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>
37 #include <sys/param.h>
40 #include <afs/afs_assert.h>
43 #include <afs/afsutil.h>
46 #include <afs/afsint.h>
50 #include "viceinode.h"
52 #include "partition.h"
54 #include "volume_inline.h"
56 #include <afs/errors.h>
58 #include <afs/errmap_nt.h>
62 #define LOCK_SH 1 /* shared lock */
63 #define LOCK_EX 2 /* exclusive lock */
64 #define LOCK_NB 4 /* don't block when locking */
65 #define LOCK_UN 8 /* unlock */
70 #include <afs/work_queue.h>
71 #include <afs/thread_pool.h>
72 #include <vol/vol-salvage.h>
75 #if !defined(HAVE_FLOCK) && !defined(AFS_NT40_ENV)
79 * This function emulates a subset of flock()
82 emul_flock(int fd, int cmd)
85 memset(&f, 0, sizeof (f));
94 return fcntl(fd, (cmd & LOCK_NB) ? F_SETLK : F_SETLKW, &f);
97 #define flock(f,c) emul_flock(f,c)
104 namei_iread(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
113 nBytes = FDH_PREAD(fdP, buf, size, offset);
119 namei_iwrite(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
128 nBytes = FDH_PWRITE(fdP, buf, size, offset);
134 /* Inode number format:
135 * low 32 bits - if a regular file or directory, the vnode; else the type.
136 * 32-36 - uniquifier tag and index into counts array for this vnode. Only
137 * two of the available bits are currently used. The rest are
138 * present in case we ever increase the number of types of volumes
139 * in the volume group.
140 * bit 37 : 1 == special, 0 == regular
142 # define NAMEI_VNODEMASK 0x00ffffffff
143 # define NAMEI_TAGSHIFT 32
144 # define NAMEI_INODESPECIAL 0x2000000000
145 # define NAMEI_SPECDIR "R"
146 # define NAMEI_SPECDIRC 'R'
147 #else /* !AFS_NT40_ENV */
148 /* Inode number format:
149 * low 26 bits - vnode number - all 1's if volume special file.
151 * next 3 bits spare (0's)
152 * high 32 bits - uniquifier (regular) or type if spare
154 # define NAMEI_VNODEMASK 0x003ffffff
155 # define NAMEI_TAGSHIFT 26
156 # define NAMEI_UNIQMASK 0xffffffff
157 # define NAMEI_UNIQSHIFT 32
158 # define NAMEI_INODESPECIAL ((Inode)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 */
164 # define NAMEI_SPECDIR "special"
165 #endif /* !AFS_NT40_ENV */
166 #define NAMEI_TAGMASK 0x7
167 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
169 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
171 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
172 * group, not counting temporary (move) volumes.
173 * This is the number of separate files, all having
174 * the same vnode number, which can occur in a volume
186 static int GetFreeTag(IHandle_t * ih, int vno);
188 /* namei_HandleToInodeDir
190 * Construct the path name of the directory holding the inode data.
191 * Format: /<vicepx>/INODEDIR
196 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
198 memset(name, '\0', sizeof(*name));
199 nt_DevToDrive(name->n_drive, ih->ih_dev);
200 strlcpy(name->n_path, name->n_drive, sizeof(name->n_path));
204 /* Format: /<vicepx>/INODEDIR */
205 #define PNAME_BLEN 64
207 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
211 memset(name, '\0', sizeof(*name));
214 * Add the /vicepXX string to the start of name->n_base and then calculate
215 * offset as the number of bytes we know we added.
217 * FIXME: This embeds knowledge of the vice partition naming scheme and
218 * mapping from device numbers. There needs to be an API that tells us
221 volutil_PartitionName_r(ih->ih_dev, name->n_base, sizeof(name->n_base));
222 offset = VICE_PREFIX_SIZE + (ih->ih_dev > 25 ? 2 : 1);
223 name->n_base[offset] = OS_DIRSEPC;
225 strlcpy(name->n_base + offset, INODEDIR, sizeof(name->n_base) - offset);
226 strlcpy(name->n_path, name->n_base, sizeof(name->n_path));
230 #define addtoname(N, C) \
232 if ((N)->n_path[strlen((N)->n_path)-1] != OS_DIRSEPC) \
233 strlcat((N)->n_path, OS_DIRSEP, sizeof((N)->n_path)); \
234 strlcat((N)->n_path, (C), sizeof((N)->n_path)); \
240 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
242 /* X:\Vol_XXXXXXX.data */
246 namei_HandleToInodeDir(name, ih);
247 /* nt_drive added to name by namei_HandleToInodeDir() */
248 namep = name->n_voldir;
249 (void)memcpy(namep, "\\Vol_", 5);
251 (void)strcpy(namep, int_to_base32(str1, ih->ih_vid));
252 namep += strlen(namep);
253 memcpy(namep, ".data", 5);
256 addtoname(name, name->n_voldir);
260 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
264 namei_HandleToInodeDir(name, ih);
265 (void)int32_to_flipbase64(tmp, (int64_t) (ih->ih_vid & 0xff));
266 strlcpy(name->n_voldir1, tmp, sizeof(name->n_voldir1));
267 addtoname(name, name->n_voldir1);
268 (void)int32_to_flipbase64(tmp, (int64_t) ih->ih_vid);
269 strlcpy(name->n_voldir2, tmp, sizeof(name->n_voldir2));
270 addtoname(name, name->n_voldir2);
274 /* namei_HandleToName
276 * Constructs a file name for the fully qualified handle.
279 /* Note that special files end up in X:\Vol_XXXXXXX.data\R */
281 namei_HandleToName(namei_t * name, IHandle_t * ih)
283 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
284 int special = (ih->ih_ino & NAMEI_INODESPECIAL)?1:0;
285 int tag = (int)((ih->ih_ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
288 namei_HandleToVolDir(name, ih);
291 name->n_dir[0] = NAMEI_SPECDIRC;
294 name->n_dir[0] = 'Q';
296 name->n_dir[0] = ((vno & 0x1f) >> 1) + 'A';
299 name->n_dir[1] = '\0';
300 addtoname(name, name->n_dir);
301 /* X:\Vol_XXXXXXX.data\X\V_XXXXXXX.XXX */
302 namep = name->n_inode;
303 (void)memcpy(namep, "\\V_", 3);
305 (void)strcpy(namep, int_to_base32(str1, vno));
306 namep += strlen(namep);
308 (void)strcpy(namep, int_to_base32(str1, tag));
309 namep += strlen(namep);
310 addtoname(name, name->n_inode);
313 /* Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special */
315 namei_HandleToName(namei_t * name, IHandle_t * ih)
317 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
320 namei_HandleToVolDir(name, ih);
322 if (vno == NAMEI_VNODESPECIAL) {
323 strlcpy(name->n_dir1, NAMEI_SPECDIR, sizeof(name->n_dir1));
324 addtoname(name, name->n_dir1);
325 name->n_dir2[0] = '\0';
327 (void)int32_to_flipbase64(str, VNO_DIR1(vno));
328 strlcpy(name->n_dir1, str, sizeof(name->n_dir1));
329 addtoname(name, name->n_dir1);
330 (void)int32_to_flipbase64(str, VNO_DIR2(vno));
331 strlcpy(name->n_dir2, str, sizeof(name->n_dir2));
332 addtoname(name, name->n_dir2);
334 (void)int64_to_flipbase64(str, (int64_t) ih->ih_ino);
335 strlcpy(name->n_inode, str, sizeof(name->n_inode));
336 addtoname(name, name->n_inode);
341 /* The following is a warning to tell sys-admins to not muck about in this
344 #define VICE_README "These files and directories are a part of the AFS \
345 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
346 ownership and permissions included.\n"
348 namei_ViceREADME(char *partition)
353 /* Create the inode directory if we're starting for the first time */
354 (void)afs_snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s", partition,
356 mkdir(filename, 0700);
358 (void)afs_snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s" OS_DIRSEP "README",
359 partition, INODEDIR);
360 fd = OS_OPEN(filename, O_WRONLY | O_CREAT | O_TRUNC, 0444);
361 if (fd != INVALID_FD) {
362 (void)OS_WRITE(fd, VICE_README, strlen(VICE_README));
369 /* namei_CreateDataDirectories
371 * If creating the file failed because of ENOENT or ENOTDIR, try
372 * creating all the directories first.
376 namei_CreateDataDirectories(namei_t * name, int *created)
383 afs_snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
385 if (mkdir(tmp) < 0) {
396 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
398 if (mkdir(tmp) < 0 && errno != EEXIST)
404 #define create_dir() \
406 if (mkdir(tmp, 0700)<0) { \
407 if (errno != EEXIST) \
415 #define create_nextdir(A) \
417 strcat(tmp, OS_DIRSEP); strcat(tmp, A); create_dir(); \
421 namei_CreateDataDirectories(namei_t * name, int *created)
427 strlcpy(tmp, name->n_base, sizeof(tmp));
430 create_nextdir(name->n_voldir1);
431 create_nextdir(name->n_voldir2);
432 create_nextdir(name->n_dir1);
433 if (name->n_dir2[0]) {
434 create_nextdir(name->n_dir2);
441 /* delTree(): Deletes an entire tree of directories (no files)
443 * root : Full path to the subtree. Should be big enough for PATH_MAX
444 * tree : the subtree to be deleted is rooted here. Specifies only the
445 * subtree beginning at tree (not the entire path). It should be
446 * a pointer into the "root" buffer.
