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
67 #define afs_fopen fopen64
68 #else /* !O_LARGEFILE */
70 #endif /*S_SPLINT_S */
72 #define afs_fstat fstat
73 #define afs_fopen fopen
74 #endif /* !O_LARGEFILE */
75 /*@=fcnmacros =macrofcndecl@*/
78 #define LOCK_SH 1 /* shared lock */
79 #define LOCK_EX 2 /* exclusive lock */
80 #define LOCK_NB 4 /* don't block when locking */
81 #define LOCK_UN 8 /* unlock */
86 #include <afs/work_queue.h>
87 #include <afs/thread_pool.h>
88 #include <vol/vol-salvage.h>
91 #if !defined(HAVE_FLOCK) && !defined(AFS_NT40_ENV)
95 * This function emulates a subset of flock()
98 emul_flock(int fd, int cmd)
101 memset(&f, 0, sizeof (f));
110 return fcntl(fd, (cmd & LOCK_NB) ? F_SETLK : F_SETLKW, &f);
113 #define flock(f,c) emul_flock(f,c)
120 namei_iread(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
129 nBytes = FDH_PREAD(fdP, buf, size, offset);
135 namei_iwrite(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
144 nBytes = FDH_PWRITE(fdP, buf, size, offset);
150 /* Inode number format:
151 * low 32 bits - if a regular file or directory, the vnode; else the type.
152 * 32-36 - uniquifier tag and index into counts array for this vnode. Only
153 * two of the available bits are currently used. The rest are
154 * present in case we ever increase the number of types of volumes
155 * in the volume group.
156 * bit 37 : 1 == special, 0 == regular
158 # define NAMEI_VNODEMASK 0x00ffffffff
159 # define NAMEI_TAGSHIFT 32
160 # define NAMEI_INODESPECIAL 0x2000000000
161 # define NAMEI_SPECDIR "R"
162 # define NAMEI_SPECDIRC 'R'
163 #else /* !AFS_NT40_ENV */
164 /* Inode number format:
165 * low 26 bits - vnode number - all 1's if volume special file.
167 * next 3 bits spare (0's)
168 * high 32 bits - uniquifier (regular) or type if spare
170 # define NAMEI_VNODEMASK 0x003ffffff
171 # define NAMEI_TAGSHIFT 26
172 # define NAMEI_UNIQMASK 0xffffffff
173 # define NAMEI_UNIQSHIFT 32
174 # define NAMEI_INODESPECIAL ((Inode)NAMEI_VNODEMASK)
175 /* dir1 is the high 8 bits of the 26 bit vnode */
176 # define VNO_DIR1(vno) ((vno >> 14) & 0xff)
177 /* dir2 is the next 9 bits */
178 # define VNO_DIR2(vno) ((vno >> 9) & 0x1ff)
179 /* "name" is the low 9 bits of the vnode, the 3 bit tag and the uniq */
180 # define NAMEI_SPECDIR "special"
181 #endif /* !AFS_NT40_ENV */
182 #define NAMEI_TAGMASK 0x7
183 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
185 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
187 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
188 * group, not counting temporary (move) volumes.
189 * This is the number of separate files, all having
190 * the same vnode number, which can occur in a volume
202 static int GetFreeTag(IHandle_t * ih, int vno);
204 /* namei_HandleToInodeDir
206 * Construct the path name of the directory holding the inode data.
207 * Format: /<vicepx>/INODEDIR
212 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
214 memset(name, '\0', sizeof(*name));
215 nt_DevToDrive(name->n_drive, ih->ih_dev);
216 strlcpy(name->n_path, name->n_drive, sizeof(name->n_path));
220 /* Format: /<vicepx>/INODEDIR */
221 #define PNAME_BLEN 64
223 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
227 memset(name, '\0', sizeof(*name));
230 * Add the /vicepXX string to the start of name->n_base and then calculate
231 * offset as the number of bytes we know we added.
