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>
24 #include <afs/assert.h>
27 #include <sys/param.h>
29 #if defined(AFS_SUN5_ENV) || defined(AFS_HPUX_ENV)
32 #include <afs/afsutil.h>
35 #include <afs/afsint.h>
39 #include "viceinode.h"
41 #include "partition.h"
43 #include <afs/errors.h>
45 /*@+fcnmacros +macrofcndecl@*/
48 extern off64_t afs_lseek(int FD, off64_t O, int F);
49 #endif /*S_SPLINT_S */
50 #define afs_lseek(FD, O, F) lseek64(FD, (off64_t)(O), F)
51 #define afs_stat stat64
52 #define afs_fstat fstat64
53 #define afs_open open64
54 #define afs_fopen fopen64
55 #else /* !O_LARGEFILE */
57 extern off_t afs_lseek(int FD, off_t O, int F);
58 #endif /*S_SPLINT_S */
59 #define afs_lseek(FD, O, F) lseek(FD, (off_t)(O), F)
61 #define afs_fstat fstat
63 #define afs_fopen fopen
64 #endif /* !O_LARGEFILE */
65 /*@=fcnmacros =macrofcndecl@*/
67 /*@printflike@*/ extern void Log(const char *format, ...);
70 #define LOCK_SH 1 /* shared lock */
71 #define LOCK_EX 2 /* exclusive lock */
72 #define LOCK_NB 4 /* don't block when locking */
73 #define LOCK_UN 8 /* unlock */
80 * This function emulates a subset of flock()
83 emul_flock(int fd, int cmd)
86 memset(&f, 0, sizeof (f));
95 return fcntl(fd, (cmd & LOCK_NB) ? F_SETLK : F_SETLKW, &f);
98 #define flock(f,c) emul_flock(f,c)
105 namei_iread(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
114 if (FDH_SEEK(fdP, offset, SEEK_SET) < 0) {
115 FDH_REALLYCLOSE(fdP);
119 nBytes = FDH_READ(fdP, buf, size);
125 namei_iwrite(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
134 if (FDH_SEEK(fdP, offset, SEEK_SET) < 0) {
135 FDH_REALLYCLOSE(fdP);
138 nBytes = FDH_WRITE(fdP, buf, size);
145 /* Inode number format:
146 * low 26 bits - vnode number - all 1's if volume special file.
148 * next 3 bits spare (0's)
149 * high 32 bits - uniquifier (regular) or type if spare
151 #define NAMEI_VNODEMASK 0x003ffffff
152 #define NAMEI_TAGMASK 0x7
153 #define NAMEI_TAGSHIFT 26
154 #define NAMEI_UNIQMASK 0xffffffff
155 #define NAMEI_UNIQSHIFT 32
156 #define NAMEI_INODESPECIAL ((Inode)NAMEI_VNODEMASK)
157 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
159 /* dir1 is the high 8 bits of the 26 bit vnode */
160 #define VNO_DIR1(vno) ((vno >> 14) & 0xff)
161 /* dir2 is the next 9 bits */
162 #define VNO_DIR2(vno) ((vno >> 9) & 0x1ff)
163 /* "name" is the low 9 bits of the vnode, the 3 bit tag and the uniq */
165 #define NAMEI_SPECDIR "special"
166 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
168 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
169 * group, not counting temporary (move) volumes.
170 * This is the number of separate files, all having
171 * the same vnode number, which can occur in a volume
182 static int namei_GetLinkCount2(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite);
184 static int GetFreeTag(IHandle_t * ih, int vno);
186 /* namei_HandleToInodeDir
188 * Construct the path name of the directory holding the inode data.
189 * Format: /<vicepx>/INODEDIR
192 #define PNAME_BLEN 64
194 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
196 char *tmp = name->n_base;
198 memset(name, '\0', sizeof(*name));
200 (void)volutil_PartitionName_r(ih->ih_dev, tmp, NAMEI_LCOMP_LEN);
201 tmp += VICE_PREFIX_SIZE;
202 tmp += ih->ih_dev > 25 ? 2 : 1;
205 (void)strcpy(tmp, INODEDIR);
206 (void)strcpy(name->n_path, name->n_base);
209 #define addtoname(N, C) \
211 strcat((N)->n_path, "/"); strcat((N)->n_path, C); \
216 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
220 namei_HandleToInodeDir(name, ih);
221 (void)int32_to_flipbase64(tmp, (int64_t) (ih->ih_vid & 0xff));
222 (void)strcpy(name->n_voldir1, tmp);
223 addtoname(name, name->n_voldir1);
224 (void)int32_to_flipbase64(tmp, (int64_t) ih->ih_vid);
225 (void)strcpy(name->n_voldir2, tmp);
226 addtoname(name, name->n_voldir2);
229 /* namei_HandleToName
231 * Constructs a file name for the fully qualified handle.