448 * errp : errno of the first error encountered during the directory cleanup.
449 * *errp should have been initialized to 0.
452 * -1 : If errors were encountered during cleanup and error is set to
456 * If there are errors, we try to work around them and delete as many
457 * directories as possible. We don't attempt to remove directories that still
458 * have non-dir entries in them.
461 delTree(char *root, char *tree, int *errp)
469 /* delete the children first */
470 cp = strchr(tree, OS_DIRSEPC);
472 delTree(root, cp + 1, errp);
475 cp = tree + strlen(tree); /* move cp to the end of string tree */
477 /* now delete all entries in this dir */
478 if ((ds = opendir(root)) != (DIR *) NULL) {
480 while ((dirp = readdir(ds))) {
481 /* ignore . and .. */
482 if (!strcmp(dirp->d_name, ".") || !strcmp(dirp->d_name, ".."))
484 /* since root is big enough, we reuse the space to
485 * concatenate the dirname to the current tree
487 strcat(root, OS_DIRSEP);
488 strcat(root, dirp->d_name);
489 if (afs_stat(root, &st) == 0 && S_ISDIR(st.st_mode)) {
490 /* delete this subtree */
491 delTree(root, cp + 1, errp);
493 *errp = *errp ? *errp : errno;
495 /* recover path to our cur tree by truncating it to
500 /* if (!errno) -- closedir not implicit if we got an error */
504 /* finally axe the current dir */
506 *errp = *errp ? *errp : errno;
508 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
510 #endif /* !AFS_PTHREAD_ENV */
515 /* if we encountered errors during cleanup, we return a -1 */
524 /* namei_RemoveDataDirectories
526 * Returns 0 on success.
527 * Returns -1 on error. Typically, callers ignore this error because we
528 * can continue running if the removes fail. The salvage process will
529 * finish tidying up for us.
534 namei_RemoveDataDirectories(namei_t * name)
541 afs_snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
544 path += strlen(path);
545 *path++ = OS_DIRSEPC;
547 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
549 if (rmdir(name->n_path) < 0 && errno != ENOENT)
554 /* Delete the Vol_NNNNNN.data directory. */
557 if (rmdir(name->n_path) < 0 && errno != ENOENT) {
565 * We only use the n_base and n_voldir1 entries
566 * and only do rmdir's.
569 namei_RemoveDataDirectories(namei_t * name)
573 int prefixlen = strlen(name->n_base), err = 0;
574 int vollen = strlen(name->n_voldir1);
575 char pbuf[MAXPATHLEN];
579 strlcpy(path, name->n_path, sizeof(pbuf));
581 /* move past the prefix and n_voldir1 */
582 path = path + prefixlen + 1 + vollen + 1; /* skip over the trailing / */
584 /* now delete all dirs upto path */
585 code = delTree(pbuf, path, &err);
587 /* We've now deleted everything under /n_base/n_voldir1/n_voldir2 that
588 * we could. Do not delete /n_base/n_voldir1, since doing such might
589 * interrupt another thread trying to create a volume. We could introduce
590 * some locking to make this safe (or only remove it for whole-partition
591 * salvages), but by not deleting it we only leave behind a maximum of
592 * 256 empty directories. So at least for now, don't bother. */
597 /* Create the file in the name space.
599 * Parameters stored as follows:
601 * p1 - volid - implied in containing directory.
602 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
603 * p3 - uniq -- bits 4-0 are in mode bits 4-0
604 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
606 * p1 - volid - creation time - dwHighDateTime
607 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
608 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
609 * p4 - parid - parent volume id - implied in containing directory.
611 * Return value is the inode number or (Inode)-1 if error.
612 * We "know" there is only one link table, so return EEXIST if there already
613 * is a link table. It's up to the calling code to test errno and increment
619 * This function is called by VCreateVolume to hide the implementation
620 * details of the inode numbers. This only allows for 7 volume special
621 * types, but if we get that far, this could should be dead by then.
624 namei_MakeSpecIno(int volid, int type)
627 ino = NAMEI_INODESPECIAL;
630 /* tag is always 0 for special */
632 type &= NAMEI_TAGMASK;
633 ino |= ((Inode) type) << NAMEI_TAGSHIFT;
634 ino |= ((Inode) volid) << NAMEI_UNIQSHIFT;
642 SetOGM(FD_t fd, int parm, int tag)
648 CheckOGM(namei_t *name, FdHandle_t *fdP, int p1)
650 WIN32_FIND_DATA info;
654 FindFirstFileEx(name->n_path, FindExInfoStandard, &info,
655 FindExSearchNameMatch, NULL,
656 FIND_FIRST_EX_CASE_SENSITIVE);
659 return -1; /* Can't get info, leave alone */
663 if (info.ftCreationTime.dwHighDateTime != (unsigned int)p1)
668 #else /* AFS_NT40_ENV */
669 /* SetOGM - set owner group and mode bits from parm and tag */
671 SetOGM(FD_t fd, int parm, int tag)
674 * owner - low 15 bits of parm.
675 * group - next 15 bits of parm.
676 * mode - 2 bits of parm, then lowest = 3 bits of tag.
678 int owner, group, mode;
680 owner = parm & 0x7fff;
681 group = (parm >> 15) & 0x7fff;
682 if (fchown(fd, owner, group) < 0)
685 mode = (parm >> 27) & 0x18;
687 if (fchmod(fd, mode) < 0)
692 /* GetOGM - get parm and tag from owner, group and mode bits. */
694 GetOGMFromStat(struct afs_stat *status, int *parm, int *tag)
696 *parm = status->st_uid | (status->st_gid << 15);
697 *parm |= (status->st_mode & 0x18) << 27;
698 *tag = status->st_mode & 0x7;
702 CheckOGM(namei_t *name, FdHandle_t *fdP, int p1)
704 struct afs_stat status;
706 if (afs_fstat(fdP->fd_fd, &status) < 0)
709 GetOGMFromStat(&status, &parm, &tag);
715 #endif /* !AFS_NT40_ENV */
717 int big_vno = 0; /* Just in case we ever do 64 bit vnodes. */
719 /* Derive the name and create it O_EXCL. If that fails we have an error.
720 * Get the tag from a free column in the link table.
