and tag
445 * owner - low 15 bits of parm.
446 * group - next 15 bits of parm.
447 * mode - 2 bits of parm, then lowest = 3 bits of tag.
450 SetOGM(int fd, int parm, int tag)
452 int owner, group, mode;
454 owner = parm & 0x7fff;
455 group = (parm >> 15) & 0x7fff;
456 if (fchown(fd, owner, group) < 0)
459 mode = (parm >> 27) & 0x18;
461 if (fchmod(fd, mode) < 0)
468 /* GetOGM - get parm and tag from owner, group and mode bits. */
470 GetOGMFromStat(struct afs_stat *status, int *parm, int *tag)
472 *parm = status->st_uid | (status->st_gid << 15);
473 *parm |= (status->st_mode & 0x18) << 27;
474 *tag = status->st_mode & 0x7;
478 GetOGM(int fd, int *parm, int *tag)
480 struct afs_stat status;
481 if (afs_fstat(fd, &status) < 0)
484 GetOGMFromStat(&status, parm, tag);
488 int big_vno = 0; /* Just in case we ever do 64 bit vnodes. */
490 /* Derive the name and create it O_EXCL. If that fails we have an error.
491 * Get the tag from a free column in the link table.
494 namei_icreate(IHandle_t * lh, char *part, int p1, int p2, int p3, int p4)
507 memset((void *)&tmp, 0, sizeof(IHandle_t));
510 tmp.ih_dev = volutil_GetPartitionID(part);
511 if (tmp.ih_dev == -1) {
517 /* Parameters for special file:
518 * p1 - volume id - goes into owner/group/mode
521 * p4 - parent volume id
526 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
527 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
529 int vno = p2 & NAMEI_VNODEMASK;
530 /* Parameters for regular file:
542 /* If GetFreeTag succeeds, it atomically sets link count to 1. */
543 tag = GetFreeTag(lh, p2);
547 /* name is <uniq(p3)><tag><vno(p2)> */
549 tmp.ih_ino = (Inode) p2;
550 tmp.ih_ino |= ((Inode) tag) << NAMEI_TAGSHIFT;
551 tmp.ih_ino |= ((Inode) p3) << NAMEI_UNIQSHIFT;
556 namei_HandleToName(&name, &tmp);
557 fd = afs_open(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
559 if (errno == ENOTDIR || errno == ENOENT) {
560 if (namei_CreateDataDirectories(&name, &created_dir) < 0)
562 fd = afs_open(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR,
570 if (SetOGM(fd, ogm_parm, tag) < 0) {
576 if (p2 == -1 && p3 == VI_LINKTABLE) {
577 /* hack at tmp to setup for set link count call. */
578 memset((void *)&tfd, 0, sizeof(FdHandle_t)); /* minimalistic still, but a little cleaner */
581 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
589 if (code || (fd < 0)) {
593 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
598 return (code || (fd < 0)) ? (Inode) - 1 : tmp.ih_ino;
604 namei_iopen(IHandle_t * h)
609 /* Convert handle to file name. */
610 namei_HandleToName(&name, h);
611 fd = afs_open(name.n_path, O_RDWR, 0666);
615 /* Need to detect vol special file and just unlink. In those cases, the
616 * handle passed in _is_ for the inode. We only check p1 for the special
620 namei_dec(IHandle_t * ih, Inode ino, int p1)
627 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
630 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
632 /* Verify this is the right file. */
633 IH_INIT(tmp, ih->ih_dev, ih->ih_vid, ino);
642 if ((GetOGM(fdP->fd_fd, &inode_p1, &tag) < 0) || (inode_p1 != p1)) {
643 FDH_REALLYCLOSE(fdP);
649 /* If it's the link table itself, decrement the link count. */
650 if (type == VI_LINKTABLE) {
651 if ((count = namei_GetLinkCount(fdP, (Inode) 0, 1)) < 0) {
652 FDH_REALLYCLOSE(fdP);
658 if (namei_SetLinkCount(fdP, (Inode) 0, count < 0 ? 0 : count, 1) <
660 FDH_REALLYCLOSE(fdP);
666 FDH_REALLYCLOSE(fdP);
672 namei_HandleToName(&name, tmp);
673 if ((code = unlink(name.n_path)) == 0) {
674 if (type == VI_LINKTABLE) {
675 /* Try to remove directory. If it fails, that's ok.