232 * Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special
235 namei_HandleToName(namei_t * name, IHandle_t * ih)
238 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
240 namei_HandleToVolDir(name, ih);
242 if (vno == NAMEI_VNODESPECIAL) {
243 (void)strcpy(name->n_dir1, NAMEI_SPECDIR);
244 addtoname(name, name->n_dir1);
245 name->n_dir2[0] = '\0';
247 (void)int32_to_flipbase64(str, VNO_DIR1(vno));
248 (void)strcpy(name->n_dir1, str);
249 addtoname(name, name->n_dir1);
250 (void)int32_to_flipbase64(str, VNO_DIR2(vno));
251 (void)strcpy(name->n_dir2, str);
252 addtoname(name, name->n_dir2);
254 (void)int64_to_flipbase64(str, (int64_t) ih->ih_ino);
255 (void)strcpy(name->n_inode, str);
256 addtoname(name, name->n_inode);
259 /* The following is a warning to tell sys-admins to not muck about in this
262 #define VICE_README "These files and directories are a part of the AFS \
263 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
264 ownership and permissions included.\n"
266 namei_ViceREADME(char *partition)
271 /* Create the inode directory if we're starting for the first time */
272 (void)afs_snprintf(filename, sizeof filename, "%s/%s", partition,
274 mkdir(filename, 0700);
276 (void)afs_snprintf(filename, sizeof filename, "%s/%s/README", partition,
278 fd = afs_open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0444);
280 (void)write(fd, VICE_README, strlen(VICE_README));
287 #define create_dir() \
289 if (mkdir(tmp, 0700)<0) { \
290 if (errno != EEXIST) \
298 #define create_nextdir(A) \
300 strcat(tmp, "/"); strcat(tmp, A); create_dir(); \
303 /* namei_CreateDataDirectories
305 * If creating the file failed because of ENOENT or ENOTDIR, try
306 * creating all the directories first.
309 namei_CreateDataDirectories(namei_t * name, int *created)
315 (void)strcpy(tmp, name->n_base);
318 create_nextdir(name->n_voldir1);
319 create_nextdir(name->n_voldir2);
320 create_nextdir(name->n_dir1);
321 if (name->n_dir2[0]) {
322 create_nextdir(name->n_dir2);
327 /* delTree(): Deletes an entire tree of directories (no files)
329 * root : Full path to the subtree. Should be big enough for PATH_MAX
330 * tree : the subtree to be deleted is rooted here. Specifies only the
331 * subtree beginning at tree (not the entire path). It should be
332 * a pointer into the "root" buffer.
334 * errp : errno of the first error encountered during the directory cleanup.
335 * *errp should have been initialized to 0.
338 * -1 : If errors were encountered during cleanup and error is set to
342 * If there are errors, we try to work around them and delete as many
343 * directories as possible. We don't attempt to remove directories that still
344 * have non-dir entries in them.
347 delTree(char *root, char *tree, int *errp)
355 /* delete the children first */
356 cp = strchr(tree, '/');
358 delTree(root, cp + 1, errp);
361 cp = tree + strlen(tree); /* move cp to the end of string tree */
363 /* now delete all entries in this dir */
364 if ((ds = opendir(root)) != (DIR *) NULL) {
366 while ((dirp = readdir(ds))) {
367 /* ignore . and .. */
368 if (!strcmp(dirp->d_name, ".") || !strcmp(dirp->d_name, ".."))
370 /* since root is big enough, we reuse the space to
371 * concatenate the dirname to the current tree
374 strcat(root, dirp->d_name);
375 if (afs_stat(root, &st) == 0 && S_ISDIR(st.st_mode)) {
376 /* delete this subtree */
377 delTree(root, cp + 1, errp);
379 *errp = *errp ? *errp : errno;
381 /* recover path to our cur tree by truncating it to
386 /* if (!errno) -- closedir not implicit if we got an error */
390 /* finally axe the current dir */
392 *errp = *errp ? *errp : errno;
394 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
396 #endif /* !AFS_PTHREAD_ENV */
401 /* if we encountered errors during cleanup, we return a -1 */
409 /* namei_RemoveDataDirectories
411 * Returns 0 on success.
412 * Returns -1 on error. Typically, callers ignore this error bcause we
413 * can continue running if the removes fail. The salvage process will
414 * finish tidying up for us. We only use the n_base and n_voldir1 entries
415 * and only do rmdir's.
419 namei_RemoveDataDirectories(namei_t * name)
421 char pbuf[MAXPATHLEN], *path = pbuf;
422 int prefixlen = strlen(name->n_base), err = 0;
424 strcpy(path, name->n_path);
426 /* move past the prefix */
427 path = path + prefixlen + 1; /* skip over the trailing / */
429 /* now delete all dirs upto path */
430 return delTree(pbuf, path, &err);
434 /* Create the file in the name space.
436 * Parameters stored as follows:
438 * p1 - volid - implied in containing directory.
439 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
440 * p3 - uniq -- bits 4-0 are in mode bits 4-0
441 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
443 * p1 - volid - creation time - dwHighDateTime
444 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
445 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
446 * p4 - parid - parent volume id - implied in containing directory.
448 * Return value is the inode number or (Inode)-1 if error.
449 * We "know" there is only one link table, so return EEXIST if there already
450 * is a link table. It's up to the calling code to test errno and increment
456 * This function is called by VCreateVolume to hide the implementation
457 * details of the inode numbers. This only allows for 7 volume special
458 * types, but if we get that far, this could should be dead by then.
461 namei_MakeSpecIno(int volid, int type)
464 ino = NAMEI_INODESPECIAL;
465 type &= NAMEI_TAGMASK;
466 ino |= ((Inode) type) << NAMEI_TAGSHIFT;
467 ino |= ((Inode) volid) << NAMEI_UNIQSHIFT;
471 /* SetOGM - set owner group and mode bits from parm and tag
473 * owner - low 15 bits of parm.
474 * group - next 15 bits of parm.
475 * mode - 2 bits of parm, then lowest = 3 bits of tag.
