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>
61 /*@+fcnmacros +macrofcndecl@*/
64 #endif /*S_SPLINT_S */
65 #define afs_stat stat64
66 #define afs_fstat fstat64
68 #define afs_open nt_open
70 #define afs_open open64
72 #define afs_fopen fopen64
73 #else /* !O_LARGEFILE */
75 #endif /*S_SPLINT_S */
77 #define afs_fstat fstat
79 #define afs_open nt_open
83 #define afs_fopen fopen
84 #endif /* !O_LARGEFILE */
85 /*@=fcnmacros =macrofcndecl@*/
88 #define LOCK_SH 1 /* shared lock */
89 #define LOCK_EX 2 /* exclusive lock */
90 #define LOCK_NB 4 /* don't block when locking */
91 #define LOCK_UN 8 /* unlock */
96 #include <afs/work_queue.h>
97 #include <afs/thread_pool.h>
98 #include <vol/vol-salvage.h>
101 #if !defined(HAVE_FLOCK) && !defined(AFS_NT40_ENV)
105 * This function emulates a subset of flock()
108 emul_flock(int fd, int cmd)
111 memset(&f, 0, sizeof (f));
120 return fcntl(fd, (cmd & LOCK_NB) ? F_SETLK : F_SETLKW, &f);
123 #define flock(f,c) emul_flock(f,c)
130 namei_iread(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
139 nBytes = FDH_PREAD(fdP, buf, size, offset);
145 namei_iwrite(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
154 nBytes = FDH_PWRITE(fdP, buf, size, offset);
160 /* Inode number format:
161 * low 32 bits - if a regular file or directory, the vnode; else the type.
162 * 32-36 - uniquifier tag and index into counts array for this vnode. Only
163 * two of the available bits are currently used. The rest are
164 * present in case we ever increase the number of types of volumes
165 * in the volume group.
166 * bit 37 : 1 == special, 0 == regular
168 # define NAMEI_VNODEMASK 0x00ffffffff
169 # define NAMEI_TAGSHIFT 32
170 # define NAMEI_INODESPECIAL 0x2000000000
171 # define NAMEI_SPECDIR "R"
172 # define NAMEI_SPECDIRC 'R'
173 #else /* !AFS_NT40_ENV */
174 /* Inode number format:
175 * low 26 bits - vnode number - all 1's if volume special file.
177 * next 3 bits spare (0's)
178 * high 32 bits - uniquifier (regular) or type if spare
180 # define NAMEI_VNODEMASK 0x003ffffff
181 # define NAMEI_TAGSHIFT 26
182 # define NAMEI_UNIQMASK 0xffffffff
183 # define NAMEI_UNIQSHIFT 32
184 # define NAMEI_INODESPECIAL ((Inode)NAMEI_VNODEMASK)
185 /* dir1 is the high 8 bits of the 26 bit vnode */
186 # define VNO_DIR1(vno) ((vno >> 14) & 0xff)
187 /* dir2 is the next 9 bits */
188 # define VNO_DIR2(vno) ((vno >> 9) & 0x1ff)
189 /* "name" is the low 9 bits of the vnode, the 3 bit tag and the uniq */
190 # define NAMEI_SPECDIR "special"
191 #endif /* !AFS_NT40_ENV */
192 #define NAMEI_TAGMASK 0x7
193 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
195 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
197 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
198 * group, not counting temporary (move) volumes.
199 * This is the number of separate files, all having
200 * the same vnode number, which can occur in a volume
212 static int GetFreeTag(IHandle_t * ih, int vno);
214 /* namei_HandleToInodeDir
216 * Construct the path name of the directory holding the inode data.
217 * Format: /<vicepx>/INODEDIR
222 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
224 memset(name, '\0', sizeof(*name));
225 nt_DevToDrive(name->n_drive, ih->ih_dev);
226 strlcpy(name->n_path, name->n_drive, sizeof(name->n_path));
230 /* Format: /<vicepx>/INODEDIR */
231 #define PNAME_BLEN 64
233 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
237 memset(name, '\0', sizeof(*name));
240 * Add the /vicepXX string to the start of name->n_base and then calculate
241 * offset as the number of bytes we know we added.
243 * FIXME: This embeds knowledge of the vice partition naming scheme and
244 * mapping from device numbers. There needs to be an API that tells us
247 volutil_PartitionName_r(ih->ih_dev, name->n_base, sizeof(name->n_base));
248 offset = VICE_PREFIX_SIZE + (ih->ih_dev > 25 ? 2 : 1);
249 name->n_base[offset] = OS_DIRSEPC;
251 strlcpy(name->n_base + offset, INODEDIR, sizeof(name->n_base) - offset);
252 strlcpy(name->n_path, name->n_base, sizeof(name->n_path));
256 #define addtoname(N, C) \
258 if ((N)->n_path[strlen((N)->n_path)-1] != OS_DIRSEPC) \
259 strlcat((N)->n_path, OS_DIRSEP, sizeof((N)->n_path)); \
260 strlcat((N)->n_path, (C), sizeof((N)->n_path)); \
266 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
268 /* X:\Vol_XXXXXXX.data */
272 namei_HandleToInodeDir(name, ih);
273 /* nt_drive added to name by namei_HandleToInodeDir() */
274 namep = name->n_voldir;
275 (void)memcpy(namep, "\\Vol_", 5);
277 (void)strcpy(namep, int_to_base32(str1, ih->ih_vid));
278 namep += strlen(namep);
279 memcpy(namep, ".data", 5);
282 addtoname(name, name->n_voldir);
286 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
290 namei_HandleToInodeDir(name, ih);
291 (void)int32_to_flipbase64(tmp, (int64_t) (ih->ih_vid & 0xff));
292 strlcpy(name->n_voldir1, tmp, sizeof(name->n_voldir1));
293 addtoname(name, name->n_voldir1);
294 (void)int32_to_flipbase64(tmp, (int64_t) ih->ih_vid);
295 strlcpy(name->n_voldir2, tmp, sizeof(name->n_voldir2));
296 addtoname(name, name->n_voldir2);
300 /* namei_HandleToName
302 * Constructs a file name for the fully qualified handle.
305 /* Note that special files end up in X:\Vol_XXXXXXX.data\R */
307 namei_HandleToName(namei_t * name, IHandle_t * ih)
309 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
310 int special = (ih->ih_ino & NAMEI_INODESPECIAL)?1:0;
311 int tag = (int)((ih->ih_ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
314 namei_HandleToVolDir(name, ih);
317 name->n_dir[0] = NAMEI_SPECDIRC;
320 name->n_dir[0] = 'Q';
322 name->n_dir[0] = ((vno & 0x1f) >> 1) + 'A';
325 name->n_dir[1] = '\0';
326 addtoname(name, name->n_dir);
327 /* X:\Vol_XXXXXXX.data\X\V_XXXXXXX.XXX */
328 namep = name->n_inode;
329 (void)memcpy(namep, "\\V_", 3);
331 (void)strcpy(namep, int_to_base32(str1, vno));
332 namep += strlen(namep);
334 (void)strcpy(namep, int_to_base32(str1, tag));
335 namep += strlen(namep);
336 addtoname(name, name->n_inode);
339 /* Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special */
341 namei_HandleToName(namei_t * name, IHandle_t * ih)
343 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
346 namei_HandleToVolDir(name, ih);
348 if (vno == NAMEI_VNODESPECIAL) {
349 strlcpy(name->n_dir1, NAMEI_SPECDIR, sizeof(name->n_dir1));
350 addtoname(name, name->n_dir1);
351 name->n_dir2[0] = '\0';
353 (void)int32_to_flipbase64(str, VNO_DIR1(vno));
354 strlcpy(name->n_dir1, str, sizeof(name->n_dir1));
355 addtoname(name, name->n_dir1);
356 (void)int32_to_flipbase64(str, VNO_DIR2(vno));
357 strlcpy(name->n_dir2, str, sizeof(name->n_dir2));
358 addtoname(name, name->n_dir2);
360 (void)int64_to_flipbase64(str, (int64_t) ih->ih_ino);
361 strlcpy(name->n_inode, str, sizeof(name->n_inode));
362 addtoname(name, name->n_inode);
367 /* The following is a warning to tell sys-admins to not muck about in this
370 #define VICE_README "These files and directories are a part of the AFS \
371 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
372 ownership and permissions included.\n"
374 namei_ViceREADME(char *partition)
379 /* Create the inode directory if we're starting for the first time */
380 (void)afs_snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s", partition,
382 mkdir(filename, 0700);
384 (void)afs_snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s" OS_DIRSEP "README",
385 partition, INODEDIR);
386 fd = afs_open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0444);
388 (void)write(fd, VICE_README, strlen(VICE_README));
395 /* namei_CreateDataDirectories
397 * If creating the file failed because of ENOENT or ENOTDIR, try
398 * creating all the directories first.