676 * Salvage will clean up.
678 (void)namei_RemoveDataDirectories(&name);
681 FDH_REALLYCLOSE(fdP);
684 /* Get a file descriptor handle for this Inode */
690 if ((count = namei_GetLinkCount(fdP, ino, 1)) < 0) {
691 FDH_REALLYCLOSE(fdP);
697 if (namei_SetLinkCount(fdP, ino, count, 1) < 0) {
698 FDH_REALLYCLOSE(fdP);
703 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
704 Log("Warning: Lost ref on ihandle dev %d vid %d ino %lld\n",
705 th->ih_dev, th->ih_vid, (int64_t) th->ih_ino);
708 /* If we're less than 0, someone presumably unlinked;
709 don't bother setting count to 0, but we need to drop a lock */
710 if (namei_SetLinkCount(fdP, ino, 0, 1) < 0) {
711 FDH_REALLYCLOSE(fdP);
717 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
719 namei_HandleToName(&name, th);
721 code = unlink(name.n_path);
730 namei_inc(IHandle_t * h, Inode ino, int p1)
736 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
737 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
738 if (type != VI_LINKTABLE)
743 /* Get a file descriptor handle for this Inode */
749 if ((count = namei_GetLinkCount(fdP, ino, 1)) < 0)
758 if (namei_SetLinkCount(fdP, ino, count, 1) < 0)
762 FDH_REALLYCLOSE(fdP);
769 /************************************************************************
770 * File Name Structure
771 ************************************************************************
773 * Each AFS file needs a unique name and it needs to be findable with
774 * minimal lookup time. Note that the constraint on the number of files and
775 * directories in a volume is the size of the vnode index files and the
776 * max file size AFS supports (for internal files) of 2^31. Since a record
777 * in the small vnode index file is 64 bytes long, we can have at most
778 * (2^31)/64 or 33554432 files. A record in the large index file is
779 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
780 * Another layout parameter is that there is roughly a 16 to 1 ratio between
781 * the number of files and the number of directories.
783 * Using this information we can see that a layout of 256 directories, each
784 * with 512 subdirectories and each of those having 512 files gives us
785 * 256*512*512 = 67108864 AFS files and directories.
787 * The volume, vnode, uniquifier and data version, as well as the tag
788 * are required, either for finding the file or for salvaging. It's best to
789 * restrict the name to something that can be mapped into 64 bits so the
790 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
791 * is used to distinguish between different versions of the same file
792 * which are currently in the RW and clones of a volume. See "Link Table
793 * Organization" below for more information on the tag. The tag is
794 * required in the name of the file to ensure a unique name.
796 * We can store data in the uid, gid and mode bits of the files, provided
797 * the directories have root only access. This gives us 15 bits for each
798 * of uid and gid (GNU chown considers 65535 to mean "don't change").
799 * There are 9 available mode bits. Adn we need to store a total of
800 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
801 * or 131 bits somewhere.
803 * The format of a file name for a regular file is:
804 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
805 * V1 - low 8 bits of RW volume id
806 * V2 - all bits of RW volume id
807 * AA - high 8 bits of vnode number.
808 * BB - next 9 bits of vnode number.