478 SetOGM(int fd, int parm, int tag)
480 int owner, group, mode;
482 owner = parm & 0x7fff;
483 group = (parm >> 15) & 0x7fff;
484 if (fchown(fd, owner, group) < 0)
487 mode = (parm >> 27) & 0x18;
489 if (fchmod(fd, mode) < 0)
496 /* GetOGM - get parm and tag from owner, group and mode bits. */
498 GetOGMFromStat(struct afs_stat *status, int *parm, int *tag)
500 *parm = status->st_uid | (status->st_gid << 15);
501 *parm |= (status->st_mode & 0x18) << 27;
502 *tag = status->st_mode & 0x7;
506 GetOGM(int fd, int *parm, int *tag)
508 struct afs_stat status;
509 if (afs_fstat(fd, &status) < 0)
512 GetOGMFromStat(&status, parm, tag);
516 int big_vno = 0; /* Just in case we ever do 64 bit vnodes. */
518 /* Derive the name and create it O_EXCL. If that fails we have an error.
519 * Get the tag from a free column in the link table.
522 namei_icreate(IHandle_t * lh, char *part, int p1, int p2, int p3, int p4)
535 memset((void *)&tmp, 0, sizeof(IHandle_t));
538 tmp.ih_dev = volutil_GetPartitionID(part);
539 if (tmp.ih_dev == -1) {
545 /* Parameters for special file:
546 * p1 - volume id - goes into owner/group/mode
549 * p4 - parent volume id
554 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
555 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
557 int vno = p2 & NAMEI_VNODEMASK;
558 /* Parameters for regular file:
570 /* If GetFreeTag succeeds, it atomically sets link count to 1. */
571 tag = GetFreeTag(lh, p2);
575 /* name is <uniq(p3)><tag><vno(p2)> */
577 tmp.ih_ino = (Inode) p2;
578 tmp.ih_ino |= ((Inode) tag) << NAMEI_TAGSHIFT;
579 tmp.ih_ino |= ((Inode) p3) << NAMEI_UNIQSHIFT;
584 namei_HandleToName(&name, &tmp);
585 fd = afs_open(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
587 if (errno == ENOTDIR || errno == ENOENT) {
588 if (namei_CreateDataDirectories(&name, &created_dir) < 0)
590 fd = afs_open(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR,
598 if (SetOGM(fd, ogm_parm, tag) < 0) {
604 if (p2 == -1 && p3 == VI_LINKTABLE) {
605 /* hack at tmp to setup for set link count call. */
606 memset((void *)&tfd, 0, sizeof(FdHandle_t)); /* minimalistic still, but a little cleaner */
609 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
617 if (code || (fd < 0)) {
621 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
626 return (code || (fd < 0)) ? (Inode) - 1 : tmp.ih_ino;
632 namei_iopen(IHandle_t * h)
637 /* Convert handle to file name. */
638 namei_HandleToName(&name, h);
639 fd = afs_open(name.n_path, O_RDWR, 0666);
643 /* Need to detect vol special file and just unlink. In those cases, the
644 * handle passed in _is_ for the inode. We only check p1 for the special
648 namei_dec(IHandle_t * ih, Inode ino, int p1)
655 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
658 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
660 /* Verify this is the right file. */
661 IH_INIT(tmp, ih->ih_dev, ih->ih_vid, ino);
670 if ((GetOGM(fdP->fd_fd, &inode_p1, &tag) < 0) || (inode_p1 != p1)) {
671 FDH_REALLYCLOSE(fdP);
677 /* If it's the link table itself, decrement the link count. */
678 if (type == VI_LINKTABLE) {
679 if ((count = namei_GetLinkCount(fdP, (Inode) 0, 1)) < 0) {
680 FDH_REALLYCLOSE(fdP);
686 if (namei_SetLinkCount(fdP, (Inode) 0, count < 0 ? 0 : count, 1) <
688 FDH_REALLYCLOSE(fdP);
694 FDH_REALLYCLOSE(fdP);
700 namei_HandleToName(&name, tmp);
701 if ((code = unlink(name.n_path)) == 0) {
702 if (type == VI_LINKTABLE) {
703 /* Try to remove directory. If it fails, that's ok.
704 * Salvage will clean up.
706 (void)namei_RemoveDataDirectories(&name);
709 FDH_REALLYCLOSE(fdP);
712 /* Get a file descriptor handle for this Inode */
718 if ((count = namei_GetLinkCount(fdP, ino, 1)) < 0) {
719 FDH_REALLYCLOSE(fdP);
725 if (namei_SetLinkCount(fdP, ino, count, 1) < 0) {
726 FDH_REALLYCLOSE(fdP);
731 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
732 Log("Warning: Lost ref on ihandle dev %d vid %d ino %" AFS_INT64_FMT "\n",
733 th->ih_dev, th->ih_vid, (int64_t) th->ih_ino);
736 /* If we're less than 0, someone presumably unlinked;
737 don't bother setting count to 0, but we need to drop a lock */
738 if (namei_SetLinkCount(fdP, ino, 0, 1) < 0) {
739 FDH_REALLYCLOSE(fdP);
745 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
747 namei_HandleToName(&name, th);
749 code = unlink(name.n_path);
758 namei_inc(IHandle_t * h, Inode ino, int p1)
764 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
765 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
766 if (type != VI_LINKTABLE)
771 /* Get a file descriptor handle for this Inode */
777 if ((count = namei_GetLinkCount(fdP, ino, 1)) < 0)
786 if (namei_SetLinkCount(fdP, ino, count, 1) < 0)
790 FDH_REALLYCLOSE(fdP);
798 namei_replace_file_by_hardlink(IHandle_t *hLink, IHandle_t *hTarget)
804 /* Convert handle to file name. */
805 namei_HandleToName(&nameLink, hLink);
806 namei_HandleToName(&nameTarget, hTarget);
808 unlink(nameLink.n_path);
809 code = link(nameTarget.n_path, nameLink.n_path);
814 namei_copy_on_write(IHandle_t *h)
816 afs_int32 fd, code = 0;
819 struct afs_stat tstat;
821 namei_HandleToName(&name, h);
822 if (afs_stat(name.n_path, &tstat) < 0)
824 if (tstat.st_nlink > 1) { /* do a copy on write */
833 strcpy(path, name.n_path);
834 strcat(path, "-tmp");
835 fd = afs_open(path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
847 size = tstat.st_size;
850 tlen = size > 8192 ? 8192 : size;
851 if (FDH_READ(fdP, buf, tlen) != tlen)
853 if (write(fd, buf, tlen) != tlen)
858 FDH_REALLYCLOSE(fdP);
864 code = rename(path, name.n_path);
870 /************************************************************************
871 * File Name Structure
872 ************************************************************************
874 * Each AFS file needs a unique name and it needs to be findable with
875 * minimal lookup time. Note that the constraint on the number of files and
876 * directories in a volume is the size of the vnode index files and the
877 * max file size AFS supports (for internal files) of 2^31. Since a record
878 * in the small vnode index file is 64 bytes long, we can have at most
879 * (2^31)/64 or 33554432 files. A record in the large index file is
880 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
881 * Another layout parameter is that there is roughly a 16 to 1 ratio between
882 * the number of files and the number of directories.