233 * FIXME: This embeds knowledge of the vice partition naming scheme and
234 * mapping from device numbers. There needs to be an API that tells us
237 volutil_PartitionName_r(ih->ih_dev, name->n_base, sizeof(name->n_base));
238 offset = VICE_PREFIX_SIZE + (ih->ih_dev > 25 ? 2 : 1);
239 name->n_base[offset] = OS_DIRSEPC;
241 strlcpy(name->n_base + offset, INODEDIR, sizeof(name->n_base) - offset);
242 strlcpy(name->n_path, name->n_base, sizeof(name->n_path));
246 #define addtoname(N, C) \
248 if ((N)->n_path[strlen((N)->n_path)-1] != OS_DIRSEPC) \
249 strlcat((N)->n_path, OS_DIRSEP, sizeof((N)->n_path)); \
250 strlcat((N)->n_path, (C), sizeof((N)->n_path)); \
256 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
258 /* X:\Vol_XXXXXXX.data */
262 namei_HandleToInodeDir(name, ih);
263 /* nt_drive added to name by namei_HandleToInodeDir() */
264 namep = name->n_voldir;
265 (void)memcpy(namep, "\\Vol_", 5);
267 (void)strcpy(namep, int_to_base32(str1, ih->ih_vid));
268 namep += strlen(namep);
269 memcpy(namep, ".data", 5);
272 addtoname(name, name->n_voldir);
276 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
280 namei_HandleToInodeDir(name, ih);
281 (void)int32_to_flipbase64(tmp, (int64_t) (ih->ih_vid & 0xff));
282 strlcpy(name->n_voldir1, tmp, sizeof(name->n_voldir1));
283 addtoname(name, name->n_voldir1);
284 (void)int32_to_flipbase64(tmp, (int64_t) ih->ih_vid);
285 strlcpy(name->n_voldir2, tmp, sizeof(name->n_voldir2));
286 addtoname(name, name->n_voldir2);
290 /* namei_HandleToName
292 * Constructs a file name for the fully qualified handle.
295 /* Note that special files end up in X:\Vol_XXXXXXX.data\R */
297 namei_HandleToName(namei_t * name, IHandle_t * ih)
299 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
300 int special = (ih->ih_ino & NAMEI_INODESPECIAL)?1:0;
301 int tag = (int)((ih->ih_ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
304 namei_HandleToVolDir(name, ih);
307 name->n_dir[0] = NAMEI_SPECDIRC;
310 name->n_dir[0] = 'Q';
312 name->n_dir[0] = ((vno & 0x1f) >> 1) + 'A';
315 name->n_dir[1] = '\0';
316 addtoname(name, name->n_dir);
317 /* X:\Vol_XXXXXXX.data\X\V_XXXXXXX.XXX */
318 namep = name->n_inode;
319 (void)memcpy(namep, "\\V_", 3);
321 (void)strcpy(namep, int_to_base32(str1, vno));
322 namep += strlen(namep);
324 (void)strcpy(namep, int_to_base32(str1, tag));
325 namep += strlen(namep);
326 addtoname(name, name->n_inode);
329 /* Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special */
331 namei_HandleToName(namei_t * name, IHandle_t * ih)
333 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
336 namei_HandleToVolDir(name, ih);
338 if (vno == NAMEI_VNODESPECIAL) {
339 strlcpy(name->n_dir1, NAMEI_SPECDIR, sizeof(name->n_dir1));
340 addtoname(name, name->n_dir1);
341 name->n_dir2[0] = '\0';
343 (void)int32_to_flipbase64(str, VNO_DIR1(vno));
344 strlcpy(name->n_dir1, str, sizeof(name->n_dir1));
345 addtoname(name, name->n_dir1);
346 (void)int32_to_flipbase64(str, VNO_DIR2(vno));
347 strlcpy(name->n_dir2, str, sizeof(name->n_dir2));
348 addtoname(name, name->n_dir2);
350 (void)int64_to_flipbase64(str, (int64_t) ih->ih_ino);
351 strlcpy(name->n_inode, str, sizeof(name->n_inode));
352 addtoname(name, name->n_inode);
357 /* The following is a warning to tell sys-admins to not muck about in this
360 #define VICE_README "These files and directories are a part of the AFS \
361 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
362 ownership and permissions included.\n"
364 namei_ViceREADME(char *partition)
369 /* Create the inode directory if we're starting for the first time */
370 (void)afs_snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s", partition,
372 mkdir(filename, 0700);
374 (void)afs_snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s" OS_DIRSEP "README",
375 partition, INODEDIR);
376 fd = OS_OPEN(filename, O_WRONLY | O_CREAT | O_TRUNC, 0444);
377 if (fd != INVALID_FD) {
378 (void)OS_WRITE(fd, VICE_README, strlen(VICE_README));
385 /* namei_CreateDataDirectories
387 * If creating the file failed because of ENOENT or ENOTDIR, try
388 * creating all the directories first.