809 * <tag><uniq><vno> - file name
811 * Volume special files are stored in a separate directory:
812 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
815 * The vnode is hashed into the directory using the high bits of the
816 * vnode number. This is so that consecutively created vnodes are in
817 * roughly the same area on the disk. This will at least be optimal if
818 * the user is creating many files in the same AFS directory. The name
819 * should be formed so that the leading characters are different as quickly
820 * as possible, leading to faster discards of incorrect matches in the
826 /************************************************************************
827 * Link Table Organization
828 ************************************************************************
830 * The link table volume special file is used to hold the link counts that
831 * are held in the inodes in inode based AFS vice filesystems. For user
832 * space access, the link counts are being kept in a separate
833 * volume special file. The file begins with the usual version stamp
834 * information and is then followed by one row per vnode number. vnode 0
835 * is used to hold the link count of the link table itself. That is because
836 * the same link table is shared among all the volumes of the volume group
837 * and is deleted only when the last volume of a volume group is deleted.
839 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
840 * link count from 0 through 7. Each colume represents a unique instance of
841 * that vnode. Say we have a file shared between the RW and a RO and a
842 * different version of the file (or a different uniquifer) for the BU volume.
843 * Then one column would be holding the link count of 2 for the RW and RO
844 * and a different column would hold the link count of 1 for the BU volume.
845 * Note that we allow only 5 volumes per file, giving 15 bits used in the
848 #define LINKTABLE_WIDTH 2
849 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
852 namei_GetLCOffsetAndIndexFromIno(Inode ino, afs_foff_t * offset, int *index)
854 int toff = (int)(ino & NAMEI_VNODEMASK);
855 int tindex = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
857 *offset = (afs_foff_t) ((toff << LINKTABLE_SHIFT) + 8); /* * 2 + sizeof stamp */
858 *index = (tindex << 1) + tindex;
862 /* namei_GetLinkCount
863 * If lockit is set, lock the file and leave it locked upon a successful
867 namei_GetLinkCount2(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite)
869 unsigned short row = 0;
874 if (!VALID_INO(ino)) return 0;
875 /* there's no linktable yet. the salvager will create one later */
876 if (h->fd_fd == -1 && fixup)
878 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
881 #if defined(AFS_AIX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
882 if (lockf(h->fd_fd, F_LOCK, 0) < 0)
884 if (flock(h->fd_fd, LOCK_EX) < 0)
889 if (afs_lseek(h->fd_fd, offset, SEEK_SET) == -1)
890 goto bad_getLinkByte;
892 rc = read(h->fd_fd, (char *)&row, sizeof(row));
893 if ((rc == 0 || !((row >> index) & NAMEI_TAGMASK)) && fixup && nowrite)
895 if (rc == 0 && fixup) {
897 if (fstat(h->fd_fd, &st) || st.st_size >= offset+sizeof(row))
898 goto bad_getLinkByte;
899 FDH_TRUNC(h, offset+sizeof(row));
901 rc = write(h->fd_fd, (char *)&row, sizeof(row));
903 if (rc != sizeof(row)) {
904 goto bad_getLinkByte;
907 if (fixup && !((row >> index) & NAMEI_TAGMASK)) {
909 if (afs_lseek(h->fd_fd, offset, SEEK_SET) == -1)
910 goto bad_getLinkByte;
911 rc = write(h->fd_fd, (char *)&row, sizeof(row));
912 if (rc != sizeof(row))
913 goto bad_getLinkByte;
916 return (int)((row >> index) & NAMEI_TAGMASK);
920 #if defined(AFS_AIX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
921 lockf(h->fd_fd, F_ULOCK, 0);
923 flock(h->fd_fd, LOCK_UN);
929 namei_GetLinkCount(FdHandle_t * h, Inode ino, int lockit)