884 * Using this information we can see that a layout of 256 directories, each
885 * with 512 subdirectories and each of those having 512 files gives us
886 * 256*512*512 = 67108864 AFS files and directories.
888 * The volume, vnode, uniquifier and data version, as well as the tag
889 * are required, either for finding the file or for salvaging. It's best to
890 * restrict the name to something that can be mapped into 64 bits so the
891 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
892 * is used to distinguish between different versions of the same file
893 * which are currently in the RW and clones of a volume. See "Link Table
894 * Organization" below for more information on the tag. The tag is
895 * required in the name of the file to ensure a unique name.
897 * We can store data in the uid, gid and mode bits of the files, provided
898 * the directories have root only access. This gives us 15 bits for each
899 * of uid and gid (GNU chown considers 65535 to mean "don't change").
900 * There are 9 available mode bits. Adn we need to store a total of
901 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
902 * or 131 bits somewhere.
904 * The format of a file name for a regular file is:
905 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
906 * V1 - low 8 bits of RW volume id
907 * V2 - all bits of RW volume id
908 * AA - high 8 bits of vnode number.
909 * BB - next 9 bits of vnode number.
910 * <tag><uniq><vno> - file name
912 * Volume special files are stored in a separate directory:
913 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
916 * The vnode is hashed into the directory using the high bits of the
917 * vnode number. This is so that consecutively created vnodes are in
918 * roughly the same area on the disk. This will at least be optimal if
919 * the user is creating many files in the same AFS directory. The name
920 * should be formed so that the leading characters are different as quickly
921 * as possible, leading to faster discards of incorrect matches in the
927 /************************************************************************
928 * Link Table Organization
929 ************************************************************************
931 * The link table volume special file is used to hold the link counts that
932 * are held in the inodes in inode based AFS vice filesystems. For user
933 * space access, the link counts are being kept in a separate
934 * volume special file. The file begins with the usual version stamp
935 * information and is then followed by one row per vnode number. vnode 0
936 * is used to hold the link count of the link table itself. That is because
937 * the same link table is shared among all the volumes of the volume group
938 * and is deleted only when the last volume of a volume group is deleted.
940 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
941 * link count from 0 through 7. Each colume represents a unique instance of
942 * that vnode. Say we have a file shared between the RW and a RO and a
943 * different version of the file (or a different uniquifer) for the BU volume.
944 * Then one column would be holding the link count of 2 for the RW and RO
945 * and a different column would hold the link count of 1 for the BU volume.
946 * Note that we allow only 5 volumes per file, giving 15 bits used in the
949 #define LINKTABLE_WIDTH 2
950 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
953 namei_GetLCOffsetAndIndexFromIno(Inode ino, afs_foff_t * offset, int *index)
955 int toff = (int)(ino & NAMEI_VNODEMASK);
956 int tindex = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
958 *offset = (afs_foff_t) ((toff << LINKTABLE_SHIFT) + 8); /* * 2 + sizeof stamp */
959 *index = (tindex << 1) + tindex;
963 /* namei_GetLinkCount
964 * If lockit is set, lock the file and leave it locked upon a successful
968 namei_GetLinkCount2(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite)
970 unsigned short row = 0;
975 /* there's no linktable yet. the salvager will create one later */
976 if (h->fd_fd == -1 && fixup)
978 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
981 if (flock(h->fd_fd, LOCK_EX) < 0)
985 if (afs_lseek(h->fd_fd, offset, SEEK_SET) == -1)
986 goto bad_getLinkByte;
988 rc = read(h->fd_fd, (char *)&row, sizeof(row));
989 if ((rc == 0 || !((row >> index) & NAMEI_TAGMASK)) && fixup && nowrite)
991 if (rc == 0 && fixup) {
993 if (fstat(h->fd_fd, &st) || st.st_size >= offset+sizeof(row))
994 goto bad_getLinkByte;
995 FDH_TRUNC(h, offset+sizeof(row));
997 rc = write(h->fd_fd, (char *)&row, sizeof(row));
999 if (rc != sizeof(row)) {
1000 goto bad_getLinkByte;
1003 if (fixup && !((row >> index) & NAMEI_TAGMASK)) {
1005 if (afs_lseek(h->fd_fd, offset, SEEK_SET) == -1)
1006 goto bad_getLinkByte;
1007 rc = write(h->fd_fd, (char *)&row, sizeof(row));
1008 if (rc != sizeof(row))
1009 goto bad_getLinkByte;