392 namei_CreateDataDirectories(namei_t * name, int *created)
399 afs_snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
401 if (mkdir(tmp) < 0) {
412 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
414 if (mkdir(tmp) < 0 && errno != EEXIST)
420 #define create_dir() \
422 if (mkdir(tmp, 0700)<0) { \
423 if (errno != EEXIST) \
431 #define create_nextdir(A) \
433 strcat(tmp, OS_DIRSEP); strcat(tmp, A); create_dir(); \
437 namei_CreateDataDirectories(namei_t * name, int *created)
443 strlcpy(tmp, name->n_base, sizeof(tmp));
446 create_nextdir(name->n_voldir1);
447 create_nextdir(name->n_voldir2);
448 create_nextdir(name->n_dir1);
449 if (name->n_dir2[0]) {
450 create_nextdir(name->n_dir2);
457 /* delTree(): Deletes an entire tree of directories (no files)
459 * root : Full path to the subtree. Should be big enough for PATH_MAX
460 * tree : the subtree to be deleted is rooted here. Specifies only the
461 * subtree beginning at tree (not the entire path). It should be
462 * a pointer into the "root" buffer.
464 * errp : errno of the first error encountered during the directory cleanup.
465 * *errp should have been initialized to 0.
468 * -1 : If errors were encountered during cleanup and error is set to
472 * If there are errors, we try to work around them and delete as many
473 * directories as possible. We don't attempt to remove directories that still
474 * have non-dir entries in them.
477 delTree(char *root, char *tree, int *errp)
485 /* delete the children first */
486 cp = strchr(tree, OS_DIRSEPC);
488 delTree(root, cp + 1, errp);
491 cp = tree + strlen(tree); /* move cp to the end of string tree */
493 /* now delete all entries in this dir */
494 if ((ds = opendir(root)) != (DIR *) NULL) {
496 while ((dirp = readdir(ds))) {
497 /* ignore . and .. */
498 if (!strcmp(dirp->d_name, ".") || !strcmp(dirp->d_name, ".."))
500 /* since root is big enough, we reuse the space to
501 * concatenate the dirname to the current tree
503 strcat(root, OS_DIRSEP);
504 strcat(root, dirp->d_name);
505 if (afs_stat(root, &st) == 0 && S_ISDIR(st.st_mode)) {
506 /* delete this subtree */
507 delTree(root, cp + 1, errp);
509 *errp = *errp ? *errp : errno;
511 /* recover path to our cur tree by truncating it to
516 /* if (!errno) -- closedir not implicit if we got an error */
520 /* finally axe the current dir */
522 *errp = *errp ? *errp : errno;
524 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
526 #endif /* !AFS_PTHREAD_ENV */
531 /* if we encountered errors during cleanup, we return a -1 */
540 /* namei_RemoveDataDirectories
542 * Returns 0 on success.
543 * Returns -1 on error. Typically, callers ignore this error because we
544 * can continue running if the removes fail. The salvage process will
545 * finish tidying up for us.
550 namei_RemoveDataDirectories(namei_t * name)
557 afs_snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
560 path += strlen(path);
561 *path++ = OS_DIRSEPC;
563 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
565 if (rmdir(name->n_path) < 0 && errno != ENOENT)
570 /* Delete the Vol_NNNNNN.data directory. */
573 if (rmdir(name->n_path) < 0 && errno != ENOENT) {
581 * We only use the n_base and n_voldir1 entries
582 * and only do rmdir's.