931 return namei_GetLinkCount2(h, ino, lockit, 0, 1);
934 /* Return a free column index for this vnode. */
936 GetFreeTag(IHandle_t * ih, int vno)
950 /* Only one manipulates at a time. */
951 #if defined(AFS_AIX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
952 if (lockf(fdP->fd_fd, F_LOCK, 0) < 0) {
954 if (flock(fdP->fd_fd, LOCK_EX) < 0) {
956 FDH_REALLYCLOSE(fdP);
960 offset = (vno << LINKTABLE_SHIFT) + 8; /* * 2 + sizeof stamp */
961 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
965 code = read(fdP->fd_fd, (char *)&row, sizeof(row));
966 if (code != sizeof(row)) {
972 /* Now find a free column in this row and claim it. */
973 for (col = 0; col < NAMEI_MAXVOLS; col++) {
974 coldata = 7 << (col * 3);
975 if ((row & coldata) == 0)
978 if (col >= NAMEI_MAXVOLS)
981 coldata = 1 << (col * 3);
984 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
987 if (write(fdP->fd_fd, (char *)&row, sizeof(row)) != sizeof(row)) {
991 #if defined(AFS_AIX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
992 lockf(fdP->fd_fd, F_ULOCK, 0);
994 flock(fdP->fd_fd, LOCK_UN);
996 FDH_REALLYCLOSE(fdP);
1000 #if defined(AFS_AIX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
1001 lockf(fdP->fd_fd, F_ULOCK, 0);
1003 flock(fdP->fd_fd, LOCK_UN);
1005 FDH_REALLYCLOSE(fdP);
1011 /* namei_SetLinkCount
1012 * If locked is set, assume file is locked. Otherwise, lock file before
1013 * proceeding to modify it.
1016 namei_SetLinkCount(FdHandle_t * fdP, Inode ino, int count, int locked)
1024 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1027 #if defined(AFS_AIX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
1028 if (lockf(fdP->fd_fd, F_LOCK, 0) < 0) {
1030 if (flock(fdP->fd_fd, LOCK_EX) < 0) {
1035 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
1037 goto bad_SetLinkCount;
1041 code = read(fdP->fd_fd, (char *)&row, sizeof(row));
1042 if (code != sizeof(row)) {
1045 goto bad_SetLinkCount;
1052 row &= (unsigned short)~junk;
1053 row |= (unsigned short)count;
1055 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
1057 goto bad_SetLinkCount;
1060 if (write(fdP->fd_fd, (char *)&row, sizeof(short)) != sizeof(short)) {
1062 goto bad_SetLinkCount;
1070 #if defined(AFS_AIX_ENV) || defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
1071 lockf(fdP->fd_fd, F_ULOCK, 0);
1073 flock(fdP->fd_fd, LOCK_UN);
1080 /* ListViceInodes - write inode data to a results file. */
1081 static int DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
1083 static int DecodeVolumeName(char *name, int *vid);
1084 static int namei_ListAFSSubDirs(IHandle_t * dirIH,
1085 int (*write_fun) (FILE *,
1086 struct ViceInodeInfo *,
1087 char *, char *), FILE * fp,
1088 int (*judgeFun) (struct ViceInodeInfo *,
1090 int singleVolumeNumber, void *rock);
1095 * Write the inode data to the results file.
1097 * Returns -2 on error, 0 on success.
1099 * This is written as a callback simply so that other listing routines
1100 * can use the same inode reading code.
1103 WriteInodeInfo(FILE * fp, struct ViceInodeInfo *info, char *dir, char *name)
1106 n = fwrite(info, sizeof(*info), 1, fp);
1107 return (n == 1) ? 0 : -2;
1111 int mode_errors; /* Number of errors found in mode bits on directories. */
1113 VerifyDirPerms(char *path)
1115 struct afs_stat status;
1117 if (afs_stat(path, &status) < 0) {
1118 Log("Unable to stat %s. Please manually verify mode bits for this"
1119 " directory\n", path);
1121 if (((status.st_mode & 0777) != 0700) || (status.st_uid != 0))
1127 * Fill the results file with the requested inode information.
1131 * -1 - complete failure, salvage should terminate.
1132 * -2 - not enough space on partition, salvager has error message for this.
1134 * This code optimizes single volume salvages by just looking at that one
1135 * volume's directory.
1137 * If the resultFile is NULL, then don't call the write routine.