402 namei_CreateDataDirectories(namei_t * name, int *created)
409 afs_snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
411 if (mkdir(tmp) < 0) {
422 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
424 if (mkdir(tmp) < 0 && errno != EEXIST)
430 #define create_dir() \
432 if (mkdir(tmp, 0700)<0) { \
433 if (errno != EEXIST) \
441 #define create_nextdir(A) \
443 strcat(tmp, OS_DIRSEP); strcat(tmp, A); create_dir(); \
447 namei_CreateDataDirectories(namei_t * name, int *created)
453 strlcpy(tmp, name->n_base, sizeof(tmp));
456 create_nextdir(name->n_voldir1);
457 create_nextdir(name->n_voldir2);
458 create_nextdir(name->n_dir1);
459 if (name->n_dir2[0]) {
460 create_nextdir(name->n_dir2);
467 /* delTree(): Deletes an entire tree of directories (no files)
469 * root : Full path to the subtree. Should be big enough for PATH_MAX
470 * tree : the subtree to be deleted is rooted here. Specifies only the
471 * subtree beginning at tree (not the entire path). It should be
472 * a pointer into the "root" buffer.
474 * errp : errno of the first error encountered during the directory cleanup.
475 * *errp should have been initialized to 0.
478 * -1 : If errors were encountered during cleanup and error is set to
482 * If there are errors, we try to work around them and delete as many
483 * directories as possible. We don't attempt to remove directories that still
484 * have non-dir entries in them.
487 delTree(char *root, char *tree, int *errp)
495 /* delete the children first */
496 cp = strchr(tree, OS_DIRSEPC);
498 delTree(root, cp + 1, errp);
501 cp = tree + strlen(tree); /* move cp to the end of string tree */
503 /* now delete all entries in this dir */
504 if ((ds = opendir(root)) != (DIR *) NULL) {
506 while ((dirp = readdir(ds))) {
507 /* ignore . and .. */
508 if (!strcmp(dirp->d_name, ".") || !strcmp(dirp->d_name, ".."))
510 /* since root is big enough, we reuse the space to
511 * concatenate the dirname to the current tree
513 strcat(root, OS_DIRSEP);
514 strcat(root, dirp->d_name);
515 if (afs_stat(root, &st) == 0 && S_ISDIR(st.st_mode)) {
516 /* delete this subtree */
517 delTree(root, cp + 1, errp);
519 *errp = *errp ? *errp : errno;
521 /* recover path to our cur tree by truncating it to
526 /* if (!errno) -- closedir not implicit if we got an error */
530 /* finally axe the current dir */
532 *errp = *errp ? *errp : errno;
534 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
536 #endif /* !AFS_PTHREAD_ENV */
541 /* if we encountered errors during cleanup, we return a -1 */
550 /* namei_RemoveDataDirectories
552 * Returns 0 on success.
553 * Returns -1 on error. Typically, callers ignore this error because we
554 * can continue running if the removes fail. The salvage process will
555 * finish tidying up for us.
560 namei_RemoveDataDirectories(namei_t * name)
567 afs_snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
570 path += strlen(path);
571 *path++ = OS_DIRSEPC;
573 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
575 if (rmdir(name->n_path) < 0 && errno != ENOENT)
580 /* Delete the Vol_NNNNNN.data directory. */
583 if (rmdir(name->n_path) < 0 && errno != ENOENT) {
591 * We only use the n_base and n_voldir1 entries
592 * and only do rmdir's.
595 namei_RemoveDataDirectories(namei_t * name)
599 int prefixlen = strlen(name->n_base), err = 0;
600 int vollen = strlen(name->n_voldir1);
601 char pbuf[MAXPATHLEN];
605 strlcpy(path, name->n_path, sizeof(pbuf));
607 /* move past the prefix and n_voldir1 */
608 path = path + prefixlen + 1 + vollen + 1; /* skip over the trailing / */
610 /* now delete all dirs upto path */
611 code = delTree(pbuf, path, &err);
613 /* We've now deleted everything under /n_base/n_voldir1/n_voldir2 that
614 * we could. Do not delete /n_base/n_voldir1, since doing such might
615 * interrupt another thread trying to create a volume. We could introduce
616 * some locking to make this safe (or only remove it for whole-partition
617 * salvages), but by not deleting it we only leave behind a maximum of
618 * 256 empty directories. So at least for now, don't bother. */
623 /* Create the file in the name space.
625 * Parameters stored as follows:
627 * p1 - volid - implied in containing directory.
628 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
629 * p3 - uniq -- bits 4-0 are in mode bits 4-0
630 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
632 * p1 - volid - creation time - dwHighDateTime
633 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
634 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
635 * p4 - parid - parent volume id - implied in containing directory.
637 * Return value is the inode number or (Inode)-1 if error.
638 * We "know" there is only one link table, so return EEXIST if there already
639 * is a link table. It's up to the calling code to test errno and increment
645 * This function is called by VCreateVolume to hide the implementation
646 * details of the inode numbers. This only allows for 7 volume special
647 * types, but if we get that far, this could should be dead by then.
650 namei_MakeSpecIno(int volid, int type)
653 ino = NAMEI_INODESPECIAL;
656 /* tag is always 0 for special */
658 type &= NAMEI_TAGMASK;
659 ino |= ((Inode) type) << NAMEI_TAGSHIFT;
660 ino |= ((Inode) volid) << NAMEI_UNIQSHIFT;
668 SetOGM(FD_t fd, int parm, int tag)
674 CheckOGM(namei_t *name, FdHandle_t *fdP, int p1)
676 WIN32_FIND_DATA info;
680 FindFirstFileEx(name->n_path, FindExInfoStandard, &info,
681 FindExSearchNameMatch, NULL,
682 FIND_FIRST_EX_CASE_SENSITIVE);
685 return -1; /* Can't get info, leave alone */
689 if (info.ftCreationTime.dwHighDateTime != (unsigned int)p1)
694 #else /* AFS_NT40_ENV */
695 /* SetOGM - set owner group and mode bits from parm and tag */
697 SetOGM(FD_t fd, int parm, int tag)
700 * owner - low 15 bits of parm.
701 * group - next 15 bits of parm.
702 * mode - 2 bits of parm, then lowest = 3 bits of tag.
704 int owner, group, mode;
706 owner = parm & 0x7fff;
707 group = (parm >> 15) & 0x7fff;
708 if (fchown(fd, owner, group) < 0)
711 mode = (parm >> 27) & 0x18;
713 if (fchmod(fd, mode) < 0)
718 /* GetOGM - get parm and tag from owner, group and mode bits. */
720 GetOGMFromStat(struct afs_stat *status, int *parm, int *tag)
722 *parm = status->st_uid | (status->st_gid << 15);
723 *parm |= (status->st_mode & 0x18) << 27;
724 *tag = status->st_mode & 0x7;
728 CheckOGM(namei_t *name, FdHandle_t *fdP, int p1)
730 struct afs_stat status;
732 if (afs_fstat(fdP->fd_fd, &status) < 0)
735 GetOGMFromStat(&status, &parm, &tag);
741 #endif /* !AFS_NT40_ENV */
743 int big_vno = 0; /* Just in case we ever do 64 bit vnodes. */
745 /* Derive the name and create it O_EXCL. If that fails we have an error.
746 * Get the tag from a free column in the link table.