724 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
727 FD_t fd = INVALID_FD;
738 memset((void *)&tmp, 0, sizeof(IHandle_t));
739 memset(&tfd, 0, sizeof(FdHandle_t));
741 tmp.ih_dev = nt_DriveToDev(part);
742 if (tmp.ih_dev == -1) {
747 if (p2 == INODESPECIAL) {
748 /* Parameters for special file:
749 * p1 - volume id - goes into owner/group/mode
750 * p2 - vnode == INODESPECIAL
752 * p4 - parent volume id
754 ftime.dwHighDateTime = p1;
755 ftime.dwLowDateTime = p2;
757 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
758 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
760 int vno = p2 & NAMEI_VNODEMASK;
761 /* Parameters for regular file:
775 tmp.ih_ino = (Inode) p2;
776 ftime.dwHighDateTime = p3;
777 ftime.dwLowDateTime = p4;
780 namei_HandleToName(&name, &tmp);
781 p = strrchr((char *)&name.n_path, '.');
783 for (tag = 0; tag < NAMEI_MAXVOLS; tag++) {
784 *p = *int_to_base32(str1, tag);
785 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_TRUNC | O_EXCL, 0666);
786 if (fd == INVALID_FD) {
787 if (errno == ENOTDIR || errno == ENOENT) {
788 if (namei_CreateDataDirectories(&name, &created_dir) == 0)
789 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_TRUNC | O_EXCL, 0666);
793 if (fd != INVALID_FD)
795 if (p2 == INODESPECIAL && p3 == VI_LINKTABLE)
798 if (fd == INVALID_FD) {
802 tmp.ih_ino &= ~(((Inode) NAMEI_TAGMASK) << NAMEI_TAGSHIFT);
803 tmp.ih_ino |= (((Inode) tag) << NAMEI_TAGSHIFT);
806 if (!SetFileTime((HANDLE) fd, &ftime, NULL, NULL)) {
807 errno = OS_ERROR(EBADF);
813 if (p2 != INODESPECIAL) {
814 if (fd == INVALID_FD) {
816 code = nt_unlink((char *)&name.n_path);
825 code = namei_SetLinkCount(fdP, tmp.ih_ino, 1, 0);
827 } else if (p2 == INODESPECIAL && p3 == VI_LINKTABLE) {
828 if (fd == INVALID_FD)
830 /* hack at tmp to setup for set link count call. */
832 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
837 if (fd != INVALID_FD)
840 if (code || (fd == INVALID_FD)) {
841 if (p2 != INODESPECIAL) {
844 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
850 int save_errno = errno;
851 namei_RemoveDataDirectories(&name);
855 return (code || (fd == INVALID_FD)) ? (Inode) -1 : tmp.ih_ino;
857 #else /* !AFS_NT40_ENV */
859 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
871 memset((void *)&tmp, 0, sizeof(IHandle_t));
872 memset(&tfd, 0, sizeof(FdHandle_t));
874 tmp.ih_dev = volutil_GetPartitionID(part);
875 if (tmp.ih_dev == -1) {
881 /* Parameters for special file:
882 * p1 - volume id - goes into owner/group/mode
885 * p4 - parent volume id
890 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
891 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
893 int vno = p2 & NAMEI_VNODEMASK;
894 /* Parameters for regular file:
906 /* If GetFreeTag succeeds, it atomically sets link count to 1. */
907 tag = GetFreeTag(lh, p2);
912 tmp.ih_ino = (Inode) p2;
913 /* name is <uniq(p3)><tag><vno(p2)> */
914 tmp.ih_ino |= ((Inode) tag) << NAMEI_TAGSHIFT;
915 tmp.ih_ino |= ((Inode) p3) << NAMEI_UNIQSHIFT;
920 namei_HandleToName(&name, &tmp);
921 fd = OS_OPEN(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
922 if (fd == INVALID_FD) {
923 if (errno == ENOTDIR || errno == ENOENT) {
924 if (namei_CreateDataDirectories(&name, &created_dir) < 0)
926 fd = OS_OPEN(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR,
928 if (fd == INVALID_FD)
934 if (SetOGM(fd, ogm_parm, tag) < 0) {
940 if (p2 == (afs_uint32)-1 && p3 == VI_LINKTABLE) {
941 /* hack at tmp to setup for set link count call. */
942 memset((void *)&tfd, 0, sizeof(FdHandle_t)); /* minimalistic still, but a little cleaner */
945 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
949 if (fd != INVALID_FD)
953 if (code || (fd == INVALID_FD)) {
957 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
962 return (code || (fd == INVALID_FD)) ? (Inode) - 1 : tmp.ih_ino;
968 namei_iopen(IHandle_t * h)
973 /* Convert handle to file name. */
974 namei_HandleToName(&name, h);
975 fd = OS_OPEN((char *)&name.n_path, O_RDWR, 0666);
979 /* Need to detect vol special file and just unlink. In those cases, the
980 * handle passed in _is_ for the inode. We only check p1 for the special
984 namei_dec(IHandle_t * ih, Inode ino, int p1)
991 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
993 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
995 /* Verify this is the right file. */
996 IH_INIT(tmp, ih->ih_dev, ih->ih_vid, ino);
998 namei_HandleToName(&name, tmp);
1003 errno = OS_ERROR(ENOENT);
1007 if (CheckOGM(&name, fdP, p1) < 0) {
1008 FDH_REALLYCLOSE(fdP);
1010 errno = OS_ERROR(EINVAL);
1014 /* If it's the link table itself, decrement the link count. */
1015 if (type == VI_LINKTABLE) {
1016 if ((count = namei_GetLinkCount(fdP, (Inode) 0, 1, 0, 1)) < 0) {
1017 FDH_REALLYCLOSE(fdP);
1023 if (namei_SetLinkCount(fdP, (Inode) 0, count < 0 ? 0 : count, 1) <
1025 FDH_REALLYCLOSE(fdP);
1031 FDH_REALLYCLOSE(fdP);
1037 if ((code = OS_UNLINK(name.n_path)) == 0) {
1038 if (type == VI_LINKTABLE) {
1039 /* Try to remove directory. If it fails, that's ok.
1040 * Salvage will clean up.
1042 (void)namei_RemoveDataDirectories(&name);
1045 FDH_REALLYCLOSE(fdP);
1048 /* Get a file descriptor handle for this Inode */
1054 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0) {
1055 FDH_REALLYCLOSE(fdP);
1061 if (namei_SetLinkCount(fdP, ino, count, 1) < 0) {
1062 FDH_REALLYCLOSE(fdP);
1067 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1068 Log("Warning: Lost ref on ihandle dev %d vid %d ino %" AFS_INT64_FMT "\n",
1069 th->ih_dev, th->ih_vid, (afs_int64)th->ih_ino);
1072 /* If we're less than 0, someone presumably unlinked;
1073 don't bother setting count to 0, but we need to drop a lock */
1074 if (namei_SetLinkCount(fdP, ino, 0, 1) < 0) {
1075 FDH_REALLYCLOSE(fdP);
1081 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1083 namei_HandleToName(&name, th);
1085 code = OS_UNLINK(name.n_path);
1094 namei_inc(IHandle_t * h, Inode ino, int p1)
1100 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1101 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1102 if (type != VI_LINKTABLE)
1107 /* Get a file descriptor handle for this Inode */
1113 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0)
1118 errno = OS_ERROR(EINVAL);
1122 if (namei_SetLinkCount(fdP, ino, count, 1) < 0)
1126 FDH_REALLYCLOSE(fdP);
1133 #ifndef AFS_NT40_ENV
1135 namei_replace_file_by_hardlink(IHandle_t *hLink, IHandle_t *hTarget)
1141 /* Convert handle to file name. */
1142 namei_HandleToName(&nameLink, hLink);
1143 namei_HandleToName(&nameTarget, hTarget);
1145 OS_UNLINK(nameLink.n_path);
1146 code = link(nameTarget.n_path, nameLink.n_path);
1151 namei_copy_on_write(IHandle_t *h)
1157 struct afs_stat tstat;
1160 namei_HandleToName(&name, h);
1161 if (afs_stat(name.n_path, &tstat) < 0)
1163 if (tstat.st_nlink > 1) { /* do a copy on write */
1172 afs_snprintf(path, sizeof(path), "%s-tmp", name.n_path);
1173 fd = OS_OPEN(path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
1174 if (fd == INVALID_FD) {
1185 size = tstat.st_size;
1188 tlen = size > 8192 ? 8192 : size;
1189 if (FDH_PREAD(fdP, buf, tlen, offset) != tlen)
1191 if (OS_WRITE(fd, buf, tlen) != tlen)
1197 FDH_REALLYCLOSE(fdP);
1202 OS_UNLINK(name.n_path);
1203 code = rename(path, name.n_path);
1210 /************************************************************************
1211 * File Name Structure
1212 ************************************************************************
1214 * Each AFS file needs a unique name and it needs to be findable with
1215 * minimal lookup time. Note that the constraint on the number of files and
1216 * directories in a volume is the size of the vnode index files and the
1217 * max file size AFS supports (for internal files) of 2^31. Since a record
1218 * in the small vnode index file is 64 bytes long, we can have at most
1219 * (2^31)/64 or 33554432 files. A record in the large index file is
1220 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
1221 * Another layout parameter is that there is roughly a 16 to 1 ratio between
1222 * the number of files and the number of directories.
1224 * Using this information we can see that a layout of 256 directories, each
1225 * with 512 subdirectories and each of those having 512 files gives us
1226 * 256*512*512 = 67108864 AFS files and directories.
1228 * The volume, vnode, uniquifier and data version, as well as the tag
1229 * are required, either for finding the file or for salvaging. It's best to
1230 * restrict the name to something that can be mapped into 64 bits so the
1231 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
1232 * is used to distinguish between different versions of the same file
1233 * which are currently in the RW and clones of a volume. See "Link Table
1234 * Organization" below for more information on the tag. The tag is
1235 * required in the name of the file to ensure a unique name.
1237 * ifdef AFS_NT40_ENV
1238 * The data for each volume group is in a separate directory. The name of the
1239 * volume is of the form: Vol_NNNNNN.data, where NNNNNN is a base 32
1240 * representation of the RW volume ID (even where the RO is the only volume
1241 * on the partition). Below that are separate subdirectories for the
1242 * AFS directories and special files. There are also 16 directories for files,
1243 * hashed on the low 5 bits (recall bit0 is always 0) of the vnode number.
1244 * These directories are named:
1245 * A - P - 16 file directories.
1246 * Q ----- data directory
1247 * R ----- special files directory
1249 * The vnode is hashed into the directory using the low bits of the
1252 * The format of a file name for a regular file is:
1253 * Y:\Vol_NNNNNN.data\X\V_IIIIII.J
1254 * Y - partition encoded as drive letter, starting with D
1255 * NNNNNN - base 32 encoded volume number of RW volume
1256 * X - hash directory, as above
1257 * IIIIII - base 32 encoded vnode number
1258 * J - base 32 encoded tag
1260 * uniq is stored in the dwHighDateTime creation time field
1261 * dv is stored in the dwLowDateTime creation time field
1263 * Special inodes are always in the R directory, as above, and are
1265 * True child volid is stored in the dwHighDateTime creation time field
1266 * vnode number is always -1 (Special)
1267 * type is the IIIIII part of the filename
1268 * uniq is the J part of the filename
1269 * parent volume id is implied in the containing directory
1272 * We can store data in the uid, gid and mode bits of the files, provided
1273 * the directories have root only access. This gives us 15 bits for each
1274 * of uid and gid (GNU chown considers 65535 to mean "don't change").