1140 ListViceInodes(char *devname, char *mountedOn, char *resultFile,
1141 int (*judgeInode) (struct ViceInodeInfo * info, int vid, void *rock),
1142 int singleVolumeNumber, int *forcep, int forceR, char *wpath,
1145 FILE *fp = (FILE *) - 1;
1147 struct afs_stat status;
1150 fp = afs_fopen(resultFile, "w");
1152 Log("Unable to create inode description file %s\n", resultFile);
1157 /* Verify protections on directories. */
1159 VerifyDirPerms(mountedOn);
1162 namei_ListAFSFiles(mountedOn, WriteInodeInfo, fp, judgeInode,
1163 singleVolumeNumber, rock);
1173 if (fflush(fp) == EOF) {
1174 Log("Unable to successfully flush inode file for %s\n", mountedOn);
1178 if (fsync(fileno(fp)) == -1) {
1179 Log("Unable to successfully fsync inode file for %s\n", mountedOn);
1183 if (fclose(fp) == EOF) {
1184 Log("Unable to successfully close inode file for %s\n", mountedOn);
1189 * Paranoia: check that the file is really the right size
1191 if (afs_stat(resultFile, &status) == -1) {
1192 Log("Unable to successfully stat inode file for %s\n", mountedOn);
1195 if (status.st_size != ninodes * sizeof(struct ViceInodeInfo)) {
1196 Log("Wrong size (%d instead of %d) in inode file for %s\n",
1197 status.st_size, ninodes * sizeof(struct ViceInodeInfo),
1205 /* namei_ListAFSFiles
1207 * Collect all the matching AFS files on the drive.
1208 * If singleVolumeNumber is non-zero, just return files for that volume.
1210 * Returns <0 on error, else number of files found to match.
1213 namei_ListAFSFiles(char *dev,
1214 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1216 int (*judgeFun) (struct ViceInodeInfo *, int, void *),
1217 int singleVolumeNumber, void *rock)
1223 struct dirent *dp1, *dp2;
1226 static void FreeZLCList(void);
1229 memset((void *)&ih, 0, sizeof(IHandle_t));
1230 ih.ih_dev = volutil_GetPartitionID(dev);
1232 if (singleVolumeNumber) {
1233 ih.ih_vid = singleVolumeNumber;
1234 namei_HandleToVolDir(&name, &ih);
1236 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1237 singleVolumeNumber, rock);
1241 /* Find all volume data directories and descend through them. */
1242 namei_HandleToInodeDir(&name, &ih);
1244 dirp1 = opendir(name.n_path);
1247 while ((dp1 = readdir(dirp1))) {
1248 if (*dp1->d_name == '.')
1250 (void)strcpy(path2, name.n_path);
1251 (void)strcat(path2, "/");
1252 (void)strcat(path2, dp1->d_name);
1253 dirp2 = opendir(path2);
1255 while ((dp2 = readdir(dirp2))) {
1256 if (*dp2->d_name == '.')
1258 if (!DecodeVolumeName(dp2->d_name, &ih.ih_vid)) {
1260 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1277 /* namei_ListAFSSubDirs
1282 * > = 0 - number of AFS files found.
1285 namei_ListAFSSubDirs(IHandle_t * dirIH,
1286 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1288 int (*judgeFun) (struct ViceInodeInfo *, int, void *),
1289 int singleVolumeNumber, void *rock)
1291 IHandle_t myIH = *dirIH;
1293 char path1[512], path2[512], path3[512];
1294 DIR *dirp1, *dirp2, *dirp3;
1295 struct dirent *dp1, *dp2, *dp3;
1296 struct ViceInodeInfo info;
1297 FdHandle_t linkHandle;
1301 static void AddToZLCDeleteList(char dir, char *name);
1302 static void DeleteZLCFiles(char *path);
1305 namei_HandleToVolDir(&name, &myIH);
1306 (void)strcpy(path1, name.n_path);
1308 /* Do the directory containing the special files first to pick up link
1311 (void)strcat(path1, "/");
1312 (void)strcat(path1, NAMEI_SPECDIR);
1314 linkHandle.fd_fd = -1;
1315 dirp1 = opendir(path1);
1317 while ((dp1 = readdir(dirp1))) {
1318 if (*dp1->d_name == '.')