1012 return (int)((row >> index) & NAMEI_TAGMASK);
1016 flock(h->fd_fd, LOCK_UN);
1021 namei_GetLinkCount(FdHandle_t * h, Inode ino, int lockit)
1023 return namei_GetLinkCount2(h, ino, lockit, 0, 1);
1026 /* Return a free column index for this vnode. */
1028 GetFreeTag(IHandle_t * ih, int vno)
1042 /* Only one manipulates at a time. */
1043 if (flock(fdP->fd_fd, LOCK_EX) < 0) {
1044 FDH_REALLYCLOSE(fdP);
1048 offset = (vno << LINKTABLE_SHIFT) + 8; /* * 2 + sizeof stamp */
1049 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
1053 code = read(fdP->fd_fd, (char *)&row, sizeof(row));
1054 if (code != sizeof(row)) {
1060 /* Now find a free column in this row and claim it. */
1061 for (col = 0; col < NAMEI_MAXVOLS; col++) {
1062 coldata = 7 << (col * 3);
1063 if ((row & coldata) == 0)
1066 if (col >= NAMEI_MAXVOLS) {
1071 coldata = 1 << (col * 3);
1074 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
1077 if (write(fdP->fd_fd, (char *)&row, sizeof(row)) != sizeof(row)) {
1081 flock(fdP->fd_fd, LOCK_UN);
1082 FDH_REALLYCLOSE(fdP);
1086 flock(fdP->fd_fd, LOCK_UN);
1087 FDH_REALLYCLOSE(fdP);
1093 /* namei_SetLinkCount
1094 * If locked is set, assume file is locked. Otherwise, lock file before
1095 * proceeding to modify it.
1098 namei_SetLinkCount(FdHandle_t * fdP, Inode ino, int count, int locked)
1106 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1109 if (flock(fdP->fd_fd, LOCK_EX) < 0) {
1113 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
1115 goto bad_SetLinkCount;
1119 code = read(fdP->fd_fd, (char *)&row, sizeof(row));
1120 if (code != sizeof(row)) {
1123 goto bad_SetLinkCount;
1130 row &= (unsigned short)~junk;
1131 row |= (unsigned short)count;
1133 if (afs_lseek(fdP->fd_fd, offset, SEEK_SET) == -1) {
1135 goto bad_SetLinkCount;
1138 if (write(fdP->fd_fd, (char *)&row, sizeof(short)) != sizeof(short)) {
1140 goto bad_SetLinkCount;
1148 flock(fdP->fd_fd, LOCK_UN);
1154 /* ListViceInodes - write inode data to a results file. */
1155 static int DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
1157 static int DecodeVolumeName(char *name, int *vid);
1158 static int namei_ListAFSSubDirs(IHandle_t * dirIH,
1159 int (*write_fun) (FILE *,
1160 struct ViceInodeInfo *,
1161 char *, char *), FILE * fp,
1162 int (*judgeFun) (struct ViceInodeInfo *,
1163 afs_uint32 vid, void *),
1164 afs_uint32 singleVolumeNumber, void *rock);
1169 * Write the inode data to the results file.
1171 * Returns -2 on error, 0 on success.
1173 * This is written as a callback simply so that other listing routines
1174 * can use the same inode reading code.
1177 WriteInodeInfo(FILE * fp, struct ViceInodeInfo *info, char *dir, char *name)
1180 n = fwrite(info, sizeof(*info), 1, fp);
1181 return (n == 1) ? 0 : -2;
1185 int mode_errors; /* Number of errors found in mode bits on directories. */
1187 VerifyDirPerms(char *path)
1189 struct afs_stat status;
1191 if (afs_stat(path, &status) < 0) {
1192 Log("Unable to stat %s. Please manually verify mode bits for this"
1193 " directory\n", path);
1195 if (((status.st_mode & 0777) != 0700) || (status.st_uid != 0))
1201 * Fill the results file with the requested inode information.
1205 * -1 - complete failure, salvage should terminate.
1206 * -2 - not enough space on partition, salvager has error message for this.
1208 * This code optimizes single volume salvages by just looking at that one
1209 * volume's directory.
1211 * If the resultFile is NULL, then don't call the write routine.
1214 ListViceInodes(char *devname, char *mountedOn, char *resultFile,
1215 int (*judgeInode) (struct ViceInodeInfo * info, afs_uint32 vid, void *rock),
1216 afs_uint32 singleVolumeNumber, int *forcep, int forceR, char *wpath,
1219 FILE *fp = (FILE *) - 1;
1221 struct afs_stat status;
1223 *forcep = 0; /* no need to salvage until further notice */
1226 fp = afs_fopen(resultFile, "w");
1228 Log("Unable to create inode description file %s\n", resultFile);
1233 /* Verify protections on directories. */
1235 VerifyDirPerms(mountedOn);
1238 namei_ListAFSFiles(mountedOn, WriteInodeInfo, fp, judgeInode,
1239 singleVolumeNumber, rock);
1249 if (fflush(fp) == EOF) {
1250 Log("Unable to successfully flush inode file for %s\n", mountedOn);
1254 if (fsync(fileno(fp)) == -1) {
1255 Log("Unable to successfully fsync inode file for %s\n", mountedOn);
1259 if (fclose(fp) == EOF) {
1260 Log("Unable to successfully close inode file for %s\n", mountedOn);
1265 * Paranoia: check that the file is really the right size
1267 if (afs_stat(resultFile, &status) == -1) {
1268 Log("Unable to successfully stat inode file for %s\n", mountedOn);
1271 if (status.st_size != ninodes * sizeof(struct ViceInodeInfo)) {
1272 Log("Wrong size (%d instead of %d) in inode file for %s\n",
1273 status.st_size, ninodes * sizeof(struct ViceInodeInfo),
1281 /* namei_ListAFSFiles
1283 * Collect all the matching AFS files on the drive.