750 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
753 FD_t fd = INVALID_FD;
764 memset((void *)&tmp, 0, sizeof(IHandle_t));
765 memset(&tfd, 0, sizeof(FdHandle_t));
767 tmp.ih_dev = nt_DriveToDev(part);
768 if (tmp.ih_dev == -1) {
773 if (p2 == INODESPECIAL) {
774 /* Parameters for special file:
775 * p1 - volume id - goes into owner/group/mode
776 * p2 - vnode == INODESPECIAL
778 * p4 - parent volume id
780 ftime.dwHighDateTime = p1;
781 ftime.dwLowDateTime = p2;
783 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
784 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
786 int vno = p2 & NAMEI_VNODEMASK;
787 /* Parameters for regular file:
801 tmp.ih_ino = (Inode) p2;
802 ftime.dwHighDateTime = p3;
803 ftime.dwLowDateTime = p4;
806 namei_HandleToName(&name, &tmp);
807 p = strrchr((char *)&name.n_path, '.');
809 for (tag = 0; tag < NAMEI_MAXVOLS; tag++) {
810 *p = *int_to_base32(str1, tag);
811 fd = afs_open((char *)&name.n_path, O_CREAT | O_RDWR | O_TRUNC | O_EXCL, 0666);
812 if (fd == INVALID_FD) {
813 if (errno == ENOTDIR || errno == ENOENT) {
814 if (namei_CreateDataDirectories(&name, &created_dir) == 0)
815 fd = afs_open((char *)&name.n_path, O_CREAT | O_RDWR | O_TRUNC | O_EXCL, 0666);
819 if (fd != INVALID_FD)
821 if (p2 == INODESPECIAL && p3 == VI_LINKTABLE)
824 if (fd == INVALID_FD) {
828 tmp.ih_ino &= ~(((Inode) NAMEI_TAGMASK) << NAMEI_TAGSHIFT);
829 tmp.ih_ino |= (((Inode) tag) << NAMEI_TAGSHIFT);
832 if (!SetFileTime((HANDLE) fd, &ftime, NULL, NULL)) {
833 errno = OS_ERROR(EBADF);
839 if (p2 != INODESPECIAL) {
840 if (fd == INVALID_FD) {
842 code = nt_unlink((char *)&name.n_path);
851 code = namei_SetLinkCount(fdP, tmp.ih_ino, 1, 0);
853 } else if (p2 == INODESPECIAL && p3 == VI_LINKTABLE) {
854 if (fd == INVALID_FD)
856 /* hack at tmp to setup for set link count call. */
858 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
863 if (fd != INVALID_FD)
866 if (code || (fd == INVALID_FD)) {
867 if (p2 != INODESPECIAL) {
870 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
876 int save_errno = errno;
877 namei_RemoveDataDirectories(&name);
881 return (code || (fd == INVALID_FD)) ? (Inode) -1 : tmp.ih_ino;
883 #else /* !AFS_NT40_ENV */
885 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
897 memset((void *)&tmp, 0, sizeof(IHandle_t));
898 memset(&tfd, 0, sizeof(FdHandle_t));
900 tmp.ih_dev = volutil_GetPartitionID(part);
901 if (tmp.ih_dev == -1) {
907 /* Parameters for special file:
908 * p1 - volume id - goes into owner/group/mode
911 * p4 - parent volume id
916 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
917 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
919 int vno = p2 & NAMEI_VNODEMASK;
920 /* Parameters for regular file:
932 /* If GetFreeTag succeeds, it atomically sets link count to 1. */
933 tag = GetFreeTag(lh, p2);
938 tmp.ih_ino = (Inode) p2;
939 /* name is <uniq(p3)><tag><vno(p2)> */
940 tmp.ih_ino |= ((Inode) tag) << NAMEI_TAGSHIFT;
941 tmp.ih_ino |= ((Inode) p3) << NAMEI_UNIQSHIFT;
946 namei_HandleToName(&name, &tmp);
947 fd = afs_open(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
949 if (errno == ENOTDIR || errno == ENOENT) {
950 if (namei_CreateDataDirectories(&name, &created_dir) < 0)
952 fd = afs_open(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR,
960 if (SetOGM(fd, ogm_parm, tag) < 0) {
966 if (p2 == (afs_uint32)-1 && p3 == VI_LINKTABLE) {
967 /* hack at tmp to setup for set link count call. */
968 memset((void *)&tfd, 0, sizeof(FdHandle_t)); /* minimalistic still, but a little cleaner */
971 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
979 if (code || (fd < 0)) {
983 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
988 return (code || (fd < 0)) ? (Inode) - 1 : tmp.ih_ino;
994 namei_iopen(IHandle_t * h)
999 /* Convert handle to file name. */
1000 namei_HandleToName(&name, h);
1001 fd = afs_open((char *)&name.n_path, O_RDWR, 0666);
1005 /* Need to detect vol special file and just unlink. In those cases, the
1006 * handle passed in _is_ for the inode. We only check p1 for the special
1010 namei_dec(IHandle_t * ih, Inode ino, int p1)
1017 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1019 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1021 /* Verify this is the right file. */
1022 IH_INIT(tmp, ih->ih_dev, ih->ih_vid, ino);
1024 namei_HandleToName(&name, tmp);
1029 errno = OS_ERROR(ENOENT);
1033 if (CheckOGM(&name, fdP, p1) < 0) {
1034 FDH_REALLYCLOSE(fdP);
1036 errno = OS_ERROR(EINVAL);
1040 /* If it's the link table itself, decrement the link count. */
1041 if (type == VI_LINKTABLE) {
1042 if ((count = namei_GetLinkCount(fdP, (Inode) 0, 1, 0, 1)) < 0) {
1043 FDH_REALLYCLOSE(fdP);
1049 if (namei_SetLinkCount(fdP, (Inode) 0, count < 0 ? 0 : count, 1) <
1051 FDH_REALLYCLOSE(fdP);
1057 FDH_REALLYCLOSE(fdP);
1063 if ((code = OS_UNLINK(name.n_path)) == 0) {
1064 if (type == VI_LINKTABLE) {
1065 /* Try to remove directory. If it fails, that's ok.
1066 * Salvage will clean up.
1068 (void)namei_RemoveDataDirectories(&name);
1071 FDH_REALLYCLOSE(fdP);
1074 /* Get a file descriptor handle for this Inode */
1080 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0) {
1081 FDH_REALLYCLOSE(fdP);
1087 if (namei_SetLinkCount(fdP, ino, count, 1) < 0) {
1088 FDH_REALLYCLOSE(fdP);
1093 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1094 Log("Warning: Lost ref on ihandle dev %d vid %d ino %" AFS_INT64_FMT "\n",
1095 th->ih_dev, th->ih_vid, (afs_int64)th->ih_ino);
1098 /* If we're less than 0, someone presumably unlinked;
1099 don't bother setting count to 0, but we need to drop a lock */
1100 if (namei_SetLinkCount(fdP, ino, 0, 1) < 0) {
1101 FDH_REALLYCLOSE(fdP);
1107 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1109 namei_HandleToName(&name, th);
1111 code = OS_UNLINK(name.n_path);
1120 namei_inc(IHandle_t * h, Inode ino, int p1)
1126 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1127 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1128 if (type != VI_LINKTABLE)
1133 /* Get a file descriptor handle for this Inode */
1139 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0)
1144 errno = OS_ERROR(EINVAL);
1148 if (namei_SetLinkCount(fdP, ino, count, 1) < 0)
1152 FDH_REALLYCLOSE(fdP);
1159 #ifndef AFS_NT40_ENV
1161 namei_replace_file_by_hardlink(IHandle_t *hLink, IHandle_t *hTarget)
1167 /* Convert handle to file name. */
1168 namei_HandleToName(&nameLink, hLink);
1169 namei_HandleToName(&nameTarget, hTarget);
1171 OS_UNLINK(nameLink.n_path);
1172 code = link(nameTarget.n_path, nameLink.n_path);
1177 namei_copy_on_write(IHandle_t *h)
1179 afs_int32 fd, code = 0;
1182 struct afs_stat tstat;
1185 namei_HandleToName(&name, h);
1186 if (afs_stat(name.n_path, &tstat) < 0)
1188 if (tstat.st_nlink > 1) { /* do a copy on write */
1197 afs_snprintf(path, sizeof(path), "%s-tmp", name.n_path);
1198 fd = afs_open(path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
1210 size = tstat.st_size;
1213 tlen = size > 8192 ? 8192 : size;
1214 if (FDH_PREAD(fdP, buf, tlen, offset) != tlen)
1216 if (write(fd, buf, tlen) != tlen)
1222 FDH_REALLYCLOSE(fdP);
1227 OS_UNLINK(name.n_path);
1228 code = rename(path, name.n_path);
1235 /************************************************************************
1236 * File Name Structure
1237 ************************************************************************
1239 * Each AFS file needs a unique name and it needs to be findable with
1240 * minimal lookup time. Note that the constraint on the number of files and
1241 * directories in a volume is the size of the vnode index files and the
1242 * max file size AFS supports (for internal files) of 2^31. Since a record
1243 * in the small vnode index file is 64 bytes long, we can have at most
1244 * (2^31)/64 or 33554432 files. A record in the large index file is
1245 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
1246 * Another layout parameter is that there is roughly a 16 to 1 ratio between
1247 * the number of files and the number of directories.
1249 * Using this information we can see that a layout of 256 directories, each
1250 * with 512 subdirectories and each of those having 512 files gives us
1251 * 256*512*512 = 67108864 AFS files and directories.
1253 * The volume, vnode, uniquifier and data version, as well as the tag
1254 * are required, either for finding the file or for salvaging. It's best to
1255 * restrict the name to something that can be mapped into 64 bits so the
1256 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
1257 * is used to distinguish between different versions of the same file
1258 * which are currently in the RW and clones of a volume. See "Link Table
1259 * Organization" below for more information on the tag. The tag is
1260 * required in the name of the file to ensure a unique name.
1262 * ifdef AFS_NT40_ENV
1263 * The data for each volume group is in a separate directory. The name of the
1264 * volume is of the form: Vol_NNNNNN.data, where NNNNNN is a base 32
1265 * representation of the RW volume ID (even where the RO is the only volume
1266 * on the partition). Below that are separate subdirectories for the
1267 * AFS directories and special files. There are also 16 directories for files,
1268 * hashed on the low 5 bits (recall bit0 is always 0) of the vnode number.
1269 * These directories are named:
1270 * A - P - 16 file directories.
1271 * Q ----- data directory
1272 * R ----- special files directory
1274 * The vnode is hashed into the directory using the low bits of the
1277 * The format of a file name for a regular file is:
1278 * Y:\Vol_NNNNNN.data\X\V_IIIIII.J
1279 * Y - partition encoded as drive letter, starting with D
1280 * NNNNNN - base 32 encoded volume number of RW volume
1281 * X - hash directory, as above
1282 * IIIIII - base 32 encoded vnode number
1283 * J - base 32 encoded tag
1285 * uniq is stored in the dwHighDateTime creation time field
1286 * dv is stored in the dwLowDateTime creation time field
1288 * Special inodes are always in the R directory, as above, and are
1290 * True child volid is stored in the dwHighDateTime creation time field
1291 * vnode number is always -1 (Special)
1292 * type is the IIIIII part of the filename
1293 * uniq is the J part of the filename
1294 * parent volume id is implied in the containing directory
1297 * We can store data in the uid, gid and mode bits of the files, provided
1298 * the directories have root only access. This gives us 15 bits for each
1299 * of uid and gid (GNU chown considers 65535 to mean "don't change").