1320 if (DecodeInode(path1, dp1->d_name, &info, myIH.ih_vid) < 0)
1322 if (info.u.param[2] != VI_LINKTABLE) {
1325 /* Open this handle */
1326 (void)afs_snprintf(path2, sizeof path2, "%s/%s", path1,
1328 linkHandle.fd_fd = afs_open(path2, Testing ? O_RDONLY : O_RDWR, 0666);
1330 namei_GetLinkCount2(&linkHandle, (Inode) 0, 1, 1, Testing);
1332 if (judgeFun && !(*judgeFun) (&info, singleVolumeNumber, rock))
1335 if ((*writeFun) (fp, &info, path1, dp1->d_name) < 0) {
1336 if (linkHandle.fd_fd >= 0)
1337 close(linkHandle.fd_fd);
1346 /* Now run through all the other subdirs */
1347 namei_HandleToVolDir(&name, &myIH);
1348 (void)strcpy(path1, name.n_path);
1350 dirp1 = opendir(path1);
1352 while ((dp1 = readdir(dirp1))) {
1353 if (*dp1->d_name == '.')
1355 if (!strcmp(dp1->d_name, NAMEI_SPECDIR))
1358 /* Now we've got a next level subdir. */
1359 (void)strcpy(path2, path1);
1360 (void)strcat(path2, "/");
1361 (void)strcat(path2, dp1->d_name);
1362 dirp2 = opendir(path2);
1364 while ((dp2 = readdir(dirp2))) {
1365 if (*dp2->d_name == '.')
1368 /* Now we've got to the actual data */
1369 (void)strcpy(path3, path2);
1370 (void)strcat(path3, "/");
1371 (void)strcat(path3, dp2->d_name);
1372 dirp3 = opendir(path3);
1374 while ((dp3 = readdir(dirp3))) {
1375 if (*dp3->d_name == '.')
1378 (path3, dp3->d_name, &info, myIH.ih_vid) < 0)
1381 namei_GetLinkCount2(&linkHandle,
1382 info.inodeNumber, 1, 1, Testing);
1383 if (info.linkCount == 0) {
1385 Log("Found 0 link count file %s/%s, deleting it.\n", path3, dp3->d_name);
1386 AddToZLCDeleteList((char)i, dp3->d_name);
1388 Log("Found 0 link count file %s/%s.\n", path3,
1394 && !(*judgeFun) (&info, singleVolumeNumber, rock))
1397 if ((*writeFun) (fp, &info, path3, dp3->d_name) <
1399 close(linkHandle.fd_fd);
1416 if (linkHandle.fd_fd >= 0)
1417 close(linkHandle.fd_fd);
1419 /* Then why does this directory exist? Blow it away. */
1420 namei_HandleToVolDir(&name, dirIH);
1421 namei_RemoveDataDirectories(&name);
1428 DecodeVolumeName(char *name, int *vid)
1430 if (strlen(name) <= 2)
1432 *vid = (int)flipbase64_to_int64(name);
1439 * Get the inode number from the name.