585 namei_RemoveDataDirectories(namei_t * name)
589 int prefixlen = strlen(name->n_base), err = 0;
590 int vollen = strlen(name->n_voldir1);
591 char pbuf[MAXPATHLEN];
595 strlcpy(path, name->n_path, sizeof(pbuf));
597 /* move past the prefix and n_voldir1 */
598 path = path + prefixlen + 1 + vollen + 1; /* skip over the trailing / */
600 /* now delete all dirs upto path */
601 code = delTree(pbuf, path, &err);
603 /* We've now deleted everything under /n_base/n_voldir1/n_voldir2 that
604 * we could. Do not delete /n_base/n_voldir1, since doing such might
605 * interrupt another thread trying to create a volume. We could introduce
606 * some locking to make this safe (or only remove it for whole-partition
607 * salvages), but by not deleting it we only leave behind a maximum of
608 * 256 empty directories. So at least for now, don't bother. */
613 /* Create the file in the name space.
615 * Parameters stored as follows:
617 * p1 - volid - implied in containing directory.
618 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
619 * p3 - uniq -- bits 4-0 are in mode bits 4-0
620 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
622 * p1 - volid - creation time - dwHighDateTime
623 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
624 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
625 * p4 - parid - parent volume id - implied in containing directory.
627 * Return value is the inode number or (Inode)-1 if error.
628 * We "know" there is only one link table, so return EEXIST if there already
629 * is a link table. It's up to the calling code to test errno and increment
635 * This function is called by VCreateVolume to hide the implementation
636 * details of the inode numbers. This only allows for 7 volume special
637 * types, but if we get that far, this could should be dead by then.
640 namei_MakeSpecIno(int volid, int type)
643 ino = NAMEI_INODESPECIAL;
646 /* tag is always 0 for special */
648 type &= NAMEI_TAGMASK;
649 ino |= ((Inode) type) << NAMEI_TAGSHIFT;
650 ino |= ((Inode) volid) << NAMEI_UNIQSHIFT;
658 SetOGM(FD_t fd, int parm, int tag)
664 CheckOGM(namei_t *name, FdHandle_t *fdP, int p1)
666 WIN32_FIND_DATA info;
670 FindFirstFileEx(name->n_path, FindExInfoStandard, &info,
671 FindExSearchNameMatch, NULL,
672 FIND_FIRST_EX_CASE_SENSITIVE);
675 return -1; /* Can't get info, leave alone */
679 if (info.ftCreationTime.dwHighDateTime != (unsigned int)p1)
684 #else /* AFS_NT40_ENV */
685 /* SetOGM - set owner group and mode bits from parm and tag */
687 SetOGM(FD_t fd, int parm, int tag)
690 * owner - low 15 bits of parm.
691 * group - next 15 bits of parm.
692 * mode - 2 bits of parm, then lowest = 3 bits of tag.
694 int owner, group, mode;
696 owner = parm & 0x7fff;
697 group = (parm >> 15) & 0x7fff;
698 if (fchown(fd, owner, group) < 0)
701 mode = (parm >> 27) & 0x18;
703 if (fchmod(fd, mode) < 0)
708 /* GetOGM - get parm and tag from owner, group and mode bits. */
710 GetOGMFromStat(struct afs_stat *status, int *parm, int *tag)
712 *parm = status->st_uid | (status->st_gid << 15);
713 *parm |= (status->st_mode & 0x18) << 27;
714 *tag = status->st_mode & 0x7;
718 CheckOGM(namei_t *name, FdHandle_t *fdP, int p1)
720 struct afs_stat status;
722 if (afs_fstat(fdP->fd_fd, &status) < 0)
725 GetOGMFromStat(&status, &parm, &tag);
731 #endif /* !AFS_NT40_ENV */
733 int big_vno = 0; /* Just in case we ever do 64 bit vnodes. */
735 /* Derive the name and create it O_EXCL. If that fails we have an error.
736 * Get the tag from a free column in the link table.