1300 * There are 9 available mode bits. Adn we need to store a total of
1301 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
1302 * or 131 bits somewhere.
1304 * The format of a file name for a regular file is:
1305 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
1306 * V1 - low 8 bits of RW volume id
1307 * V2 - all bits of RW volume id
1308 * AA - high 8 bits of vnode number.
1309 * BB - next 9 bits of vnode number.
1310 * <tag><uniq><vno> - file name
1312 * Volume special files are stored in a separate directory:
1313 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
1316 * The vnode is hashed into the directory using the high bits of the
1317 * vnode number. This is so that consecutively created vnodes are in
1318 * roughly the same area on the disk. This will at least be optimal if
1319 * the user is creating many files in the same AFS directory. The name
1320 * should be formed so that the leading characters are different as quickly
1321 * as possible, leading to faster discards of incorrect matches in the
1329 /************************************************************************
1330 * Link Table Organization
1331 ************************************************************************
1333 * The link table volume special file is used to hold the link counts that
1334 * are held in the inodes in inode based AFS vice filesystems. For user
1335 * space access, the link counts are being kept in a separate
1336 * volume special file. The file begins with the usual version stamp
1337 * information and is then followed by one row per vnode number. vnode 0
1338 * is used to hold the link count of the link table itself. That is because
1339 * the same link table is shared among all the volumes of the volume group
1340 * and is deleted only when the last volume of a volume group is deleted.
1342 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
1343 * link count from 0 through 7. Each colume represents a unique instance of
1344 * that vnode. Say we have a file shared between the RW and a RO and a
1345 * different version of the file (or a different uniquifer) for the BU volume.
1346 * Then one column would be holding the link count of 2 for the RW and RO
1347 * and a different column would hold the link count of 1 for the BU volume.
1348 * # ifdef AFS_NT40_ENV
1349 * The column used is determined for NT by the uniquifier tag applied to
1350 * generate a unique file name in the NTFS namespace. The file name is
1351 * of the form "V_<vno>.<tag>" . And the <tag> is also the column number
1352 * in the link table.
1354 * Note that we allow only 5 volumes per file, giving 15 bits used in the
1358 #define LINKTABLE_WIDTH 2
1359 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
1362 * compute namei link table file and bit offset from inode number.
1364 * @param[in] ino inode number
1365 * @param[out] offset link table file offset
1366 * @param[out] index bit offset within 2-byte record
1371 namei_GetLCOffsetAndIndexFromIno(Inode ino, afs_foff_t * offset, int *index)
1373 int toff = (int)(ino & NAMEI_VNODEMASK);
1374 int tindex = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1376 *offset = (afs_foff_t) ((toff << LINKTABLE_SHIFT) + 8); /* * 2 + sizeof stamp */
1377 *index = (tindex << 1) + tindex;
1380 #ifdef AFS_PTHREAD_ENV
1381 /* XXX do static initializers work for WINNT/pthread? */
1382 pthread_mutex_t _namei_glc_lock = PTHREAD_MUTEX_INITIALIZER;
1383 #define NAMEI_GLC_LOCK MUTEX_ENTER(&_namei_glc_lock)
1384 #define NAMEI_GLC_UNLOCK MUTEX_EXIT(&_namei_glc_lock)
1385 #else /* !AFS_PTHREAD_ENV */
1386 #define NAMEI_GLC_LOCK
1387 #define NAMEI_GLC_UNLOCK
1388 #endif /* !AFS_PTHREAD_ENV */
1391 * get the link count of an inode.
1393 * @param[in] h namei link count table file handle
1394 * @param[in] ino inode number for which we are requesting a link count
1395 * @param[in] lockit if asserted, return with lock held on link table file
1396 * @param[in] fixup if asserted, write 1 to link count when read() returns
1398 * @param[in] nowrite return success on zero byte read or ZLC
1400 * @post if lockit asserted and lookup was successful, will return with write
1401 * lock on link table file descriptor
1403 * @return link count
1404 * @retval -1 namei link table i/o error
1409 namei_GetLinkCount(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite)
1411 unsigned short row = 0;
1416 /* there's no linktable yet. the salvager will create one later */
1417 if (h->fd_fd == INVALID_FD && fixup)
1419 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1422 if (FDH_LOCKFILE(h, offset) != 0)
1426 rc = FDH_PREAD(h, (char*)&row, sizeof(row), offset);
1427 if ((rc == 0 || !((row >> index) & NAMEI_TAGMASK)) && fixup && nowrite)
1429 if (rc == 0 && fixup) {
1431 * extend link table and write a link count of 1 for ino
1433 * in order to make MT-safe, truncation (extension really)
1434 * must happen under a mutex
1437 if (FDH_SIZE(h) >= offset+sizeof(row)) {
1439 goto bad_getLinkByte;
1441 FDH_TRUNC(h, offset+sizeof(row));
1443 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1446 if (rc != sizeof(row)) {
1447 goto bad_getLinkByte;
1450 if (fixup && !((row >> index) & NAMEI_TAGMASK)) {
1454 * in order to make this mt-safe, we need to do the read-modify-write
1455 * under a mutex. thus, we repeat the read inside the lock.
1458 rc = FDH_PREAD(h, (char *)&row, sizeof(row), offset);
1459 if (rc == sizeof(row)) {
1461 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1464 if (rc != sizeof(row))
1465 goto bad_getLinkByte;
1468 return (int)((row >> index) & NAMEI_TAGMASK);
1472 FDH_UNLOCKFILE(h, offset);
1477 namei_SetNonZLC(FdHandle_t * h, Inode ino)
1479 return namei_GetLinkCount(h, ino, 0, 1, 0);
1482 /* Return a free column index for this vnode. */
1484 GetFreeTag(IHandle_t * ih, int vno)
1498 /* Only one manipulates at a time. */
1499 if (FDH_LOCKFILE(fdP, offset) != 0) {
1500 FDH_REALLYCLOSE(fdP);
1504 offset = (vno << LINKTABLE_SHIFT) + 8; /* * 2 + sizeof stamp */
1506 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1507 if (nBytes != sizeof(row)) {
1513 /* Now find a free column in this row and claim it. */
1514 for (col = 0; col < NAMEI_MAXVOLS; col++) {
1515 coldata = 7 << (col * 3);
1516 if ((row & coldata) == 0)
1519 if (col >= NAMEI_MAXVOLS) {
1524 coldata = 1 << (col * 3);
1527 if (FDH_PWRITE(fdP, (char *)&row, sizeof(row), offset) != sizeof(row)) {
1531 FDH_UNLOCKFILE(fdP, offset);
1532 FDH_REALLYCLOSE(fdP);
1536 FDH_UNLOCKFILE(fdP, offset);
1537 FDH_REALLYCLOSE(fdP);
1543 /* namei_SetLinkCount
1544 * If locked is set, assume file is locked. Otherwise, lock file before
1545 * proceeding to modify it.
1548 namei_SetLinkCount(FdHandle_t * fdP, Inode ino, int count, int locked)
1554 ssize_t nBytes = -1;
1556 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1559 if (FDH_LOCKFILE(fdP, offset) != 0) {
1564 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1565 if (nBytes != sizeof(row)) {
1567 errno = OS_ERROR(EBADF);
1568 goto bad_SetLinkCount;
1573 bytesRead = 7 << index;
1575 row &= (unsigned short)~bytesRead;
1576 row |= (unsigned short)count;
1578 if (FDH_PWRITE(fdP, (char *)&row, sizeof(short), offset) != sizeof(short)) {
1579 errno = OS_ERROR(EBADF);
1580 goto bad_SetLinkCount;
1588 FDH_UNLOCKFILE(fdP, offset);
1590 /* disallowed above 7, so... */
1595 /* ListViceInodes - write inode data to a results file. */
1596 static int DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
1597 unsigned int volid);
1598 static int DecodeVolumeName(char *name, unsigned int *vid);
1599 static int namei_ListAFSSubDirs(IHandle_t * dirIH,
1600 int (*write_fun) (FILE *,
1601 struct ViceInodeInfo *,
1602 char *, char *), FILE * fp,
1603 int (*judgeFun) (struct ViceInodeInfo *,
1604 afs_uint32 vid, void *),
1605 afs_uint32 singleVolumeNumber, void *rock);
1610 * Write the inode data to the results file.
1612 * Returns -2 on error, 0 on success.
1614 * This is written as a callback simply so that other listing routines
1615 * can use the same inode reading code.