1443 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info, int volid)
1446 struct afs_stat status;
1449 (void)strcpy(fpath, dpath);
1450 (void)strcat(fpath, "/");
1451 (void)strcat(fpath, name);
1453 if (afs_stat(fpath, &status) < 0) {
1457 info->byteCount = status.st_size;
1458 info->inodeNumber = (Inode) flipbase64_to_int64(name);
1460 GetOGMFromStat(&status, &parm, &tag);
1461 if ((info->inodeNumber & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1462 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
1463 info->u.param[0] = parm;
1464 info->u.param[1] = -1;
1465 info->u.param[2] = tag;
1466 info->u.param[3] = volid;
1468 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
1469 info->u.param[0] = volid;
1470 info->u.param[1] = (int)(info->inodeNumber & NAMEI_VNODEMASK);
1471 info->u.param[2] = (int)((info->inodeNumber >> NAMEI_UNIQSHIFT)
1472 & (Inode) NAMEI_UNIQMASK);
1473 info->u.param[3] = parm;
1479 * Convert the VolumeInfo file from RO to RW
1480 * this routine is called by namei_convertROtoRWvolume()
1484 convertVolumeInfo(fdr, fdw, vid)
1489 struct VolumeDiskData vd;
1492 if (read(fdr, &vd, sizeof(struct VolumeDiskData)) !=
1493 sizeof(struct VolumeDiskData)) {
1494 Log("1 convertVolumeInfo: read failed for %lu with code %d\n", vid,
1498 vd.restoredFromId = vd.id; /* remember the RO volume here */
1500 vd.id = vd.parentId;
1503 vd.uniquifier += 5000; /* just in case there are still file copies from
1504 * the old RW volume around */
1505 p = strrchr(vd.name, '.');
1506 if (p && !strcmp(p, ".readonly")) {
1509 if (write(fdw, &vd, sizeof(struct VolumeDiskData)) !=
1510 sizeof(struct VolumeDiskData)) {
1511 Log("1 convertVolumeInfo: write failed for %lu with code %d\n", vid,
1519 * Convert a RO-volume into a RW-volume
1521 * This function allows to recover very fast from the loss of a partition
1522 * from RO-copies if all RO-Copies exist on another partition.
1523 * Then these RO-volumes can be made to the new RW-volumes.
1524 * Backup of RW-volumes then consists in "vos release".
1526 * We must make sure in this partition exists only the RO-volume which
1527 * is typical for remote replicas.
1529 * Then the linktable is already ok,
1530 * the vnode files need to be renamed
1531 * the volinfo file needs to be replaced by another one with
1532 * slightly different contents and new name.
1533 * The volume header file of the RO-volume in the /vicep<x> directory
1534 * is destroyed by this call. A new header file for the RW-volume must
1535 * be created after return from this routine.
1539 namei_ConvertROtoRWvolume(IHandle_t * h, afs_uint32 vid)
1542 char dir_name[512], oldpath[512], newpath[512];
1555 struct ViceInodeInfo info;
1557 namei_HandleToName(&n, h);
1558 strcpy(dir_name, n.n_path);
1559 p = strrchr(dir_name, '/');
1561 dirp = opendir(dir_name);
1563 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name);
1567 while ((dp = readdir(dirp))) {
1568 /* struct ViceInodeInfo info; */
1570 if (*dp->d_name == '.')