740 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
743 FD_t fd = INVALID_FD;
754 memset((void *)&tmp, 0, sizeof(IHandle_t));
755 memset(&tfd, 0, sizeof(FdHandle_t));
757 tmp.ih_dev = nt_DriveToDev(part);
758 if (tmp.ih_dev == -1) {
763 if (p2 == INODESPECIAL) {
764 /* Parameters for special file:
765 * p1 - volume id - goes into owner/group/mode
766 * p2 - vnode == INODESPECIAL
768 * p4 - parent volume id
770 ftime.dwHighDateTime = p1;
771 ftime.dwLowDateTime = p2;
773 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
774 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
776 int vno = p2 & NAMEI_VNODEMASK;
777 /* Parameters for regular file:
791 tmp.ih_ino = (Inode) p2;
792 ftime.dwHighDateTime = p3;
793 ftime.dwLowDateTime = p4;
796 namei_HandleToName(&name, &tmp);
797 p = strrchr((char *)&name.n_path, '.');
799 for (tag = 0; tag < NAMEI_MAXVOLS; tag++) {
800 *p = *int_to_base32(str1, tag);
801 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_TRUNC | O_EXCL, 0666);
802 if (fd == INVALID_FD) {
803 if (errno == ENOTDIR || errno == ENOENT) {
804 if (namei_CreateDataDirectories(&name, &created_dir) == 0)
805 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_TRUNC | O_EXCL, 0666);
809 if (fd != INVALID_FD)
811 if (p2 == INODESPECIAL && p3 == VI_LINKTABLE)
814 if (fd == INVALID_FD) {
818 tmp.ih_ino &= ~(((Inode) NAMEI_TAGMASK) << NAMEI_TAGSHIFT);
819 tmp.ih_ino |= (((Inode) tag) << NAMEI_TAGSHIFT);
822 if (!SetFileTime((HANDLE) fd, &ftime, NULL, NULL)) {
823 errno = OS_ERROR(EBADF);
829 if (p2 != INODESPECIAL) {
830 if (fd == INVALID_FD) {
832 code = nt_unlink((char *)&name.n_path);
841 code = namei_SetLinkCount(fdP, tmp.ih_ino, 1, 0);
843 } else if (p2 == INODESPECIAL && p3 == VI_LINKTABLE) {
844 if (fd == INVALID_FD)
846 /* hack at tmp to setup for set link count call. */
848 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
853 if (fd != INVALID_FD)
856 if (code || (fd == INVALID_FD)) {
857 if (p2 != INODESPECIAL) {
860 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
866 int save_errno = errno;
867 namei_RemoveDataDirectories(&name);
871 return (code || (fd == INVALID_FD)) ? (Inode) -1 : tmp.ih_ino;
873 #else /* !AFS_NT40_ENV */
875 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
887 memset((void *)&tmp, 0, sizeof(IHandle_t));
888 memset(&tfd, 0, sizeof(FdHandle_t));
890 tmp.ih_dev = volutil_GetPartitionID(part);
891 if (tmp.ih_dev == -1) {
897 /* Parameters for special file:
898 * p1 - volume id - goes into owner/group/mode
901 * p4 - parent volume id
906 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
907 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
909 int vno = p2 & NAMEI_VNODEMASK;
910 /* Parameters for regular file:
922 /* If GetFreeTag succeeds, it atomically sets link count to 1. */
923 tag = GetFreeTag(lh, p2);
928 tmp.ih_ino = (Inode) p2;
929 /* name is <uniq(p3)><tag><vno(p2)> */
930 tmp.ih_ino |= ((Inode) tag) << NAMEI_TAGSHIFT;
931 tmp.ih_ino |= ((Inode) p3) << NAMEI_UNIQSHIFT;
936 namei_HandleToName(&name, &tmp);
937 fd = OS_OPEN(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
938 if (fd == INVALID_FD) {
939 if (errno == ENOTDIR || errno == ENOENT) {
940 if (namei_CreateDataDirectories(&name, &created_dir) < 0)
942 fd = OS_OPEN(name.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR,
944 if (fd == INVALID_FD)
950 if (SetOGM(fd, ogm_parm, tag) < 0) {
956 if (p2 == (afs_uint32)-1 && p3 == VI_LINKTABLE) {
957 /* hack at tmp to setup for set link count call. */
958 memset((void *)&tfd, 0, sizeof(FdHandle_t)); /* minimalistic still, but a little cleaner */
961 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
965 if (fd != INVALID_FD)
969 if (code || (fd == INVALID_FD)) {
973 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
978 return (code || (fd == INVALID_FD)) ? (Inode) - 1 : tmp.ih_ino;
984 namei_iopen(IHandle_t * h)
989 /* Convert handle to file name. */
990 namei_HandleToName(&name, h);
991 fd = OS_OPEN((char *)&name.n_path, O_RDWR, 0666);
995 /* Need to detect vol special file and just unlink. In those cases, the
996 * handle passed in _is_ for the inode. We only check p1 for the special
1000 namei_dec(IHandle_t * ih, Inode ino, int p1)
1007 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1009 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1011 /* Verify this is the right file. */
1012 IH_INIT(tmp, ih->ih_dev, ih->ih_vid, ino);
1014 namei_HandleToName(&name, tmp);
1019 errno = OS_ERROR(ENOENT);
1023 if (CheckOGM(&name, fdP, p1) < 0) {
1024 FDH_REALLYCLOSE(fdP);
1026 errno = OS_ERROR(EINVAL);
1030 /* If it's the link table itself, decrement the link count. */
1031 if (type == VI_LINKTABLE) {
1032 if ((count = namei_GetLinkCount(fdP, (Inode) 0, 1, 0, 1)) < 0) {
1033 FDH_REALLYCLOSE(fdP);
1039 if (namei_SetLinkCount(fdP, (Inode) 0, count < 0 ? 0 : count, 1) <
1041 FDH_REALLYCLOSE(fdP);
1047 FDH_REALLYCLOSE(fdP);
1053 if ((code = OS_UNLINK(name.n_path)) == 0) {
1054 if (type == VI_LINKTABLE) {
1055 /* Try to remove directory. If it fails, that's ok.