1284 * If singleVolumeNumber is non-zero, just return files for that volume.
1286 * Returns <0 on error, else number of files found to match.
1289 namei_ListAFSFiles(char *dev,
1290 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1292 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1293 afs_uint32 singleVolumeNumber, void *rock)
1299 struct dirent *dp1, *dp2;
1302 static void FreeZLCList(void);
1305 memset((void *)&ih, 0, sizeof(IHandle_t));
1306 ih.ih_dev = volutil_GetPartitionID(dev);
1308 if (singleVolumeNumber) {
1309 ih.ih_vid = singleVolumeNumber;
1310 namei_HandleToVolDir(&name, &ih);
1312 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1313 singleVolumeNumber, rock);
1317 /* Find all volume data directories and descend through them. */
1318 namei_HandleToInodeDir(&name, &ih);
1320 dirp1 = opendir(name.n_path);
1323 while ((dp1 = readdir(dirp1))) {
1324 if (*dp1->d_name == '.')
1326 (void)strcpy(path2, name.n_path);
1327 (void)strcat(path2, "/");
1328 (void)strcat(path2, dp1->d_name);
1329 dirp2 = opendir(path2);
1331 while ((dp2 = readdir(dirp2))) {
1332 if (*dp2->d_name == '.')
1334 if (!DecodeVolumeName(dp2->d_name, &ih.ih_vid)) {
1336 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1353 /* namei_ListAFSSubDirs
1358 * > = 0 - number of AFS files found.
1361 namei_ListAFSSubDirs(IHandle_t * dirIH,
1362 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1364 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1365 afs_uint32 singleVolumeNumber, void *rock)
1367 IHandle_t myIH = *dirIH;
1369 char path1[512], path2[512], path3[512];
1370 DIR *dirp1, *dirp2, *dirp3;
1371 struct dirent *dp1, *dp2, *dp3;
1372 struct ViceInodeInfo info;
1373 FdHandle_t linkHandle;
1377 static void AddToZLCDeleteList(char dir, char *name);
1378 static void DeleteZLCFiles(char *path);
1381 namei_HandleToVolDir(&name, &myIH);
1382 (void)strcpy(path1, name.n_path);
1384 /* Do the directory containing the special files first to pick up link
1387 (void)strcat(path1, "/");
1388 (void)strcat(path1, NAMEI_SPECDIR);
1390 linkHandle.fd_fd = -1;
1391 dirp1 = opendir(path1);
1393 while ((dp1 = readdir(dirp1))) {
1394 if (*dp1->d_name == '.')
1396 if (DecodeInode(path1, dp1->d_name, &info, myIH.ih_vid) < 0)
1398 if (info.u.param[2] != VI_LINKTABLE) {
1401 /* Open this handle */
1402 (void)afs_snprintf(path2, sizeof path2, "%s/%s", path1,
1404 linkHandle.fd_fd = afs_open(path2, Testing ? O_RDONLY : O_RDWR, 0666);
1406 namei_GetLinkCount2(&linkHandle, (Inode) 0, 1, 1, Testing);
1408 if (judgeFun && !(*judgeFun) (&info, singleVolumeNumber, rock))
1411 if ((*writeFun) (fp, &info, path1, dp1->d_name) < 0) {
1412 if (linkHandle.fd_fd >= 0)
1413 close(linkHandle.fd_fd);
1422 /* Now run through all the other subdirs */
1423 namei_HandleToVolDir(&name, &myIH);
1424 (void)strcpy(path1, name.n_path);
1426 dirp1 = opendir(path1);
1428 while ((dp1 = readdir(dirp1))) {
1429 if (*dp1->d_name == '.')
1431 if (!strcmp(dp1->d_name, NAMEI_SPECDIR))
1434 /* Now we've got a next level subdir. */
1435 (void)strcpy(path2, path1);
1436 (void)strcat(path2, "/");
1437 (void)strcat(path2, dp1->d_name);
1438 dirp2 = opendir(path2);
1440 while ((dp2 = readdir(dirp2))) {
1441 if (*dp2->d_name == '.')
1444 /* Now we've got to the actual data */
1445 (void)strcpy(path3, path2);
1446 (void)strcat(path3, "/");
1447 (void)strcat(path3, dp2->d_name);
1448 dirp3 = opendir(path3);
1450 while ((dp3 = readdir(dirp3))) {
1451 if (*dp3->d_name == '.')
1454 (path3, dp3->d_name, &info, myIH.ih_vid) < 0)
1457 namei_GetLinkCount2(&linkHandle,
1458 info.inodeNumber, 1, 1, Testing);
1459 if (info.linkCount == 0) {
1461 Log("Found 0 link count file %s/%s, deleting it.\n", path3, dp3->d_name);
1462 AddToZLCDeleteList((char)i, dp3->d_name);
1464 Log("Found 0 link count file %s/%s.\n", path3,
1470 && !(*judgeFun) (&info, singleVolumeNumber, rock))
1473 if ((*writeFun) (fp, &info, path3, dp3->d_name) <
1475 close(linkHandle.fd_fd);
1492 if (linkHandle.fd_fd >= 0)
1493 close(linkHandle.fd_fd);
1495 /* Then why does this directory exist? Blow it away. */
1496 namei_HandleToVolDir(&name, dirIH);
1497 namei_RemoveDataDirectories(&name);
1504 DecodeVolumeName(char *name, int *vid)
1506 if (strlen(name) <= 2)
1508 *vid = (int)flipbase64_to_int64(name);
1515 * Get the inode number from the name.