1275 * There are 9 available mode bits. Adn we need to store a total of
1276 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
1277 * or 131 bits somewhere.
1279 * The format of a file name for a regular file is:
1280 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
1281 * V1 - low 8 bits of RW volume id
1282 * V2 - all bits of RW volume id
1283 * AA - high 8 bits of vnode number.
1284 * BB - next 9 bits of vnode number.
1285 * <tag><uniq><vno> - file name
1287 * Volume special files are stored in a separate directory:
1288 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
1291 * The vnode is hashed into the directory using the high bits of the
1292 * vnode number. This is so that consecutively created vnodes are in
1293 * roughly the same area on the disk. This will at least be optimal if
1294 * the user is creating many files in the same AFS directory. The name
1295 * should be formed so that the leading characters are different as quickly
1296 * as possible, leading to faster discards of incorrect matches in the
1304 /************************************************************************
1305 * Link Table Organization
1306 ************************************************************************
1308 * The link table volume special file is used to hold the link counts that
1309 * are held in the inodes in inode based AFS vice filesystems. For user
1310 * space access, the link counts are being kept in a separate
1311 * volume special file. The file begins with the usual version stamp
1312 * information and is then followed by one row per vnode number. vnode 0
1313 * is used to hold the link count of the link table itself. That is because
1314 * the same link table is shared among all the volumes of the volume group
1315 * and is deleted only when the last volume of a volume group is deleted.
1317 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
1318 * link count from 0 through 7. Each colume represents a unique instance of
1319 * that vnode. Say we have a file shared between the RW and a RO and a
1320 * different version of the file (or a different uniquifer) for the BU volume.
1321 * Then one column would be holding the link count of 2 for the RW and RO
1322 * and a different column would hold the link count of 1 for the BU volume.
1323 * # ifdef AFS_NT40_ENV
1324 * The column used is determined for NT by the uniquifier tag applied to
1325 * generate a unique file name in the NTFS namespace. The file name is
1326 * of the form "V_<vno>.<tag>" . And the <tag> is also the column number
1327 * in the link table.
1329 * Note that we allow only 5 volumes per file, giving 15 bits used in the
1333 #define LINKTABLE_WIDTH 2
1334 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
1337 * compute namei link table file and bit offset from inode number.
1339 * @param[in] ino inode number
1340 * @param[out] offset link table file offset
1341 * @param[out] index bit offset within 2-byte record
1346 namei_GetLCOffsetAndIndexFromIno(Inode ino, afs_foff_t * offset, int *index)
1348 int toff = (int)(ino & NAMEI_VNODEMASK);
1349 int tindex = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1351 *offset = (afs_foff_t) ((toff << LINKTABLE_SHIFT) + 8); /* * 2 + sizeof stamp */
1352 *index = (tindex << 1) + tindex;
1355 #ifdef AFS_PTHREAD_ENV
1356 /* XXX do static initializers work for WINNT/pthread? */
1357 pthread_mutex_t _namei_glc_lock = PTHREAD_MUTEX_INITIALIZER;
1358 #define NAMEI_GLC_LOCK MUTEX_ENTER(&_namei_glc_lock)
1359 #define NAMEI_GLC_UNLOCK MUTEX_EXIT(&_namei_glc_lock)
1360 #else /* !AFS_PTHREAD_ENV */
1361 #define NAMEI_GLC_LOCK
1362 #define NAMEI_GLC_UNLOCK
1363 #endif /* !AFS_PTHREAD_ENV */
1366 * get the link count of an inode.
1368 * @param[in] h namei link count table file handle
1369 * @param[in] ino inode number for which we are requesting a link count
1370 * @param[in] lockit if asserted, return with lock held on link table file
1371 * @param[in] fixup if asserted, write 1 to link count when read() returns
1373 * @param[in] nowrite return success on zero byte read or ZLC
1375 * @post if lockit asserted and lookup was successful, will return with write
1376 * lock on link table file descriptor
1378 * @return link count
1379 * @retval -1 namei link table i/o error
1384 namei_GetLinkCount(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite)
1386 unsigned short row = 0;
1391 /* there's no linktable yet. the salvager will create one later */
1392 if (h->fd_fd == INVALID_FD && fixup)
1394 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1397 if (FDH_LOCKFILE(h, offset) != 0)
1401 rc = FDH_PREAD(h, (char*)&row, sizeof(row), offset);
1402 if ((rc == 0 || !((row >> index) & NAMEI_TAGMASK)) && fixup && nowrite)
1404 if (rc == 0 && fixup) {
1406 * extend link table and write a link count of 1 for ino
1408 * in order to make MT-safe, truncation (extension really)
1409 * must happen under a mutex
1412 if (FDH_SIZE(h) >= offset+sizeof(row)) {
1414 goto bad_getLinkByte;
1416 FDH_TRUNC(h, offset+sizeof(row));
1418 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1421 if (rc != sizeof(row)) {
1422 goto bad_getLinkByte;
1425 if (fixup && !((row >> index) & NAMEI_TAGMASK)) {
1429 * in order to make this mt-safe, we need to do the read-modify-write
1430 * under a mutex. thus, we repeat the read inside the lock.
1433 rc = FDH_PREAD(h, (char *)&row, sizeof(row), offset);
1434 if (rc == sizeof(row)) {
1436 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1439 if (rc != sizeof(row))
1440 goto bad_getLinkByte;
1443 return (int)((row >> index) & NAMEI_TAGMASK);
1447 FDH_UNLOCKFILE(h, offset);
1452 namei_SetNonZLC(FdHandle_t * h, Inode ino)
1454 return namei_GetLinkCount(h, ino, 0, 1, 0);
1457 /* Return a free column index for this vnode. */
1459 GetFreeTag(IHandle_t * ih, int vno)
1473 /* Only one manipulates at a time. */
1474 if (FDH_LOCKFILE(fdP, offset) != 0) {
1475 FDH_REALLYCLOSE(fdP);
1479 offset = (vno << LINKTABLE_SHIFT) + 8; /* * 2 + sizeof stamp */
1481 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1482 if (nBytes != sizeof(row)) {
1488 /* Now find a free column in this row and claim it. */
1489 for (col = 0; col < NAMEI_MAXVOLS; col++) {
1490 coldata = 7 << (col * 3);
1491 if ((row & coldata) == 0)
1494 if (col >= NAMEI_MAXVOLS) {
1499 coldata = 1 << (col * 3);
1502 if (FDH_PWRITE(fdP, (char *)&row, sizeof(row), offset) != sizeof(row)) {
1506 FDH_UNLOCKFILE(fdP, offset);
1507 FDH_REALLYCLOSE(fdP);
1511 FDH_UNLOCKFILE(fdP, offset);
1512 FDH_REALLYCLOSE(fdP);
1518 /* namei_SetLinkCount
1519 * If locked is set, assume file is locked. Otherwise, lock file before
1520 * proceeding to modify it.
1523 namei_SetLinkCount(FdHandle_t * fdP, Inode ino, int count, int locked)
1529 ssize_t nBytes = -1;
1531 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1534 if (FDH_LOCKFILE(fdP, offset) != 0) {
1539 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1540 if (nBytes != sizeof(row)) {
1542 errno = OS_ERROR(EBADF);
1543 goto bad_SetLinkCount;
1548 bytesRead = 7 << index;
1550 row &= (unsigned short)~bytesRead;
1551 row |= (unsigned short)count;
1553 if (FDH_PWRITE(fdP, (char *)&row, sizeof(short), offset) != sizeof(short)) {
1554 errno = OS_ERROR(EBADF);
1555 goto bad_SetLinkCount;
1563 FDH_UNLOCKFILE(fdP, offset);
1565 /* disallowed above 7, so... */
1570 /* ListViceInodes - write inode data to a results file. */
1571 static int DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
1572 unsigned int volid);
1573 static int DecodeVolumeName(char *name, unsigned int *vid);
1574 static int namei_ListAFSSubDirs(IHandle_t * dirIH,
1575 int (*write_fun) (FILE *,
1576 struct ViceInodeInfo *,
1577 char *, char *), FILE * fp,
1578 int (*judgeFun) (struct ViceInodeInfo *,
1579 afs_uint32 vid, void *),
1580 afs_uint32 singleVolumeNumber, void *rock);
1585 * Write the inode data to the results file.
1587 * Returns -2 on error, 0 on success.
1589 * This is written as a callback simply so that other listing routines
1590 * can use the same inode reading code.
1593 WriteInodeInfo(FILE * fp, struct ViceInodeInfo *info, char *dir, char *name)
1596 n = fwrite(info, sizeof(*info), 1, fp);
1597 return (n == 1) ? 0 : -2;
1601 int mode_errors; /* Number of errors found in mode bits on directories. */
1603 VerifyDirPerms(char *path)
1605 struct afs_stat status;
1607 if (afs_stat(path, &status) < 0) {
1608 Log("Unable to stat %s. Please manually verify mode bits for this"
1609 " directory\n", path);
1611 if (((status.st_mode & 0777) != 0700) || (status.st_uid != 0))
1617 * Fill the results file with the requested inode information.