1618 WriteInodeInfo(FILE * fp, struct ViceInodeInfo *info, char *dir, char *name)
1621 n = fwrite(info, sizeof(*info), 1, fp);
1622 return (n == 1) ? 0 : -2;
1626 int mode_errors; /* Number of errors found in mode bits on directories. */
1628 VerifyDirPerms(char *path)
1630 struct afs_stat status;
1632 if (afs_stat(path, &status) < 0) {
1633 Log("Unable to stat %s. Please manually verify mode bits for this"
1634 " directory\n", path);
1636 if (((status.st_mode & 0777) != 0700) || (status.st_uid != 0))
1642 * Fill the results file with the requested inode information.
1644 * This code optimizes single volume salvages by just looking at that one
1645 * volume's directory.
1647 * @param[in] devname device name string
1648 * @param[in] moutnedOn vice partition mount point
1649 * @param[in] resultFile result file in which to write inode
1650 * metadata. If NULL, write routine is not
1652 * @param[in] judgeInode filter function pointer. if not NULL, only
1653 * inodes for which this routine returns non-
1654 * zero will be written to the results file.
1655 * @param[in] singleVolumeNumber volume id filter
1656 * @param[out] forcep always set to 0 for namei impl
1657 * @param[in] forceR not used by namei impl
1658 * @param[in] wpath not used by namei impl
1659 * @param[in] rock opaque pointer passed to judgeInode
1661 * @return operation status
1663 * @retval -1 complete failure, salvage should terminate.
1664 * @retval -2 not enough space on partition, salvager has error message
1668 ListViceInodes(char *devname, char *mountedOn, FILE *inodeFile,
1669 int (*judgeInode) (struct ViceInodeInfo * info, afs_uint32 vid, void *rock),
1670 afs_uint32 singleVolumeNumber, int *forcep, int forceR, char *wpath,
1674 struct afs_stat status;
1676 *forcep = 0; /* no need to salvage until further notice */
1678 /* Verify protections on directories. */
1680 VerifyDirPerms(mountedOn);
1683 namei_ListAFSFiles(mountedOn, WriteInodeInfo, inodeFile, judgeInode,
1684 singleVolumeNumber, rock);
1693 if (fflush(inodeFile) == EOF) {
1694 Log("Unable to successfully flush inode file for %s\n", mountedOn);
1697 if (fsync(fileno(inodeFile)) == -1) {
1698 Log("Unable to successfully fsync inode file for %s\n", mountedOn);
1703 * Paranoia: check that the file is really the right size
1705 if (afs_fstat(fileno(inodeFile), &status) == -1) {
1706 Log("Unable to successfully stat inode file for %s\n", mountedOn);
1709 if (status.st_size != ninodes * sizeof(struct ViceInodeInfo)) {
1710 Log("Wrong size (%d instead of %lu) in inode file for %s\n",
1711 (int) status.st_size,
1712 (long unsigned int) ninodes * sizeof(struct ViceInodeInfo),
1721 * Collect all the matching AFS files on the drive.
1722 * If singleVolumeNumber is non-zero, just return files for that volume.
1724 * @param[in] dev vice partition path
1725 * @param[in] writeFun function pointer to a function which
1726 * writes inode information to FILE fp
1727 * @param[in] fp file stream where inode metadata is sent
1728 * @param[in] judgeFun filter function pointer. if not NULL,
1729 * only entries for which a non-zero value
1730 * is returned are written to fp
1731 * @param[in] singleVolumeNumber volume id filter. if nonzero, only
1732 * process files for that specific volume id
1733 * @param[in] rock opaque pointer passed into writeFun and
1736 * @return operation status
1738 * @retval >=0 number of matching files found
1741 namei_ListAFSFiles(char *dev,
1742 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1745 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1746 afs_uint32 singleVolumeNumber, void *rock)
1753 #ifndef AFS_NT40_ENV
1759 static void FreeZLCList(void);
1762 memset((void *)&ih, 0, sizeof(IHandle_t));
1764 ih.ih_dev = nt_DriveToDev(dev);
1766 ih.ih_dev = volutil_GetPartitionID(dev);
1769 if (singleVolumeNumber) {
1770 ih.ih_vid = singleVolumeNumber;
1771 namei_HandleToVolDir(&name, &ih);
1773 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1774 singleVolumeNumber, rock);
1778 /* Find all volume data directories and descend through them. */
1779 namei_HandleToInodeDir(&name, &ih);
1781 dirp1 = opendir(name.n_path);
1784 while ((dp1 = readdir(dirp1))) {
1786 /* Heirarchy is one level on Windows */
1787 if (!DecodeVolumeName(dp1->d_name, &ih.ih_vid)) {
1789 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1793 if (*dp1->d_name == '.')
1795 afs_snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", name.n_path,
1797 dirp2 = opendir(path2);
1799 while ((dp2 = readdir(dirp2))) {
1800 if (*dp2->d_name == '.')
1802 if (!DecodeVolumeName(dp2->d_name, &ih.ih_vid)) {
1804 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1821 static void AddToZLCDeleteList(char dir, char *name);
1822 static void DeleteZLCFiles(char *path);
1826 * examine a namei volume special file.
1828 * @param[in] path1 volume special directory path
1829 * @param[in] dname directory entry name
1830 * @param[in] myIH inode handle to volume directory
1831 * @param[out] linkHandle namei link count fd handle. if
1832 * the inode in question is the link
1833 * table, then the FdHandle is populated
1834 * @param[in] writeFun metadata write function pointer
1835 * @param[in] fp file pointer where inode metadata
1836 * is written by (*writeFun)()
1837 * @param[in] judgeFun inode filter function pointer. if
1838 * not NULL, only inodes for which this
1839 * function returns non-zero are recorded
1840 * into fp by writeFun
1841 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1842 * inodes associated with this volume id
1843 * are recorded by writeFun
1844 * @param[in] rock opaque pointer passed to writeFun and
1847 * @return operation status
1848 * @retval 1 count this inode
1849 * @retval 0 don't count this inode
1850 * @retval -1 failure
1855 _namei_examine_special(char * path1,
1858 FdHandle_t * linkHandle,
1859 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1862 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1863 int singleVolumeNumber,
1867 struct ViceInodeInfo info;
1868 afs_uint32 inode_vgid;
1870 if (DecodeInode(path1, dname, &info, myIH->ih_vid) < 0) {
1876 inode_vgid = myIH->ih_vid;
1878 inode_vgid = (info.inodeNumber >> NAMEI_UNIQSHIFT) & NAMEI_UNIQMASK;
1881 if (info.u.param[2] != VI_LINKTABLE) {
1883 } else if ((info.u.param[0] != myIH->ih_vid) ||
1884 (inode_vgid != myIH->ih_vid)) {
1885 /* VGID encoded in linktable filename and/or OGM data isn't
1886 * consistent with VGID encoded in namei path */
1887 Log("namei_ListAFSSubDirs: warning: inconsistent linktable "
1888 "filename \"%s" OS_DIRSEP "%s\"; salvager will delete it "
1889 "(dir_vgid=%u, inode_vgid=%u, ogm_vgid=%u)\n",
1890 path1, dname, myIH->ih_vid,
1891 (unsigned int)inode_vgid,
1895 /* Open this handle */
1896 (void)afs_snprintf(path2, sizeof(path2),
1897 "%s" OS_DIRSEP "%s", path1, dname);
1898 linkHandle->fd_fd = afs_open(path2, Testing ? O_RDONLY : O_RDWR, 0666);
1900 namei_GetLinkCount(linkHandle, (Inode) 0, 1, 1, Testing);
1904 (*judgeFun) (&info, singleVolumeNumber, rock)) {
1906 if ((*writeFun) (fp, &info, path1, dname) < 0) {
1916 * examine a namei file.
1918 * @param[in] path3 volume special directory path
1919 * @param[in] dname directory entry name
1920 * @param[in] myIH inode handle to volume directory
1921 * @param[in] linkHandle namei link count fd handle.
1922 * @param[in] writeFun metadata write function pointer
1923 * @param[in] fp file pointer where inode metadata
1924 * is written by (*writeFun)()
1925 * @param[in] judgeFun inode filter function pointer. if
1926 * not NULL, only inodes for which this
1927 * function returns non-zero are recorded
1928 * into fp by writeFun
1929 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1930 * inodes associated with this volume id
1931 * are recorded by writeFun
1932 * @param[in] rock opaque pointer passed to writeFun and
1935 * @return operation status
1936 * @retval 1 count this inode
1937 * @retval 0 don't count this inode
1938 * @retval -1 failure
1939 * @retval -2 request ZLC delete
1944 _namei_examine_reg(char * path3,
1947 FdHandle_t * linkHandle,
1948 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1951 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1952 int singleVolumeNumber,
1956 struct ViceInodeInfo info;
1958 int dirl; /* Windows-only (one level hash dir) */
1961 if (DecodeInode(path3, dname, &info, myIH->ih_vid) < 0) {
1966 namei_GetLinkCount(linkHandle,
1967 info.inodeNumber, 1, 1, Testing);
1968 if (info.linkCount == 0) {
1970 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n", path3, dname);
1971 #ifdef AFS_SALSRV_ENV
1972 /* defer -- the AddToZLCDeleteList() interface is not MT-safe */
1974 #else /* !AFS_SALSRV_ENV */
1975 dirl = path3[strlen(path3)-1];
1976 AddToZLCDeleteList((char)dirl, dname);
1977 #endif /* !AFS_SALSRV_ENV */
1978 #else /* !DELETE_ZLC */
1979 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path3,
1986 (*judgeFun) (&info, singleVolumeNumber, rock)) {
1988 if ((*writeFun) (fp, &info, path3, dname) < 0) {
1998 * listsubdirs work queue node.