1519 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info, int volid)
1522 struct afs_stat status;
1524 lb64_string_t check;
1526 (void)strcpy(fpath, dpath);
1527 (void)strcat(fpath, "/");
1528 (void)strcat(fpath, name);
1530 if (afs_stat(fpath, &status) < 0) {
1534 info->byteCount = status.st_size;
1535 info->inodeNumber = (Inode) flipbase64_to_int64(name);
1537 int64_to_flipbase64(check, info->inodeNumber);
1538 if (strcmp(name, check))
1541 GetOGMFromStat(&status, &parm, &tag);
1542 if ((info->inodeNumber & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1543 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
1544 info->u.param[0] = parm;
1545 info->u.param[1] = -1;
1546 info->u.param[2] = tag;
1547 info->u.param[3] = volid;
1549 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
1550 info->u.param[0] = volid;
1551 info->u.param[1] = (int)(info->inodeNumber & NAMEI_VNODEMASK);
1552 info->u.param[2] = (int)((info->inodeNumber >> NAMEI_UNIQSHIFT)
1553 & (Inode) NAMEI_UNIQMASK);
1554 info->u.param[3] = parm;
1560 * Convert the VolumeInfo file from RO to RW
1561 * this routine is called by namei_convertROtoRWvolume()
1565 convertVolumeInfo(int fdr, int fdw, afs_uint32 vid)
1567 struct VolumeDiskData vd;
1570 if (read(fdr, &vd, sizeof(struct VolumeDiskData)) !=
1571 sizeof(struct VolumeDiskData)) {
1572 Log("1 convertVolumeInfo: read failed for %lu with code %d\n", vid,
1576 vd.restoredFromId = vd.id; /* remember the RO volume here */
1578 vd.id = vd.parentId;
1581 vd.uniquifier += 5000; /* just in case there are still file copies from
1582 * the old RW volume around */
1583 p = strrchr(vd.name, '.');
1584 if (p && !strcmp(p, ".readonly")) {
1587 if (write(fdw, &vd, sizeof(struct VolumeDiskData)) !=
1588 sizeof(struct VolumeDiskData)) {
1589 Log("1 convertVolumeInfo: write failed for %lu with code %d\n", vid,
1597 * Convert a RO-volume into a RW-volume
1599 * This function allows to recover very fast from the loss of a partition
1600 * from RO-copies if all RO-Copies exist on another partition.
1601 * Then these RO-volumes can be made to the new RW-volumes.
1602 * Backup of RW-volumes then consists in "vos release".
1604 * We must make sure in this partition exists only the RO-volume which
1605 * is typical for remote replicas.
1607 * Then the linktable is already ok,
1608 * the vnode files need to be renamed
1609 * the volinfo file needs to be replaced by another one with
1610 * slightly different contents and new name.
1611 * The volume header file of the RO-volume in the /vicep<x> directory
1612 * is destroyed by this call. A new header file for the RW-volume must
1613 * be created after return from this routine.
1617 namei_ConvertROtoRWvolume(char *pname, afs_uint32 volumeId)
1619 #ifdef FSSYNC_BUILD_CLIENT
1621 char dir_name[512], oldpath[512], newpath[512];
1636 struct DiskPartition64 *partP;
1637 struct ViceInodeInfo info;
1638 struct VolumeDiskHeader h;
1639 char headername[16];
1640 afs_int32 error = 0;
1642 (void)afs_snprintf(headername, sizeof headername, VFORMAT, afs_cast_uint32(volumeId));
1643 (void)afs_snprintf(oldpath, sizeof oldpath, "%s/%s", pname, headername);
1644 fd = open(oldpath, O_RDONLY);
1646 Log("1 namei_ConvertROtoRWvolume: Couldn't open header for RO-volume %lu.\n", volumeId);
1649 if (read(fd, &h, sizeof(h)) != sizeof(h)) {
1650 Log("1 namei_ConvertROtoRWvolume: Couldn't read header for RO-volume %lu.\n", volumeId);
1655 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_BREAKCBKS, 0, NULL);
1657 for (partP = DiskPartitionList; partP && strcmp(partP->name, pname);
1658 partP = partP->next);
1660 Log("1 namei_ConvertROtoRWvolume: Couldn't find DiskPartition for %s\n", pname);
1663 ino = namei_MakeSpecIno(h.parent, VI_LINKTABLE);
1664 IH_INIT(ih, partP->device, h.parent, ino);
1666 namei_HandleToName(&n, ih);
1667 strcpy(dir_name, n.n_path);
1668 p = strrchr(dir_name, '/');
1670 dirp = opendir(dir_name);
1672 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name);
1676 while ((dp = readdir(dirp))) {
1677 /* struct ViceInodeInfo info; */
1679 if (*dp->d_name == '.')