1619 * This code optimizes single volume salvages by just looking at that one
1620 * volume's directory.
1622 * @param[in] devname device name string
1623 * @param[in] moutnedOn vice partition mount point
1624 * @param[in] resultFile result file in which to write inode
1625 * metadata. If NULL, write routine is not
1627 * @param[in] judgeInode filter function pointer. if not NULL, only
1628 * inodes for which this routine returns non-
1629 * zero will be written to the results file.
1630 * @param[in] singleVolumeNumber volume id filter
1631 * @param[out] forcep always set to 0 for namei impl
1632 * @param[in] forceR not used by namei impl
1633 * @param[in] wpath not used by namei impl
1634 * @param[in] rock opaque pointer passed to judgeInode
1636 * @return operation status
1638 * @retval -1 complete failure, salvage should terminate.
1639 * @retval -2 not enough space on partition, salvager has error message
1643 ListViceInodes(char *devname, char *mountedOn, FILE *inodeFile,
1644 int (*judgeInode) (struct ViceInodeInfo * info, afs_uint32 vid, void *rock),
1645 afs_uint32 singleVolumeNumber, int *forcep, int forceR, char *wpath,
1649 struct afs_stat status;
1651 *forcep = 0; /* no need to salvage until further notice */
1653 /* Verify protections on directories. */
1655 VerifyDirPerms(mountedOn);
1658 namei_ListAFSFiles(mountedOn, WriteInodeInfo, inodeFile, judgeInode,
1659 singleVolumeNumber, rock);
1668 if (fflush(inodeFile) == EOF) {
1669 Log("Unable to successfully flush inode file for %s\n", mountedOn);
1672 if (fsync(fileno(inodeFile)) == -1) {
1673 Log("Unable to successfully fsync inode file for %s\n", mountedOn);
1678 * Paranoia: check that the file is really the right size
1680 if (afs_fstat(fileno(inodeFile), &status) == -1) {
1681 Log("Unable to successfully stat inode file for %s\n", mountedOn);
1684 if (status.st_size != ninodes * sizeof(struct ViceInodeInfo)) {
1685 Log("Wrong size (%d instead of %lu) in inode file for %s\n",
1686 (int) status.st_size,
1687 (long unsigned int) ninodes * sizeof(struct ViceInodeInfo),
1696 * Collect all the matching AFS files on the drive.
1697 * If singleVolumeNumber is non-zero, just return files for that volume.
1699 * @param[in] dev vice partition path
1700 * @param[in] writeFun function pointer to a function which
1701 * writes inode information to FILE fp
1702 * @param[in] fp file stream where inode metadata is sent
1703 * @param[in] judgeFun filter function pointer. if not NULL,
1704 * only entries for which a non-zero value
1705 * is returned are written to fp
1706 * @param[in] singleVolumeNumber volume id filter. if nonzero, only
1707 * process files for that specific volume id
1708 * @param[in] rock opaque pointer passed into writeFun and
1711 * @return operation status
1713 * @retval >=0 number of matching files found
1716 namei_ListAFSFiles(char *dev,
1717 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1720 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1721 afs_uint32 singleVolumeNumber, void *rock)
1728 #ifndef AFS_NT40_ENV
1734 static void FreeZLCList(void);
1737 memset((void *)&ih, 0, sizeof(IHandle_t));
1739 ih.ih_dev = nt_DriveToDev(dev);
1741 ih.ih_dev = volutil_GetPartitionID(dev);
1744 if (singleVolumeNumber) {
1745 ih.ih_vid = singleVolumeNumber;
1746 namei_HandleToVolDir(&name, &ih);
1748 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1749 singleVolumeNumber, rock);
1753 /* Find all volume data directories and descend through them. */
1754 namei_HandleToInodeDir(&name, &ih);
1756 dirp1 = opendir(name.n_path);
1759 while ((dp1 = readdir(dirp1))) {
1761 /* Heirarchy is one level on Windows */
1762 if (!DecodeVolumeName(dp1->d_name, &ih.ih_vid)) {
1764 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1768 if (*dp1->d_name == '.')
1770 afs_snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", name.n_path,
1772 dirp2 = opendir(path2);
1774 while ((dp2 = readdir(dirp2))) {
1775 if (*dp2->d_name == '.')
1777 if (!DecodeVolumeName(dp2->d_name, &ih.ih_vid)) {
1779 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1796 static void AddToZLCDeleteList(char dir, char *name);
1797 static void DeleteZLCFiles(char *path);
1801 * examine a namei volume special file.
1803 * @param[in] path1 volume special directory path
1804 * @param[in] dname directory entry name
1805 * @param[in] myIH inode handle to volume directory
1806 * @param[out] linkHandle namei link count fd handle. if
1807 * the inode in question is the link
1808 * table, then the FdHandle is populated
1809 * @param[in] writeFun metadata write function pointer
1810 * @param[in] fp file pointer where inode metadata
1811 * is written by (*writeFun)()
1812 * @param[in] judgeFun inode filter function pointer. if
1813 * not NULL, only inodes for which this
1814 * function returns non-zero are recorded
1815 * into fp by writeFun
1816 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1817 * inodes associated with this volume id
1818 * are recorded by writeFun
1819 * @param[in] rock opaque pointer passed to writeFun and
1822 * @return operation status
1823 * @retval 1 count this inode
1824 * @retval 0 don't count this inode
1825 * @retval -1 failure
1830 _namei_examine_special(char * path1,
1833 FdHandle_t * linkHandle,
1834 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1837 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1838 int singleVolumeNumber,
1842 struct ViceInodeInfo info;
1843 afs_uint32 inode_vgid;
1845 if (DecodeInode(path1, dname, &info, myIH->ih_vid) < 0) {
1851 inode_vgid = myIH->ih_vid;
1853 inode_vgid = (info.inodeNumber >> NAMEI_UNIQSHIFT) & NAMEI_UNIQMASK;
1856 if (info.u.param[2] != VI_LINKTABLE) {
1858 } else if ((info.u.param[0] != myIH->ih_vid) ||
1859 (inode_vgid != myIH->ih_vid)) {
1860 /* VGID encoded in linktable filename and/or OGM data isn't
1861 * consistent with VGID encoded in namei path */
1862 Log("namei_ListAFSSubDirs: warning: inconsistent linktable "
1863 "filename \"%s" OS_DIRSEP "%s\"; salvager will delete it "
1864 "(dir_vgid=%u, inode_vgid=%u, ogm_vgid=%u)\n",
1865 path1, dname, myIH->ih_vid,
1866 (unsigned int)inode_vgid,
1870 /* Open this handle */
1871 (void)afs_snprintf(path2, sizeof(path2),
1872 "%s" OS_DIRSEP "%s", path1, dname);
1873 linkHandle->fd_fd = OS_OPEN(path2, Testing ? O_RDONLY : O_RDWR, 0666);
1875 namei_GetLinkCount(linkHandle, (Inode) 0, 1, 1, Testing);
1879 (*judgeFun) (&info, singleVolumeNumber, rock)) {
1881 if ((*writeFun) (fp, &info, path1, dname) < 0) {
1891 * examine a namei file.
1893 * @param[in] path3 volume special directory path
1894 * @param[in] dname directory entry name
1895 * @param[in] myIH inode handle to volume directory
1896 * @param[in] linkHandle namei link count fd handle.
1897 * @param[in] writeFun metadata write function pointer
1898 * @param[in] fp file pointer where inode metadata
1899 * is written by (*writeFun)()
1900 * @param[in] judgeFun inode filter function pointer. if
1901 * not NULL, only inodes for which this
1902 * function returns non-zero are recorded
1903 * into fp by writeFun
1904 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1905 * inodes associated with this volume id
1906 * are recorded by writeFun
1907 * @param[in] rock opaque pointer passed to writeFun and
1910 * @return operation status
1911 * @retval 1 count this inode
1912 * @retval 0 don't count this inode
1913 * @retval -1 failure
1914 * @retval -2 request ZLC delete
1919 _namei_examine_reg(char * path3,
1922 FdHandle_t * linkHandle,
1923 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1926 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1927 int singleVolumeNumber,
1931 struct ViceInodeInfo info;
1933 int dirl; /* Windows-only (one level hash dir) */
1936 if (DecodeInode(path3, dname, &info, myIH->ih_vid) < 0) {
1941 namei_GetLinkCount(linkHandle,
1942 info.inodeNumber, 1, 1, Testing);
1943 if (info.linkCount == 0) {
1945 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n", path3, dname);
1946 #ifdef AFS_SALSRV_ENV
1947 /* defer -- the AddToZLCDeleteList() interface is not MT-safe */
1949 #else /* !AFS_SALSRV_ENV */
1950 dirl = path3[strlen(path3)-1];
1951 AddToZLCDeleteList((char)dirl, dname);
1952 #endif /* !AFS_SALSRV_ENV */
1953 #else /* !DELETE_ZLC */
1954 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path3,
1961 (*judgeFun) (&info, singleVolumeNumber, rock)) {
1963 if ((*writeFun) (fp, &info, path3, dname) < 0) {
1973 * listsubdirs work queue node.