1056 * Salvage will clean up.
1058 (void)namei_RemoveDataDirectories(&name);
1061 FDH_REALLYCLOSE(fdP);
1064 /* Get a file descriptor handle for this Inode */
1070 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0) {
1071 FDH_REALLYCLOSE(fdP);
1077 if (namei_SetLinkCount(fdP, ino, count, 1) < 0) {
1078 FDH_REALLYCLOSE(fdP);
1083 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1084 Log("Warning: Lost ref on ihandle dev %d vid %d ino %" AFS_INT64_FMT "\n",
1085 th->ih_dev, th->ih_vid, (afs_int64)th->ih_ino);
1088 /* If we're less than 0, someone presumably unlinked;
1089 don't bother setting count to 0, but we need to drop a lock */
1090 if (namei_SetLinkCount(fdP, ino, 0, 1) < 0) {
1091 FDH_REALLYCLOSE(fdP);
1097 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1099 namei_HandleToName(&name, th);
1101 code = OS_UNLINK(name.n_path);
1110 namei_inc(IHandle_t * h, Inode ino, int p1)
1116 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1117 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1118 if (type != VI_LINKTABLE)
1123 /* Get a file descriptor handle for this Inode */
1129 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0)
1134 errno = OS_ERROR(EINVAL);
1138 if (namei_SetLinkCount(fdP, ino, count, 1) < 0)
1142 FDH_REALLYCLOSE(fdP);
1149 #ifndef AFS_NT40_ENV
1151 namei_replace_file_by_hardlink(IHandle_t *hLink, IHandle_t *hTarget)
1157 /* Convert handle to file name. */
1158 namei_HandleToName(&nameLink, hLink);
1159 namei_HandleToName(&nameTarget, hTarget);
1161 OS_UNLINK(nameLink.n_path);
1162 code = link(nameTarget.n_path, nameLink.n_path);
1167 namei_copy_on_write(IHandle_t *h)
1173 struct afs_stat tstat;
1176 namei_HandleToName(&name, h);
1177 if (afs_stat(name.n_path, &tstat) < 0)
1179 if (tstat.st_nlink > 1) { /* do a copy on write */
1188 afs_snprintf(path, sizeof(path), "%s-tmp", name.n_path);
1189 fd = OS_OPEN(path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
1190 if (fd == INVALID_FD) {
1201 size = tstat.st_size;
1204 tlen = size > 8192 ? 8192 : size;
1205 if (FDH_PREAD(fdP, buf, tlen, offset) != tlen)
1207 if (OS_WRITE(fd, buf, tlen) != tlen)
1213 FDH_REALLYCLOSE(fdP);
1218 OS_UNLINK(name.n_path);
1219 code = rename(path, name.n_path);
1226 /************************************************************************
1227 * File Name Structure
1228 ************************************************************************
1230 * Each AFS file needs a unique name and it needs to be findable with
1231 * minimal lookup time. Note that the constraint on the number of files and
1232 * directories in a volume is the size of the vnode index files and the
1233 * max file size AFS supports (for internal files) of 2^31. Since a record
1234 * in the small vnode index file is 64 bytes long, we can have at most
1235 * (2^31)/64 or 33554432 files. A record in the large index file is
1236 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
1237 * Another layout parameter is that there is roughly a 16 to 1 ratio between
1238 * the number of files and the number of directories.