2000 struct listsubdirs_work_node {
2001 #ifdef AFS_SALSRV_ENV
2002 int *error; /**< *error set if an error was
2003 * encountered in any listsubdirs
2007 IHandle_t * IH; /**< volume directory handle */
2008 FdHandle_t *linkHandle; /**< namei link count fd handle. when
2009 * examinining the link table special
2010 * inode, this will be pointed at the
2013 FILE * fp; /**< file pointer for writeFun */
2015 /** function which will write inode metadata to fp */
2016 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *, char *);
2018 /** inode filter function */
2019 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *);
2020 int singleVolumeNumber; /**< volume id filter */
2021 void * rock; /**< pointer passed to writeFun and judgeFun */
2022 int code; /**< return code from examine function */
2023 int special; /**< asserted when this is a volume
2028 * simple wrapper around _namei_examine_special and _namei_examine_reg.
2030 * @param[in] work the struct listsubdirs_work_node for the associated
2031 * "list subdirs" job
2032 * @param[in] dir the directory to examine
2033 * @param[in] filename the filename in 'dir' to examine
2035 * @return operation status
2036 * @retval 1 count this inode
2037 * @retval 0 don't count this inode
2038 * @retval -1 failure
2041 _namei_examine_file(const struct listsubdirs_work_node *work, char *dir,
2044 if (work->special) {
2045 return _namei_examine_special(dir, filename, work->IH,
2046 work->linkHandle, work->writeFun, work->fp,
2047 work->judgeFun, work->singleVolumeNumber,
2050 return _namei_examine_reg(dir, filename, work->IH,
2051 work->linkHandle, work->writeFun, work->fp,
2052 work->judgeFun, work->singleVolumeNumber,
2058 #ifdef AFS_SALSRV_ENV
2059 /** @addtogroup afs_vol_salsrv_pario */
2063 * arguments for the _namei_examine_file_cbk callback function.
2065 struct listsubdirs_args {
2066 const struct listsubdirs_work_node *work; /**< arguments that are the same
2067 * for all invocations of
2068 * _namei_examine_file_cbk, in
2070 int *result; /**< where we can store the return code of _namei_examine_file */
2072 char dir[512]; /**< directory to examine */
2073 char filename[256]; /**< filename in 'dir' to examine */
2077 * a node in the list of results of listsubdir jobs.
2079 struct listsubdirs_result {
2081 int inodes; /**< return value from _namei_examine_file */
2085 * clean up a list of 'struct listsubdirs_result's and interpret the results.
2087 * @param[in] resultlist a list of 'struct listsubdirs_result's
2089 * @return number of inodes found
2093 _namei_listsubdirs_cleanup_results(struct rx_queue *resultlist)
2095 struct listsubdirs_result *res, *nres;
2098 for(queue_Scan(resultlist, res, nres, listsubdirs_result)) {
2100 /* noop, retain erroneous error code */
2101 } else if (res->inodes < 0) {
2116 * wait for the spawned listsubdirs jobs to finish, and return how many inodes
2119 * @param[in] queue queue to wait to finish
2120 * @param[in] resultlist list of 'struct listsubdirs_result's that the queued
2121 * jobs are storing their results in
2123 * @return number of inodes found
2127 _namei_listsubdirs_wait(struct afs_work_queue *queue, struct rx_queue *resultlist)
2131 code = afs_wq_wait_all(queue);
2136 return _namei_listsubdirs_cleanup_results(resultlist);
2140 * work queue entry point for examining namei files.
2142 * @param[in] queue pointer to struct Vwork_queue
2143 * @param[in] node pointer to struct Vwork_queue_node
2144 * @param[in] queue_rock opaque pointer to struct salsrv_pool_state
2145 * @param[in] node_rock opaque pointer to struct listsubdirs_work_node
2146 * @param[in] caller_rock opaque pointer to struct salsrv_worker_thread_state
2148 * @return operation status
2150 * @see Vwork_queue_callback_func_t
2155 _namei_examine_file_cbk(struct afs_work_queue *queue,
2156 struct afs_work_queue_node *node,
2162 struct listsubdirs_args *args = node_rock;
2163 const struct listsubdirs_work_node * work = args->work;
2164 char *dir = args->dir;
2165 char *filename = args->filename;
2167 code = _namei_examine_file(work, dir, filename);
2169 *(args->result) = code;
2173 /* we've errored, so no point in letting more jobs continue */
2174 afs_wq_shutdown(queue);
2180 static pthread_once_t wq_once = PTHREAD_ONCE_INIT;
2181 static pthread_key_t wq_key;
2184 * create the wq_key key for storing a work queue.
2187 _namei_wq_keycreate(void)
2189 osi_Assert(pthread_key_create(&wq_key, NULL) == 0);
2193 * set the work queue for this thread to use for backgrounding namei ops.
2195 * The work queue will be used in ListAFSSubdirs (called indirectly by
2196 * ListViceInodes) to examine files in parallel.
2198 * @param[in] wq the work queue to use
2201 namei_SetWorkQueue(struct afs_work_queue *wq)
2203 osi_Assert(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2205 osi_Assert(pthread_setspecific(wq_key, wq) == 0);
2209 * enqueue an examine file work unit.
2211 * @param[in] work the _namei_examine_file arguments that are common to
2212 * all callers within the same ListAFSFiles operation
2213 * @param[in] dir the specific directory to look at (string will be
2214 * copied; can be stack/temporary memory)
2215 * @param[in] filename the filename to look at (string will be copied; can be
2216 * stack/temporary memory)
2217 * @param[in] wq work queue to enqueue this work unit to
2218 * @param[in] resultlist the list to append the 'struct listsubdirs_result' to
2219 * for the enqueued work unit
2221 * @return operation status
2223 * @retval -1 fatal error
2225 * @note errors MUST be indicated by a -1 error code and nothing else, to be
2226 * compatible with _namei_examine_reg and _namei_examine_special
2231 _namei_examine_file_spawn(const struct listsubdirs_work_node *work,
2232 const char *dir, const char *filename,
2233 struct afs_work_queue *wq,
2234 struct rx_queue *resultlist)
2237 struct listsubdirs_args *args = NULL;
2238 struct listsubdirs_result *result = NULL;
2239 struct afs_work_queue_node *node = NULL;
2240 struct afs_work_queue_add_opts opts;
2242 args = malloc(sizeof(*args));
2248 result = malloc(sizeof(*result));
2249 if (result == NULL) {
2254 code = afs_wq_node_alloc(&node);
2259 code = afs_wq_node_set_detached(node);
2266 args->result = &result->inodes;
2267 strlcpy(args->dir, dir, sizeof(args->dir));
2268 strlcpy(args->filename, filename, sizeof(args->filename));
2270 code = afs_wq_node_set_callback(node,
2271 &_namei_examine_file_cbk,
2279 afs_wq_add_opts_init(&opts);
2282 code = afs_wq_add(wq, node, &opts);
2289 queue_Append(resultlist, result);
2294 afs_wq_node_put(node);
2310 #else /* !AFS_SALSRV_ENV */
2311 # define _namei_examine_file_spawn(work, dir, file, wq, resultlist) \
2312 _namei_examine_file(work, dir, file)
2313 #endif /* !AFS_SALSRV_ENV */
2316 * traverse and check inodes.
2318 * @param[in] dirIH volume group directory handle
2319 * @param[in] writeFun function pointer which will write inode
2320 * metadata to FILE stream fp
2321 * @param[in] fp file stream where inode metadata gets
2323 * @param[in] judgeFun inode filter function. if not NULL, only
2324 * inodes for which the filter returns non-zero
2325 * will be written out by writeFun
2326 * @param[in] singleVolumeNumber volume id filter. only inodes matching this
2327 * filter are written out by writeFun
2328 * @param[in] rock opaque pointer passed to judgeFun and writeFun
2330 * @return operation status
2332 * @retval >=0 number of matching inodes found
2334 * @todo the salsrv implementation may consume a lot of
2335 * memory for a large volume. at some point we should
2336 * probably write a background thread to asynchronously
2337 * clean up the resultlist nodes to reduce memory footprint
2342 namei_ListAFSSubDirs(IHandle_t * dirIH,
2343 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
2346 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
2347 afs_uint32 singleVolumeNumber, void *rock)
2349 int code = 0, ret = 0;
2350 IHandle_t myIH = *dirIH;
2352 char path1[512], path3[512];
2354 #ifndef AFS_NT40_ENV
2359 struct dirent *dp1, *dp3;
2360 FdHandle_t linkHandle;
2362 struct listsubdirs_work_node work;
2363 #ifdef AFS_SALSRV_ENV
2365 struct afs_work_queue *wq;
2367 struct rx_queue resultlist;
2370 namei_HandleToVolDir(&name, &myIH);
2371 strlcpy(path1, name.n_path, sizeof(path1));
2373 /* Do the directory containing the special files first to pick up link
2376 (void)strcat(path1, OS_DIRSEP);
2377 (void)strcat(path1, NAMEI_SPECDIR);
2379 linkHandle.fd_fd = INVALID_FD;
2380 #ifdef AFS_SALSRV_ENV
2381 osi_Assert(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2383 wq = pthread_getspecific(wq_key);
2389 queue_Init(&resultlist);
2392 memset(&work, 0, sizeof(work));
2393 work.linkHandle = &linkHandle;
2396 work.writeFun = writeFun;
2397 work.judgeFun = judgeFun;
2398 work.singleVolumeNumber = singleVolumeNumber;
2401 #ifdef AFS_SALSRV_ENV
2402 work.error = &error;
2405 dirp1 = opendir(path1);
2407 while ((dp1 = readdir(dirp1))) {
2408 if (*dp1->d_name == '.')