1681 if (DecodeInode(dir_name, dp->d_name, &info, ih->ih_vid) < 0) {
1682 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s/%s\n",
1683 dir_name, dp->d_name);
1687 if (info.u.param[1] != -1) {
1688 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s/%s\n", dir_name, dp->d_name);
1692 if (info.u.param[0] != volumeId) {
1693 if (info.u.param[0] == ih->ih_vid) {
1694 if (info.u.param[2] == VI_LINKTABLE) { /* link table */
1699 Log("1 namei_ConvertROtoRWvolume: found special file %s/%s for volume %lu\n", dir_name, dp->d_name, info.u.param[0]);
1703 if (info.u.param[2] == VI_VOLINFO) { /* volume info file */
1704 strcpy(infoName, dp->d_name);
1706 } else if (info.u.param[2] == VI_SMALLINDEX) { /* small vnodes file */
1707 strcpy(smallName, dp->d_name);
1709 } else if (info.u.param[2] == VI_LARGEINDEX) { /* large vnodes file */
1710 strcpy(largeName, dp->d_name);
1714 Log("1 namei_ConvertROtoRWvolume: unknown type %d of special file found : %s/%s\n", info.u.param[2], dir_name, dp->d_name);
1720 if (!infoSeen || !smallSeen || !largeSeen || !linkSeen) {
1721 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name);
1726 * If we come here then there was only a RO-volume and we can safely
1730 memset(&t_ih, 0, sizeof(t_ih));
1731 t_ih.ih_dev = ih->ih_dev;
1732 t_ih.ih_vid = ih->ih_vid;
1734 (void)afs_snprintf(oldpath, sizeof oldpath, "%s/%s", dir_name, infoName);
1735 fd = afs_open(oldpath, O_RDWR, 0);
1737 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
1741 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_VOLINFO);
1742 namei_HandleToName(&n, &t_ih);
1743 fd2 = afs_open(n.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
1745 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n.n_path);
1749 code = convertVolumeInfo(fd, fd2, ih->ih_vid);
1756 SetOGM(fd2, ih->ih_vid, 1);
1759 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_SMALLINDEX);
1760 namei_HandleToName(&n, &t_ih);
1761 (void)afs_snprintf(newpath, sizeof newpath, "%s/%s", dir_name, smallName);
1762 fd = afs_open(newpath, O_RDWR, 0);
1764 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath);
1767 SetOGM(fd, ih->ih_vid, 2);
1769 link(newpath, n.n_path);
1772 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_LARGEINDEX);
1773 namei_HandleToName(&n, &t_ih);
1774 (void)afs_snprintf(newpath, sizeof newpath, "%s/%s", dir_name, largeName);
1775 fd = afs_open(newpath, O_RDWR, 0);
1777 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath);
1780 SetOGM(fd, ih->ih_vid, 3);
1782 link(newpath, n.n_path);
1788 h.volumeInfo_hi = h.id;
1789 h.smallVnodeIndex_hi = h.id;
1790 h.largeVnodeIndex_hi = h.id;
1791 h.linkTable_hi = h.id;
1792 (void)afs_snprintf(headername, sizeof headername, VFORMAT, afs_cast_uint32(h.id));
1793 (void)afs_snprintf(newpath, sizeof newpath, "%s/%s", pname, headername);
1794 fd = open(newpath, O_CREAT | O_EXCL | O_RDWR, 0644);
1796 Log("1 namei_ConvertROtoRWvolume: Couldn't create header for RW-volume %lu.\n", h.id);
1799 if (write(fd, &h, sizeof(h)) != sizeof(h)) {
1800 Log("1 namei_ConvertROtoRWvolume: Couldn't write header for RW-volume\
1806 if (unlink(oldpath) < 0) {
1807 Log("1 namei_ConvertROtoRWvolume: Couldn't unlink RO header, error = %d\n", error);
1809 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_DONE, 0, NULL);
1810 FSYNC_VolOp(h.id, pname, FSYNC_VOL_ON, 0, NULL);
1817 * returns a static string used to print either 32 or 64 bit inode numbers.
1820 PrintInode(char *s, Inode ino)
1822 static afs_ino_str_t result;
1826 (void)afs_snprintf(s, sizeof(afs_ino_str_t), "%llu", (afs_uintmax_t) ino);
1833 /* Routines to facilitate removing zero link count files. */
1834 #define MAX_ZLC_NAMES 32
1835 #define MAX_ZLC_NAMELEN 16
1836 typedef struct zlcList_s {
1837 struct zlcList_s *zlc_next;
1839 char zlc_names[MAX_ZLC_NAMES][MAX_ZLC_NAMELEN];
1842 static zlcList_t *zlcAnchor = NULL;
1843 static zlcList_t *zlcCur = NULL;
1846 AddToZLCDeleteList(char dir, char *name)
1848 assert(strlen(name) <= MAX_ZLC_NAMELEN - 3);
1850 if (!zlcCur || zlcCur->zlc_n >= MAX_ZLC_NAMES) {
1851 if (zlcCur && zlcCur->zlc_next)
1852 zlcCur = zlcCur->zlc_next;
1854 zlcList_t *tmp = (zlcList_t *) malloc(sizeof(zlcList_t));
1860 zlcCur->zlc_next = tmp;
1864 zlcCur->zlc_next = NULL;
1868 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%c\\%s", dir, name);
1873 DeleteZLCFiles(char *path)
1879 for (z = zlcAnchor; z; z = z->zlc_next) {
1880 for (i = 0; i < z->zlc_n; i++) {
1881 (void)sprintf(fname, "%s\\%s", path, z->zlc_names[i]);
1882 if (namei_unlink(fname) < 0) {
1883 Log("ZLC: Can't unlink %s, dos error = %d\n", fname,
1887 z->zlc_n = 0; /* Can reuse space. */
1900 tnext = i->zlc_next;
1904 zlcCur = zlcAnchor = NULL;
1908 #endif /* AFS_NAMEI_ENV */