1240 * Using this information we can see that a layout of 256 directories, each
1241 * with 512 subdirectories and each of those having 512 files gives us
1242 * 256*512*512 = 67108864 AFS files and directories.
1244 * The volume, vnode, uniquifier and data version, as well as the tag
1245 * are required, either for finding the file or for salvaging. It's best to
1246 * restrict the name to something that can be mapped into 64 bits so the
1247 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
1248 * is used to distinguish between different versions of the same file
1249 * which are currently in the RW and clones of a volume. See "Link Table
1250 * Organization" below for more information on the tag. The tag is
1251 * required in the name of the file to ensure a unique name.
1253 * ifdef AFS_NT40_ENV
1254 * The data for each volume group is in a separate directory. The name of the
1255 * volume is of the form: Vol_NNNNNN.data, where NNNNNN is a base 32
1256 * representation of the RW volume ID (even where the RO is the only volume
1257 * on the partition). Below that are separate subdirectories for the
1258 * AFS directories and special files. There are also 16 directories for files,
1259 * hashed on the low 5 bits (recall bit0 is always 0) of the vnode number.
1260 * These directories are named:
1261 * A - P - 16 file directories.
1262 * Q ----- data directory
1263 * R ----- special files directory
1265 * The vnode is hashed into the directory using the low bits of the
1268 * The format of a file name for a regular file is:
1269 * Y:\Vol_NNNNNN.data\X\V_IIIIII.J
1270 * Y - partition encoded as drive letter, starting with D
1271 * NNNNNN - base 32 encoded volume number of RW volume
1272 * X - hash directory, as above
1273 * IIIIII - base 32 encoded vnode number
1274 * J - base 32 encoded tag
1276 * uniq is stored in the dwHighDateTime creation time field
1277 * dv is stored in the dwLowDateTime creation time field
1279 * Special inodes are always in the R directory, as above, and are
1281 * True child volid is stored in the dwHighDateTime creation time field
1282 * vnode number is always -1 (Special)
1283 * type is the IIIIII part of the filename
1284 * uniq is the J part of the filename
1285 * parent volume id is implied in the containing directory
1288 * We can store data in the uid, gid and mode bits of the files, provided
1289 * the directories have root only access. This gives us 15 bits for each
1290 * of uid and gid (GNU chown considers 65535 to mean "don't change").
1291 * There are 9 available mode bits. Adn we need to store a total of
1292 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
1293 * or 131 bits somewhere.
1295 * The format of a file name for a regular file is:
1296 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
1297 * V1 - low 8 bits of RW volume id
1298 * V2 - all bits of RW volume id
1299 * AA - high 8 bits of vnode number.
1300 * BB - next 9 bits of vnode number.
1301 * <tag><uniq><vno> - file name
1303 * Volume special files are stored in a separate directory:
1304 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
1307 * The vnode is hashed into the directory using the high bits of the
1308 * vnode number. This is so that consecutively created vnodes are in
1309 * roughly the same area on the disk. This will at least be optimal if
1310 * the user is creating many files in the same AFS directory. The name
1311 * should be formed so that the leading characters are different as quickly
1312 * as possible, leading to faster discards of incorrect matches in the
1320 /************************************************************************
1321 * Link Table Organization
1322 ************************************************************************
1324 * The link table volume special file is used to hold the link counts that
1325 * are held in the inodes in inode based AFS vice filesystems. For user
1326 * space access, the link counts are being kept in a separate
1327 * volume special file. The file begins with the usual version stamp
1328 * information and is then followed by one row per vnode number. vnode 0
1329 * is used to hold the link count of the link table itself. That is because
1330 * the same link table is shared among all the volumes of the volume group
1331 * and is deleted only when the last volume of a volume group is deleted.
1333 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
1334 * link count from 0 through 7. Each colume represents a unique instance of
1335 * that vnode. Say we have a file shared between the RW and a RO and a
1336 * different version of the file (or a different uniquifer) for the BU volume.