1975 struct listsubdirs_work_node {
1976 #ifdef AFS_SALSRV_ENV
1977 int *error; /**< *error set if an error was
1978 * encountered in any listsubdirs
1982 IHandle_t * IH; /**< volume directory handle */
1983 FdHandle_t *linkHandle; /**< namei link count fd handle. when
1984 * examinining the link table special
1985 * inode, this will be pointed at the
1988 FILE * fp; /**< file pointer for writeFun */
1990 /** function which will write inode metadata to fp */
1991 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *, char *);
1993 /** inode filter function */
1994 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *);
1995 int singleVolumeNumber; /**< volume id filter */
1996 void * rock; /**< pointer passed to writeFun and judgeFun */
1997 int code; /**< return code from examine function */
1998 int special; /**< asserted when this is a volume
2003 * simple wrapper around _namei_examine_special and _namei_examine_reg.
2005 * @param[in] work the struct listsubdirs_work_node for the associated
2006 * "list subdirs" job
2007 * @param[in] dir the directory to examine
2008 * @param[in] filename the filename in 'dir' to examine
2010 * @return operation status
2011 * @retval 1 count this inode
2012 * @retval 0 don't count this inode
2013 * @retval -1 failure
2016 _namei_examine_file(const struct listsubdirs_work_node *work, char *dir,
2019 if (work->special) {
2020 return _namei_examine_special(dir, filename, work->IH,
2021 work->linkHandle, work->writeFun, work->fp,
2022 work->judgeFun, work->singleVolumeNumber,
2025 return _namei_examine_reg(dir, filename, work->IH,
2026 work->linkHandle, work->writeFun, work->fp,
2027 work->judgeFun, work->singleVolumeNumber,
2033 #ifdef AFS_SALSRV_ENV
2034 /** @addtogroup afs_vol_salsrv_pario */
2038 * arguments for the _namei_examine_file_cbk callback function.
2040 struct listsubdirs_args {
2041 const struct listsubdirs_work_node *work; /**< arguments that are the same
2042 * for all invocations of
2043 * _namei_examine_file_cbk, in
2045 int *result; /**< where we can store the return code of _namei_examine_file */
2047 char dir[512]; /**< directory to examine */
2048 char filename[256]; /**< filename in 'dir' to examine */
2052 * a node in the list of results of listsubdir jobs.
2054 struct listsubdirs_result {
2056 int inodes; /**< return value from _namei_examine_file */
2060 * clean up a list of 'struct listsubdirs_result's and interpret the results.
2062 * @param[in] resultlist a list of 'struct listsubdirs_result's
2064 * @return number of inodes found
2068 _namei_listsubdirs_cleanup_results(struct rx_queue *resultlist)
2070 struct listsubdirs_result *res, *nres;
2073 for(queue_Scan(resultlist, res, nres, listsubdirs_result)) {
2075 /* noop, retain erroneous error code */
2076 } else if (res->inodes < 0) {
2091 * wait for the spawned listsubdirs jobs to finish, and return how many inodes
2094 * @param[in] queue queue to wait to finish
2095 * @param[in] resultlist list of 'struct listsubdirs_result's that the queued
2096 * jobs are storing their results in
2098 * @return number of inodes found
2102 _namei_listsubdirs_wait(struct afs_work_queue *queue, struct rx_queue *resultlist)
2106 code = afs_wq_wait_all(queue);
2111 return _namei_listsubdirs_cleanup_results(resultlist);
2115 * work queue entry point for examining namei files.
2117 * @param[in] queue pointer to struct Vwork_queue
2118 * @param[in] node pointer to struct Vwork_queue_node
2119 * @param[in] queue_rock opaque pointer to struct salsrv_pool_state
2120 * @param[in] node_rock opaque pointer to struct listsubdirs_work_node
2121 * @param[in] caller_rock opaque pointer to struct salsrv_worker_thread_state
2123 * @return operation status
2125 * @see Vwork_queue_callback_func_t
2130 _namei_examine_file_cbk(struct afs_work_queue *queue,
2131 struct afs_work_queue_node *node,
2137 struct listsubdirs_args *args = node_rock;
2138 const struct listsubdirs_work_node * work = args->work;
2139 char *dir = args->dir;
2140 char *filename = args->filename;
2142 code = _namei_examine_file(work, dir, filename);
2144 *(args->result) = code;
2148 /* we've errored, so no point in letting more jobs continue */
2149 afs_wq_shutdown(queue);
2155 static pthread_once_t wq_once = PTHREAD_ONCE_INIT;
2156 static pthread_key_t wq_key;
2159 * create the wq_key key for storing a work queue.
2162 _namei_wq_keycreate(void)
2164 osi_Assert(pthread_key_create(&wq_key, NULL) == 0);
2168 * set the work queue for this thread to use for backgrounding namei ops.
2170 * The work queue will be used in ListAFSSubdirs (called indirectly by
2171 * ListViceInodes) to examine files in parallel.
2173 * @param[in] wq the work queue to use
2176 namei_SetWorkQueue(struct afs_work_queue *wq)
2178 osi_Assert(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2180 osi_Assert(pthread_setspecific(wq_key, wq) == 0);
2184 * enqueue an examine file work unit.
2186 * @param[in] work the _namei_examine_file arguments that are common to
2187 * all callers within the same ListAFSFiles operation
2188 * @param[in] dir the specific directory to look at (string will be
2189 * copied; can be stack/temporary memory)
2190 * @param[in] filename the filename to look at (string will be copied; can be
2191 * stack/temporary memory)
2192 * @param[in] wq work queue to enqueue this work unit to
2193 * @param[in] resultlist the list to append the 'struct listsubdirs_result' to
2194 * for the enqueued work unit
2196 * @return operation status
2198 * @retval -1 fatal error
2200 * @note errors MUST be indicated by a -1 error code and nothing else, to be
2201 * compatible with _namei_examine_reg and _namei_examine_special
2206 _namei_examine_file_spawn(const struct listsubdirs_work_node *work,
2207 const char *dir, const char *filename,
2208 struct afs_work_queue *wq,
2209 struct rx_queue *resultlist)
2212 struct listsubdirs_args *args = NULL;
2213 struct listsubdirs_result *result = NULL;
2214 struct afs_work_queue_node *node = NULL;
2215 struct afs_work_queue_add_opts opts;
2217 args = malloc(sizeof(*args));
2223 result = malloc(sizeof(*result));
2224 if (result == NULL) {
2229 code = afs_wq_node_alloc(&node);
2234 code = afs_wq_node_set_detached(node);
2241 args->result = &result->inodes;
2242 strlcpy(args->dir, dir, sizeof(args->dir));
2243 strlcpy(args->filename, filename, sizeof(args->filename));
2245 code = afs_wq_node_set_callback(node,
2246 &_namei_examine_file_cbk,
2254 afs_wq_add_opts_init(&opts);
2257 code = afs_wq_add(wq, node, &opts);
2264 queue_Append(resultlist, result);
2269 afs_wq_node_put(node);
2285 #else /* !AFS_SALSRV_ENV */
2286 # define _namei_examine_file_spawn(work, dir, file, wq, resultlist) \
2287 _namei_examine_file(work, dir, file)
2288 #endif /* !AFS_SALSRV_ENV */
2291 * traverse and check inodes.
2293 * @param[in] dirIH volume group directory handle
2294 * @param[in] writeFun function pointer which will write inode
2295 * metadata to FILE stream fp
2296 * @param[in] fp file stream where inode metadata gets
2298 * @param[in] judgeFun inode filter function. if not NULL, only
2299 * inodes for which the filter returns non-zero
2300 * will be written out by writeFun
2301 * @param[in] singleVolumeNumber volume id filter. only inodes matching this
2302 * filter are written out by writeFun
2303 * @param[in] rock opaque pointer passed to judgeFun and writeFun
2305 * @return operation status
2307 * @retval >=0 number of matching inodes found
2309 * @todo the salsrv implementation may consume a lot of
2310 * memory for a large volume. at some point we should
2311 * probably write a background thread to asynchronously
2312 * clean up the resultlist nodes to reduce memory footprint
2317 namei_ListAFSSubDirs(IHandle_t * dirIH,
2318 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
2321 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
2322 afs_uint32 singleVolumeNumber, void *rock)
2324 int code = 0, ret = 0;
2325 IHandle_t myIH = *dirIH;
2327 char path1[512], path3[512];
2329 #ifndef AFS_NT40_ENV
2334 struct dirent *dp1, *dp3;
2335 FdHandle_t linkHandle;
2337 struct listsubdirs_work_node work;
2338 #ifdef AFS_SALSRV_ENV
2340 struct afs_work_queue *wq;
2342 struct rx_queue resultlist;
2345 namei_HandleToVolDir(&name, &myIH);
2346 strlcpy(path1, name.n_path, sizeof(path1));
2348 /* Do the directory containing the special files first to pick up link
2351 (void)strcat(path1, OS_DIRSEP);
2352 (void)strcat(path1, NAMEI_SPECDIR);
2354 linkHandle.fd_fd = INVALID_FD;
2355 #ifdef AFS_SALSRV_ENV
2356 osi_Assert(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2358 wq = pthread_getspecific(wq_key);
2364 queue_Init(&resultlist);
2367 memset(&work, 0, sizeof(work));
2368 work.linkHandle = &linkHandle;
2371 work.writeFun = writeFun;
2372 work.judgeFun = judgeFun;
2373 work.singleVolumeNumber = singleVolumeNumber;
2376 #ifdef AFS_SALSRV_ENV
2377 work.error = &error;
2380 dirp1 = opendir(path1);
2382 while ((dp1 = readdir(dirp1))) {
2383 if (*dp1->d_name == '.')