2411 #ifdef AFS_SALSRV_ENV
2417 #endif /* AFS_SALSRV_ENV */
2419 code = _namei_examine_file_spawn(&work, path1, dp1->d_name, wq, &resultlist);
2429 /* count this inode */
2430 #ifndef AFS_SALSRV_ENV
2439 #ifdef AFS_SALSRV_ENV
2440 /* effectively a barrier */
2441 code = _namei_listsubdirs_wait(wq, &resultlist);
2442 if (code < 0 || error) {
2450 if (linkHandle.fd_fd == INVALID_FD) {
2451 Log("namei_ListAFSSubDirs: warning: VG %u does not have a link table; "
2452 "salvager will recreate it.\n", dirIH->ih_vid);
2455 /* Now run through all the other subdirs */
2456 namei_HandleToVolDir(&name, &myIH);
2457 strlcpy(path1, name.n_path, sizeof(path1));
2461 dirp1 = opendir(path1);
2463 while ((dp1 = readdir(dirp1))) {
2464 #ifndef AFS_NT40_ENV
2465 if (*dp1->d_name == '.')
2468 if (!strcmp(dp1->d_name, NAMEI_SPECDIR))
2471 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2472 /* Now we've got a next level subdir. */
2473 afs_snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", path1, dp1->d_name);
2474 dirp2 = opendir(path2);
2476 while ((dp2 = readdir(dirp2))) {
2477 if (*dp2->d_name == '.')
2480 /* Now we've got to the actual data */
2481 afs_snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s", path2,
2484 /* Now we've got to the actual data */
2485 afs_snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s", path1,
2488 dirp3 = opendir(path3);
2490 while ((dp3 = readdir(dirp3))) {
2491 #ifndef AFS_NT40_ENV
2492 if (*dp3->d_name == '.')
2496 #ifdef AFS_SALSRV_ENV
2499 #ifndef AFS_NT40_ENV
2506 #endif /* AFS_SALSRV_ENV */
2508 code = _namei_examine_file_spawn(&work, path3,
2515 #ifndef AFS_NT40_ENV
2523 #ifndef AFS_SALSRV_ENV
2531 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2540 #ifdef AFS_SALSRV_ENV
2541 /* effectively a barrier */
2542 code = _namei_listsubdirs_wait(wq, &resultlist);
2543 if (code < 0 || error) {
2553 /* Then why does this directory exist? Blow it away. */
2554 namei_HandleToVolDir(&name, dirIH);
2555 namei_RemoveDataDirectories(&name);
2559 #ifdef AFS_SALSRV_ENV
2561 afs_wq_wait_all(wq);
2563 _namei_listsubdirs_cleanup_results(&resultlist);
2565 if (linkHandle.fd_fd != INVALID_FD)
2566 OS_CLOSE(linkHandle.fd_fd);
2578 DecodeVolumeName(char *name, unsigned int *vid)
2580 /* Name begins with "Vol_" and ends with .data. See nt_HandleToVolDir() */
2587 if (strncmp(name, "Vol_", 4))
2589 if (strcmp(name + len - 5, ".data"))
2592 stmp[len - 5] = '\0';
2593 *vid = base32_to_int(stmp + 4);
2598 DecodeVolumeName(char *name, unsigned int *vid)
2600 if (strlen(name) < 1)
2602 *vid = (unsigned int)flipbase64_to_int64(name);
2610 * Get the inode number from the name.
2615 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2620 WIN32_FIND_DATA data;
2624 FdHandle_t linkHandle;
2627 afs_snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2629 dirH = FindFirstFileEx(fpath, FindExInfoStandard, &data,
2630 FindExSearchNameMatch, NULL,
2631 FIND_FIRST_EX_CASE_SENSITIVE);
2632 if (dirH == INVALID_HANDLE_VALUE)
2635 (void)strcpy(stmp, name);
2636 s = strrchr(stmp, '_');
2640 t = strrchr(s, '.');
2645 vno = base32_to_int(s); /* type for special files */
2646 tag = base32_to_int(t+1);
2647 info->inodeNumber = ((Inode) tag) << NAMEI_TAGSHIFT;
2648 info->inodeNumber |= vno;
2649 info->byteCount = data.nFileSizeLow;
2651 dirl = dpath[strlen(dpath)-1];
2652 if (dirl == NAMEI_SPECDIRC) { /* Special inode. */
2653 info->inodeNumber |= NAMEI_INODESPECIAL;
2654 info->u.param[0] = data.ftCreationTime.dwHighDateTime;
2655 info->u.param[1] = data.ftCreationTime.dwLowDateTime;
2656 info->u.param[2] = vno; /* type */
2657 info->u.param[3] = volid;
2658 if (vno != VI_LINKTABLE)
2659 info->linkCount = 1;
2661 /* Open this handle */
2663 (void)sprintf(lpath, "%s" OS_DIRSEP "%s", fpath, data.cFileName);
2664 linkHandle.fd_fd = afs_open(lpath, O_RDONLY, 0666);
2666 namei_GetLinkCount(&linkHandle, (Inode) 0, 0, 0, 0);
2670 namei_GetLinkCount(&linkHandle, info->inodeNumber, 0, 0, 0);
2671 if (info->linkCount == 0) {
2673 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n",
2674 fpath, data.cFileName);
2675 AddToZLCDeleteList(dirl, data.cFileName);
2677 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path,
2681 info->u.param[2] = data.ftCreationTime.dwHighDateTime;
2682 info->u.param[3] = data.ftCreationTime.dwLowDateTime;
2683 info->u.param[1] = vno;
2684 info->u.param[0] = volid;
2691 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2695 struct afs_stat status;
2697 lb64_string_t check;
2699 afs_snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2701 if (afs_stat(fpath, &status) < 0) {
2705 info->byteCount = status.st_size;
2706 info->inodeNumber = (Inode) flipbase64_to_int64(name);
2708 int64_to_flipbase64(check, info->inodeNumber);
2709 if (strcmp(name, check))
2712 GetOGMFromStat(&status, &parm, &tag);
2713 if ((info->inodeNumber & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
2714 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
2715 info->u.param[0] = parm;
2716 info->u.param[1] = -1;
2717 info->u.param[2] = tag;
2718 info->u.param[3] = volid;
2720 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
2721 info->u.param[0] = volid;
2722 info->u.param[1] = (int)(info->inodeNumber & NAMEI_VNODEMASK);
2723 info->u.param[2] = (int)((info->inodeNumber >> NAMEI_UNIQSHIFT)
2724 & (Inode) NAMEI_UNIQMASK);
2725 info->u.param[3] = parm;
2732 * Convert the VolumeInfo file from RO to RW
2733 * this routine is called by namei_convertROtoRWvolume()
2736 #ifdef FSSYNC_BUILD_CLIENT
2738 convertVolumeInfo(FD_t fdr, FD_t fdw, afs_uint32 vid)
2740 struct VolumeDiskData vd;
2743 if (OS_READ(fdr, &vd, sizeof(struct VolumeDiskData)) !=
2744 sizeof(struct VolumeDiskData)) {
2745 Log("1 convertVolumeInfo: read failed for %lu with code %d\n",
2746 afs_printable_uint32_lu(vid),
2750 vd.restoredFromId = vd.id; /* remember the RO volume here */
2752 vd.id = vd.parentId;
2756 vd.uniquifier += 5000; /* just in case there are still file copies from
2757 * the old RW volume around */
2759 /* For ROs, the copyDate contains the time that the RO volume was actually
2760 * created, and the creationDate just contains the last time the RO was
2761 * copied from the RW data. So, make the new RW creationDate more accurate
2762 * by setting it to copyDate, if copyDate is older. Since, we know the
2763 * volume is at least as old as copyDate. */
2764 if (vd.copyDate < vd.creationDate) {
2765 vd.creationDate = vd.copyDate;
2767 /* If copyDate is newer, just make copyDate and creationDate the same,
2768 * for consistency with other RWs */
2769 vd.copyDate = vd.creationDate;
2772 p = strrchr(vd.name, '.');
2773 if (p && !strcmp(p, ".readonly")) {
2776 if (OS_WRITE(fdw, &vd, sizeof(struct VolumeDiskData)) !=
2777 sizeof(struct VolumeDiskData)) {
2778 Log("1 convertVolumeInfo: write failed for %lu with code %d\n",
2779 afs_printable_uint32_lu(vid),
2788 * Convert a RO-volume into a RW-volume
2790 * This function allows to recover very fast from the loss of a partition
2791 * from RO-copies if all RO-Copies exist on another partition.