1572 if (DecodeInode(dir_name, dp->d_name, &info, h->ih_vid) < 0) {
1573 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s/%s\n",
1574 dir_name, dp->d_name);
1578 if (info.u.param[1] != -1) {
1579 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s/%s\n", dir_name, dp->d_name);
1583 if (info.u.param[0] != vid) {
1584 if (info.u.param[0] == h->ih_vid) {
1585 if (info.u.param[2] == VI_LINKTABLE) { /* link table */
1590 Log("1 namei_ConvertROtoRWvolume: found special file %s/%s for volume %lu\n", dir_name, dp->d_name, info.u.param[0]);
1594 if (info.u.param[2] == VI_VOLINFO) { /* volume info file */
1595 strcpy(infoName, dp->d_name);
1597 } else if (info.u.param[2] == VI_SMALLINDEX) { /* small vnodes file */
1598 strcpy(smallName, dp->d_name);
1600 } else if (info.u.param[2] == VI_LARGEINDEX) { /* large vnodes file */
1601 strcpy(largeName, dp->d_name);
1605 Log("1 namei_ConvertROtoRWvolume: unknown type %d of special file found : %s/%s\n", info.u.param[2], dir_name, dp->d_name);
1611 if (!infoSeen || !smallSeen || !largeSeen || !linkSeen) {
1612 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name);
1617 * If we come here then there was only a RO-volume and we can safely
1621 memset(&t_ih, 0, sizeof(t_ih));
1622 t_ih.ih_dev = h->ih_dev;
1623 t_ih.ih_vid = h->ih_vid;
1625 (void)afs_snprintf(oldpath, sizeof oldpath, "%s/%s", dir_name, infoName);
1626 fd = afs_open(oldpath, O_RDWR, 0);
1628 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
1632 t_ih.ih_ino = namei_MakeSpecIno(h->ih_vid, VI_VOLINFO);
1633 namei_HandleToName(&n, &t_ih);
1634 fd2 = afs_open(n.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
1636 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n.n_path);
1640 code = convertVolumeInfo(fd, fd2, h->ih_vid);
1647 SetOGM(fd2, h->ih_vid, 1);
1650 t_ih.ih_ino = namei_MakeSpecIno(h->ih_vid, VI_SMALLINDEX);
1651 namei_HandleToName(&n, &t_ih);
1652 (void)afs_snprintf(newpath, sizeof newpath, "%s/%s", dir_name, smallName);
1653 fd = afs_open(newpath, O_RDWR, 0);
1655 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath);
1658 SetOGM(fd, h->ih_vid, 2);
1660 link(newpath, n.n_path);
1663 t_ih.ih_ino = namei_MakeSpecIno(h->ih_vid, VI_LARGEINDEX);
1664 namei_HandleToName(&n, &t_ih);
1665 (void)afs_snprintf(newpath, sizeof newpath, "%s/%s", dir_name, largeName);
1666 fd = afs_open(newpath, O_RDWR, 0);
1668 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath);
1671 SetOGM(fd, h->ih_vid, 3);
1673 link(newpath, n.n_path);
1682 * returns a static string used to print either 32 or 64 bit inode numbers.
1685 PrintInode(char *s, Inode ino)
1687 static afs_ino_str_t result;
1691 (void)afs_snprintf(s, sizeof(afs_ino_str_t), "%llu", (afs_uintmax_t) ino);
1698 /* Routines to facilitate removing zero link count files. */
1699 #define MAX_ZLC_NAMES 32
1700 #define MAX_ZLC_NAMELEN 16
1701 typedef struct zlcList_s {
1702 struct zlcList_s *zlc_next;
1704 char zlc_names[MAX_ZLC_NAMES][MAX_ZLC_NAMELEN];
1707 static zlcList_t *zlcAnchor = NULL;
1708 static zlcList_t *zlcCur = NULL;
1711 AddToZLCDeleteList(char dir, char *name)
1713 assert(strlen(name) <= MAX_ZLC_NAMELEN - 3);
1715 if (!zlcCur || zlcCur->zlc_n >= MAX_ZLC_NAMES) {
1716 if (zlcCur && zlcCur->zlc_next)
1717 zlcCur = zlcCur->zlc_next;
1719 zlcList_t *tmp = (zlcList_t *) malloc(sizeof(zlcList_t));
1725 zlcCur->zlc_next = tmp;
1729 zlcCur->zlc_next = NULL;
1733 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%c\\%s", dir, name);
1738 DeleteZLCFiles(char *path)
1744 for (z = zlcAnchor; z; z = z->zlc_next) {
1745 for (i = 0; i < z->zlc_n; i++) {
1746 (void)sprintf(fname, "%s\\%s", path, z->zlc_names[i]);
1747 if (namei_unlink(fname) < 0) {
1748 Log("ZLC: Can't unlink %s, dos error = %d\n", fname,
1752 z->zlc_n = 0; /* Can reuse space. */
1765 tnext = i->zlc_next;
1769 zlcCur = zlcAnchor = NULL;
1773 #endif /* AFS_NAMEI_ENV */