2386 #ifdef AFS_SALSRV_ENV
2392 #endif /* AFS_SALSRV_ENV */
2394 code = _namei_examine_file_spawn(&work, path1, dp1->d_name, wq, &resultlist);
2404 /* count this inode */
2405 #ifndef AFS_SALSRV_ENV
2414 #ifdef AFS_SALSRV_ENV
2415 /* effectively a barrier */
2416 code = _namei_listsubdirs_wait(wq, &resultlist);
2417 if (code < 0 || error) {
2425 if (linkHandle.fd_fd == INVALID_FD) {
2426 Log("namei_ListAFSSubDirs: warning: VG %u does not have a link table; "
2427 "salvager will recreate it.\n", dirIH->ih_vid);
2430 /* Now run through all the other subdirs */
2431 namei_HandleToVolDir(&name, &myIH);
2432 strlcpy(path1, name.n_path, sizeof(path1));
2436 dirp1 = opendir(path1);
2438 while ((dp1 = readdir(dirp1))) {
2439 #ifndef AFS_NT40_ENV
2440 if (*dp1->d_name == '.')
2443 if (!strcmp(dp1->d_name, NAMEI_SPECDIR))
2446 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2447 /* Now we've got a next level subdir. */
2448 afs_snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", path1, dp1->d_name);
2449 dirp2 = opendir(path2);
2451 while ((dp2 = readdir(dirp2))) {
2452 if (*dp2->d_name == '.')
2455 /* Now we've got to the actual data */
2456 afs_snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s", path2,
2459 /* Now we've got to the actual data */
2460 afs_snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s", path1,
2463 dirp3 = opendir(path3);
2465 while ((dp3 = readdir(dirp3))) {
2466 #ifndef AFS_NT40_ENV
2467 if (*dp3->d_name == '.')
2471 #ifdef AFS_SALSRV_ENV
2474 #ifndef AFS_NT40_ENV
2481 #endif /* AFS_SALSRV_ENV */
2483 code = _namei_examine_file_spawn(&work, path3,
2490 #ifndef AFS_NT40_ENV
2498 #ifndef AFS_SALSRV_ENV
2506 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2515 #ifdef AFS_SALSRV_ENV
2516 /* effectively a barrier */
2517 code = _namei_listsubdirs_wait(wq, &resultlist);
2518 if (code < 0 || error) {
2528 /* Then why does this directory exist? Blow it away. */
2529 namei_HandleToVolDir(&name, dirIH);
2530 namei_RemoveDataDirectories(&name);
2534 #ifdef AFS_SALSRV_ENV
2536 afs_wq_wait_all(wq);
2538 _namei_listsubdirs_cleanup_results(&resultlist);
2540 if (linkHandle.fd_fd != INVALID_FD)
2541 OS_CLOSE(linkHandle.fd_fd);
2553 DecodeVolumeName(char *name, unsigned int *vid)
2555 /* Name begins with "Vol_" and ends with .data. See nt_HandleToVolDir() */
2562 if (strncmp(name, "Vol_", 4))
2564 if (strcmp(name + len - 5, ".data"))
2567 stmp[len - 5] = '\0';
2568 *vid = base32_to_int(stmp + 4);
2573 DecodeVolumeName(char *name, unsigned int *vid)
2575 if (strlen(name) < 1)
2577 *vid = (unsigned int)flipbase64_to_int64(name);
2585 * Get the inode number from the name.
2590 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2595 WIN32_FIND_DATA data;
2599 FdHandle_t linkHandle;
2602 afs_snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2604 dirH = FindFirstFileEx(fpath, FindExInfoStandard, &data,
2605 FindExSearchNameMatch, NULL,
2606 FIND_FIRST_EX_CASE_SENSITIVE);
2607 if (dirH == INVALID_HANDLE_VALUE)
2610 (void)strcpy(stmp, name);
2611 s = strrchr(stmp, '_');
2615 t = strrchr(s, '.');
2620 vno = base32_to_int(s); /* type for special files */
2621 tag = base32_to_int(t+1);
2622 info->inodeNumber = ((Inode) tag) << NAMEI_TAGSHIFT;
2623 info->inodeNumber |= vno;
2624 info->byteCount = data.nFileSizeLow;
2626 dirl = dpath[strlen(dpath)-1];
2627 if (dirl == NAMEI_SPECDIRC) { /* Special inode. */
2628 info->inodeNumber |= NAMEI_INODESPECIAL;
2629 info->u.param[0] = data.ftCreationTime.dwHighDateTime;
2630 info->u.param[1] = data.ftCreationTime.dwLowDateTime;
2631 info->u.param[2] = vno; /* type */
2632 info->u.param[3] = volid;
2633 if (vno != VI_LINKTABLE)
2634 info->linkCount = 1;
2636 /* Open this handle */
2638 (void)sprintf(lpath, "%s" OS_DIRSEP "%s", fpath, data.cFileName);
2639 linkHandle.fd_fd = OS_OPEN(lpath, O_RDONLY, 0666);
2641 namei_GetLinkCount(&linkHandle, (Inode) 0, 0, 0, 0);
2645 namei_GetLinkCount(&linkHandle, info->inodeNumber, 0, 0, 0);
2646 if (info->linkCount == 0) {
2648 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n",
2649 fpath, data.cFileName);
2650 AddToZLCDeleteList(dirl, data.cFileName);
2652 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path,
2656 info->u.param[2] = data.ftCreationTime.dwHighDateTime;
2657 info->u.param[3] = data.ftCreationTime.dwLowDateTime;
2658 info->u.param[1] = vno;
2659 info->u.param[0] = volid;
2666 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2670 struct afs_stat status;
2672 lb64_string_t check;
2674 afs_snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2676 if (afs_stat(fpath, &status) < 0) {
2680 info->byteCount = status.st_size;
2681 info->inodeNumber = (Inode) flipbase64_to_int64(name);
2683 int64_to_flipbase64(check, info->inodeNumber);
2684 if (strcmp(name, check))
2687 GetOGMFromStat(&status, &parm, &tag);
2688 if ((info->inodeNumber & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
2689 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
2690 info->u.param[0] = parm;
2691 info->u.param[1] = -1;
2692 info->u.param[2] = tag;
2693 info->u.param[3] = volid;
2695 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
2696 info->u.param[0] = volid;
2697 info->u.param[1] = (int)(info->inodeNumber & NAMEI_VNODEMASK);
2698 info->u.param[2] = (int)((info->inodeNumber >> NAMEI_UNIQSHIFT)
2699 & (Inode) NAMEI_UNIQMASK);
2700 info->u.param[3] = parm;
2707 * Convert the VolumeInfo file from RO to RW
2708 * this routine is called by namei_convertROtoRWvolume()
2711 #ifdef FSSYNC_BUILD_CLIENT
2713 convertVolumeInfo(FD_t fdr, FD_t fdw, afs_uint32 vid)
2715 struct VolumeDiskData vd;
2718 if (OS_READ(fdr, &vd, sizeof(struct VolumeDiskData)) !=
2719 sizeof(struct VolumeDiskData)) {
2720 Log("1 convertVolumeInfo: read failed for %lu with code %d\n",
2721 afs_printable_uint32_lu(vid),
2725 vd.restoredFromId = vd.id; /* remember the RO volume here */
2727 vd.id = vd.parentId;
2731 vd.uniquifier += 5000; /* just in case there are still file copies from
2732 * the old RW volume around */
2734 /* For ROs, the copyDate contains the time that the RO volume was actually
2735 * created, and the creationDate just contains the last time the RO was
2736 * copied from the RW data. So, make the new RW creationDate more accurate
2737 * by setting it to copyDate, if copyDate is older. Since, we know the
2738 * volume is at least as old as copyDate. */
2739 if (vd.copyDate < vd.creationDate) {
2740 vd.creationDate = vd.copyDate;
2742 /* If copyDate is newer, just make copyDate and creationDate the same,
2743 * for consistency with other RWs */
2744 vd.copyDate = vd.creationDate;
2747 p = strrchr(vd.name, '.');
2748 if (p && !strcmp(p, ".readonly")) {
2751 if (OS_WRITE(fdw, &vd, sizeof(struct VolumeDiskData)) !=
2752 sizeof(struct VolumeDiskData)) {
2753 Log("1 convertVolumeInfo: write failed for %lu with code %d\n",
2754 afs_printable_uint32_lu(vid),
2763 * Convert a RO-volume into a RW-volume
2765 * This function allows to recover very fast from the loss of a partition
2766 * from RO-copies if all RO-Copies exist on another partition.