2792 * Then these RO-volumes can be made to the new RW-volumes.
2793 * Backup of RW-volumes then consists in "vos release".
2795 * We must make sure in this partition exists only the RO-volume which
2796 * is typical for remote replicas.
2798 * Then the linktable is already ok,
2799 * the vnode files need to be renamed
2800 * the volinfo file needs to be replaced by another one with
2801 * slightly different contents and new name.
2802 * The volume header file of the RO-volume in the /vicep<x> directory
2803 * is destroyed by this call. A new header file for the RW-volume must
2804 * be created after return from this routine.
2808 namei_ConvertROtoRWvolume(char *pname, afs_uint32 volumeId)
2811 #ifdef FSSYNC_BUILD_CLIENT
2813 char dir_name[512], oldpath[512], newpath[512];
2828 struct DiskPartition64 *partP;
2829 struct ViceInodeInfo info;
2830 struct VolumeDiskHeader h;
2831 # ifdef AFS_DEMAND_ATTACH_FS
2833 # endif /* AFS_DEMAND_ATTACH_FS */
2835 for (partP = DiskPartitionList; partP && strcmp(partP->name, pname);
2836 partP = partP->next);
2838 Log("1 namei_ConvertROtoRWvolume: Couldn't find DiskPartition for %s\n", pname);
2843 # ifdef AFS_DEMAND_ATTACH_FS
2844 locktype = VVolLockType(V_VOLUPD, 1);
2845 code = VLockVolumeByIdNB(volumeId, partP, locktype);
2851 # endif /* AFS_DEMAND_ATTACH_FS */
2853 if (VReadVolumeDiskHeader(volumeId, partP, &h)) {
2854 Log("1 namei_ConvertROtoRWvolume: Couldn't read header for RO-volume %lu.\n",
2855 afs_printable_uint32_lu(volumeId));
2860 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_BREAKCBKS, 0, NULL);
2862 ino = namei_MakeSpecIno(h.parent, VI_LINKTABLE);
2863 IH_INIT(ih, partP->device, h.parent, ino);
2865 namei_HandleToName(&n, ih);
2866 strlcpy(dir_name, n.n_path, sizeof(dir_name));
2867 p = strrchr(dir_name, OS_DIRSEPC);
2869 dirp = opendir(dir_name);
2871 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name);
2876 while ((dp = readdir(dirp))) {
2877 /* struct ViceInodeInfo info; */
2878 #ifndef AFS_NT40_ENV
2879 if (*dp->d_name == '.')
2882 if (DecodeInode(dir_name, dp->d_name, &info, ih->ih_vid) < 0) {
2883 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s" OS_DIRSEP "%s\n",
2884 dir_name, dp->d_name);
2889 if (info.u.param[1] != -1) {
2890 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s" OS_DIRSEP "%s\n", dir_name, dp->d_name);
2895 if (info.u.param[0] != volumeId) {
2896 if (info.u.param[0] == ih->ih_vid) {
2897 if (info.u.param[2] == VI_LINKTABLE) { /* link table */
2902 Log("1 namei_ConvertROtoRWvolume: found special file %s" OS_DIRSEP "%s"
2903 " for volume %lu\n", dir_name, dp->d_name,
2904 afs_printable_uint32_lu(info.u.param[0]));
2909 if (info.u.param[2] == VI_VOLINFO) { /* volume info file */
2910 strlcpy(infoName, dp->d_name, sizeof(infoName));
2912 } else if (info.u.param[2] == VI_SMALLINDEX) { /* small vnodes file */
2913 strlcpy(smallName, dp->d_name, sizeof(smallName));
2915 } else if (info.u.param[2] == VI_LARGEINDEX) { /* large vnodes file */
2916 strlcpy(largeName, dp->d_name, sizeof(largeName));
2920 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);
2927 if (!infoSeen || !smallSeen || !largeSeen || !linkSeen) {
2928 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name);
2934 * If we come here then there was only a RO-volume and we can safely
2938 memset(&t_ih, 0, sizeof(t_ih));
2939 t_ih.ih_dev = ih->ih_dev;
2940 t_ih.ih_vid = ih->ih_vid;
2942 (void)afs_snprintf(oldpath, sizeof oldpath, "%s" OS_DIRSEP "%s", dir_name,
2944 fd = afs_open(oldpath, O_RDWR, 0);
2945 if (fd == INVALID_FD) {
2946 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
2951 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_VOLINFO);
2952 namei_HandleToName(&n, &t_ih);
2953 fd2 = afs_open(n.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
2954 if (fd2 == INVALID_FD) {
2955 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n.n_path);
2960 code = convertVolumeInfo(fd, fd2, ih->ih_vid);
2964 OS_UNLINK(n.n_path);
2968 SetOGM(fd2, ih->ih_vid, 1);
2971 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_SMALLINDEX);
2972 namei_HandleToName(&n, &t_ih);
2973 (void)afs_snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
2975 fd = afs_open(newpath, O_RDWR, 0);
2976 if (fd == INVALID_FD) {
2977 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath);
2981 SetOGM(fd, ih->ih_vid, 2);
2984 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
2986 link(newpath, n.n_path);
2990 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_LARGEINDEX);
2991 namei_HandleToName(&n, &t_ih);
2992 (void)afs_snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
2994 fd = afs_open(newpath, O_RDWR, 0);
2995 if (fd == INVALID_FD) {
2996 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath);
3000 SetOGM(fd, ih->ih_vid, 3);
3003 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
3005 link(newpath, n.n_path);
3012 h.volumeInfo_hi = h.id;
3013 h.smallVnodeIndex_hi = h.id;
3014 h.largeVnodeIndex_hi = h.id;
3015 h.linkTable_hi = h.id;
3017 if (VCreateVolumeDiskHeader(&h, partP)) {
3018 Log("1 namei_ConvertROtoRWvolume: Couldn't write header for RW-volume %lu\n",
3019 afs_printable_uint32_lu(h.id));
3024 if (VDestroyVolumeDiskHeader(partP, volumeId, h.parent)) {
3025 Log("1 namei_ConvertROtoRWvolume: Couldn't unlink header for RO-volume %lu\n",
3026 afs_printable_uint32_lu(volumeId));
3029 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_DONE, 0, NULL);
3030 FSYNC_VolOp(h.id, pname, FSYNC_VOL_ON, 0, NULL);
3033 # ifdef AFS_DEMAND_ATTACH_FS
3035 VUnlockVolumeById(volumeId, partP);
3037 # endif /* AFS_DEMAND_ATTACH_FS */
3045 * returns a static string used to print either 32 or 64 bit inode numbers.
3048 PrintInode(char *s, Inode ino)
3050 static afs_ino_str_t result;
3054 (void)afs_snprintf(s, sizeof(afs_ino_str_t), "%" AFS_UINT64_FMT, (afs_uintmax_t) ino);
3061 /* Routines to facilitate removing zero link count files. */
3062 #define MAX_ZLC_NAMES 32
3063 #define MAX_ZLC_NAMELEN 16
3064 typedef struct zlcList_s {
3065 struct zlcList_s *zlc_next;
3067 char zlc_names[MAX_ZLC_NAMES][MAX_ZLC_NAMELEN];
3070 static zlcList_t *zlcAnchor = NULL;
3071 static zlcList_t *zlcCur = NULL;
3074 AddToZLCDeleteList(char dir, char *name)
3076 osi_Assert(strlen(name) <= MAX_ZLC_NAMELEN - 3);
3078 if (!zlcCur || zlcCur->zlc_n >= MAX_ZLC_NAMES) {
3079 if (zlcCur && zlcCur->zlc_next)
3080 zlcCur = zlcCur->zlc_next;
3082 zlcList_t *tmp = (zlcList_t *) malloc(sizeof(zlcList_t));
3088 zlcCur->zlc_next = tmp;
3092 zlcCur->zlc_next = NULL;
3097 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%c" OS_DIRSEP "%s", dir, name);
3099 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%s", name);
3105 DeleteZLCFiles(char *path)
3111 for (z = zlcAnchor; z; z = z->zlc_next) {
3112 for (i = 0; i < z->zlc_n; i++) {
3114 (void)sprintf(fname, "%s" OS_DIRSEP "%s", path, z->zlc_names[i]);
3116 (void)sprintf(fname, "%s", z->zlc_names[i]);
3117 if (namei_unlink(fname) < 0) {
3118 Log("ZLC: Can't unlink %s, dos error = %d\n", fname,
3122 z->zlc_n = 0; /* Can reuse space. */
3135 tnext = i->zlc_next;
3139 zlcCur = zlcAnchor = NULL;
3143 #endif /* AFS_NAMEI_ENV */