2767 * Then these RO-volumes can be made to the new RW-volumes.
2768 * Backup of RW-volumes then consists in "vos release".
2770 * We must make sure in this partition exists only the RO-volume which
2771 * is typical for remote replicas.
2773 * Then the linktable is already ok,
2774 * the vnode files need to be renamed
2775 * the volinfo file needs to be replaced by another one with
2776 * slightly different contents and new name.
2777 * The volume header file of the RO-volume in the /vicep<x> directory
2778 * is destroyed by this call. A new header file for the RW-volume must
2779 * be created after return from this routine.
2783 namei_ConvertROtoRWvolume(char *pname, afs_uint32 volumeId)
2786 #ifdef FSSYNC_BUILD_CLIENT
2788 char dir_name[512], oldpath[512], newpath[512];
2803 struct DiskPartition64 *partP;
2804 struct ViceInodeInfo info;
2805 struct VolumeDiskHeader h;
2806 # ifdef AFS_DEMAND_ATTACH_FS
2808 # endif /* AFS_DEMAND_ATTACH_FS */
2810 for (partP = DiskPartitionList; partP && strcmp(partP->name, pname);
2811 partP = partP->next);
2813 Log("1 namei_ConvertROtoRWvolume: Couldn't find DiskPartition for %s\n", pname);
2818 # ifdef AFS_DEMAND_ATTACH_FS
2819 locktype = VVolLockType(V_VOLUPD, 1);
2820 code = VLockVolumeByIdNB(volumeId, partP, locktype);
2826 # endif /* AFS_DEMAND_ATTACH_FS */
2828 if (VReadVolumeDiskHeader(volumeId, partP, &h)) {
2829 Log("1 namei_ConvertROtoRWvolume: Couldn't read header for RO-volume %lu.\n",
2830 afs_printable_uint32_lu(volumeId));
2835 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_BREAKCBKS, 0, NULL);
2837 ino = namei_MakeSpecIno(h.parent, VI_LINKTABLE);
2838 IH_INIT(ih, partP->device, h.parent, ino);
2840 namei_HandleToName(&n, ih);
2841 strlcpy(dir_name, n.n_path, sizeof(dir_name));
2842 p = strrchr(dir_name, OS_DIRSEPC);
2844 dirp = opendir(dir_name);
2846 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name);
2851 while ((dp = readdir(dirp))) {
2852 /* struct ViceInodeInfo info; */
2853 #ifndef AFS_NT40_ENV
2854 if (*dp->d_name == '.')
2857 if (DecodeInode(dir_name, dp->d_name, &info, ih->ih_vid) < 0) {
2858 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s" OS_DIRSEP "%s\n",
2859 dir_name, dp->d_name);
2864 if (info.u.param[1] != -1) {
2865 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s" OS_DIRSEP "%s\n", dir_name, dp->d_name);
2870 if (info.u.param[0] != volumeId) {
2871 if (info.u.param[0] == ih->ih_vid) {
2872 if (info.u.param[2] == VI_LINKTABLE) { /* link table */
2877 Log("1 namei_ConvertROtoRWvolume: found special file %s" OS_DIRSEP "%s"
2878 " for volume %lu\n", dir_name, dp->d_name,
2879 afs_printable_uint32_lu(info.u.param[0]));
2884 if (info.u.param[2] == VI_VOLINFO) { /* volume info file */
2885 strlcpy(infoName, dp->d_name, sizeof(infoName));
2887 } else if (info.u.param[2] == VI_SMALLINDEX) { /* small vnodes file */
2888 strlcpy(smallName, dp->d_name, sizeof(smallName));
2890 } else if (info.u.param[2] == VI_LARGEINDEX) { /* large vnodes file */
2891 strlcpy(largeName, dp->d_name, sizeof(largeName));
2895 Log("1 namei_ConvertROtoRWvolume: unknown type %d of special file found : %s" OS_DIRSEP "%s\n", info.u.param[2], dir_name, dp->d_name);
2902 if (!infoSeen || !smallSeen || !largeSeen || !linkSeen) {
2903 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name);
2909 * If we come here then there was only a RO-volume and we can safely
2913 memset(&t_ih, 0, sizeof(t_ih));
2914 t_ih.ih_dev = ih->ih_dev;
2915 t_ih.ih_vid = ih->ih_vid;
2917 (void)afs_snprintf(oldpath, sizeof oldpath, "%s" OS_DIRSEP "%s", dir_name,
2919 fd = OS_OPEN(oldpath, O_RDWR, 0);
2920 if (fd == INVALID_FD) {
2921 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
2926 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_VOLINFO);
2927 namei_HandleToName(&n, &t_ih);
2928 fd2 = OS_OPEN(n.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
2929 if (fd2 == INVALID_FD) {
2930 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n.n_path);
2935 code = convertVolumeInfo(fd, fd2, ih->ih_vid);
2939 OS_UNLINK(n.n_path);
2943 SetOGM(fd2, ih->ih_vid, 1);
2946 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_SMALLINDEX);
2947 namei_HandleToName(&n, &t_ih);
2948 (void)afs_snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
2950 fd = OS_OPEN(newpath, O_RDWR, 0);
2951 if (fd == INVALID_FD) {
2952 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath);
2956 SetOGM(fd, ih->ih_vid, 2);
2959 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
2961 link(newpath, n.n_path);
2965 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_LARGEINDEX);
2966 namei_HandleToName(&n, &t_ih);
2967 (void)afs_snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
2969 fd = OS_OPEN(newpath, O_RDWR, 0);
2970 if (fd == INVALID_FD) {
2971 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath);
2975 SetOGM(fd, ih->ih_vid, 3);
2978 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
2980 link(newpath, n.n_path);
2987 h.volumeInfo_hi = h.id;
2988 h.smallVnodeIndex_hi = h.id;
2989 h.largeVnodeIndex_hi = h.id;
2990 h.linkTable_hi = h.id;
2992 if (VCreateVolumeDiskHeader(&h, partP)) {
2993 Log("1 namei_ConvertROtoRWvolume: Couldn't write header for RW-volume %lu\n",
2994 afs_printable_uint32_lu(h.id));
2999 if (VDestroyVolumeDiskHeader(partP, volumeId, h.parent)) {
3000 Log("1 namei_ConvertROtoRWvolume: Couldn't unlink header for RO-volume %lu\n",
3001 afs_printable_uint32_lu(volumeId));
3004 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_DONE, 0, NULL);
3005 FSYNC_VolOp(h.id, pname, FSYNC_VOL_ON, 0, NULL);
3008 # ifdef AFS_DEMAND_ATTACH_FS
3010 VUnlockVolumeById(volumeId, partP);
3012 # endif /* AFS_DEMAND_ATTACH_FS */
3020 * returns a static string used to print either 32 or 64 bit inode numbers.
3023 PrintInode(char *s, Inode ino)
3025 static afs_ino_str_t result;
3029 (void)afs_snprintf(s, sizeof(afs_ino_str_t), "%" AFS_UINT64_FMT, (afs_uintmax_t) ino);
3036 /* Routines to facilitate removing zero link count files. */
3037 #define MAX_ZLC_NAMES 32
3038 #define MAX_ZLC_NAMELEN 16
3039 typedef struct zlcList_s {
3040 struct zlcList_s *zlc_next;
3042 char zlc_names[MAX_ZLC_NAMES][MAX_ZLC_NAMELEN];
3045 static zlcList_t *zlcAnchor = NULL;
3046 static zlcList_t *zlcCur = NULL;
3049 AddToZLCDeleteList(char dir, char *name)
3051 osi_Assert(strlen(name) <= MAX_ZLC_NAMELEN - 3);
3053 if (!zlcCur || zlcCur->zlc_n >= MAX_ZLC_NAMES) {
3054 if (zlcCur && zlcCur->zlc_next)
3055 zlcCur = zlcCur->zlc_next;
3057 zlcList_t *tmp = (zlcList_t *) malloc(sizeof(zlcList_t));
3063 zlcCur->zlc_next = tmp;
3067 zlcCur->zlc_next = NULL;
3072 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%c" OS_DIRSEP "%s", dir, name);
3074 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%s", name);
3080 DeleteZLCFiles(char *path)
3086 for (z = zlcAnchor; z; z = z->zlc_next) {
3087 for (i = 0; i < z->zlc_n; i++) {
3089 (void)sprintf(fname, "%s" OS_DIRSEP "%s", path, z->zlc_names[i]);
3091 (void)sprintf(fname, "%s", z->zlc_names[i]);
3092 if (namei_unlink(fname) < 0) {
3093 Log("ZLC: Can't unlink %s, dos error = %d\n", fname,
3097 z->zlc_n = 0; /* Can reuse space. */
3110 tnext = i->zlc_next;
3114 zlcCur = zlcAnchor = NULL;
3118 #endif /* AFS_NAMEI_ENV */