1337 * Then one column would be holding the link count of 2 for the RW and RO
1338 * and a different column would hold the link count of 1 for the BU volume.
1339 * # ifdef AFS_NT40_ENV
1340 * The column used is determined for NT by the uniquifier tag applied to
1341 * generate a unique file name in the NTFS namespace. The file name is
1342 * of the form "V_<vno>.<tag>" . And the <tag> is also the column number
1343 * in the link table.
1345 * Note that we allow only 5 volumes per file, giving 15 bits used in the
1349 #define LINKTABLE_WIDTH 2
1350 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
1353 * compute namei link table file and bit offset from inode number.
1355 * @param[in] ino inode number
1356 * @param[out] offset link table file offset
1357 * @param[out] index bit offset within 2-byte record
1362 namei_GetLCOffsetAndIndexFromIno(Inode ino, afs_foff_t * offset, int *index)
1364 int toff = (int)(ino & NAMEI_VNODEMASK);
1365 int tindex = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1367 *offset = (afs_foff_t) ((toff << LINKTABLE_SHIFT) + 8); /* * 2 + sizeof stamp */
1368 *index = (tindex << 1) + tindex;
1371 #ifdef AFS_PTHREAD_ENV
1372 /* XXX do static initializers work for WINNT/pthread? */
1373 pthread_mutex_t _namei_glc_lock = PTHREAD_MUTEX_INITIALIZER;
1374 #define NAMEI_GLC_LOCK MUTEX_ENTER(&_namei_glc_lock)
1375 #define NAMEI_GLC_UNLOCK MUTEX_EXIT(&_namei_glc_lock)
1376 #else /* !AFS_PTHREAD_ENV */
1377 #define NAMEI_GLC_LOCK
1378 #define NAMEI_GLC_UNLOCK
1379 #endif /* !AFS_PTHREAD_ENV */
1382 * get the link count of an inode.
1384 * @param[in] h namei link count table file handle
1385 * @param[in] ino inode number for which we are requesting a link count
1386 * @param[in] lockit if asserted, return with lock held on link table file
1387 * @param[in] fixup if asserted, write 1 to link count when read() returns
1389 * @param[in] nowrite return success on zero byte read or ZLC
1391 * @post if lockit asserted and lookup was successful, will return with write
1392 * lock on link table file descriptor
1394 * @return link count
1395 * @retval -1 namei link table i/o error
1400 namei_GetLinkCount(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite)
1402 unsigned short row = 0;
1407 /* there's no linktable yet. the salvager will create one later */
1408 if (h->fd_fd == INVALID_FD && fixup)
1410 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1413 if (FDH_LOCKFILE(h, offset) != 0)
1417 rc = FDH_PREAD(h, (char*)&row, sizeof(row), offset);
1418 if ((rc == 0 || !((row >> index) & NAMEI_TAGMASK)) && fixup && nowrite)
1420 if (rc == 0 && fixup) {
1422 * extend link table and write a link count of 1 for ino
1424 * in order to make MT-safe, truncation (extension really)
1425 * must happen under a mutex
1428 if (FDH_SIZE(h) >= offset+sizeof(row)) {
1430 goto bad_getLinkByte;
1432 FDH_TRUNC(h, offset+sizeof(row));
1434 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1437 if (rc != sizeof(row)) {
1438 goto bad_getLinkByte;
1441 if (fixup && !((row >> index) & NAMEI_TAGMASK)) {
1445 * in order to make this mt-safe, we need to do the read-modify-write
1446 * under a mutex. thus, we repeat the read inside the lock.
1449 rc = FDH_PREAD(h, (char *)&row, sizeof(row), offset);
1450 if (rc == sizeof(row)) {
1452 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1455 if (rc != sizeof(row))
1456 goto bad_getLinkByte;
1459 return (int)((row >> index) & NAMEI_TAGMASK);
1463 FDH_UNLOCKFILE(h, offset);
1468 namei_SetNonZLC(FdHandle_t * h, Inode ino)
1470 return namei_GetLinkCount(h, ino, 0, 1, 0);
1473 /* Return a free column index for this vnode. */
1475 GetFreeTag(IHandle_t * ih, int vno)
1489 /* Only one manipulates at a time. */
1490 if (FDH_LOCKFILE(fdP, offset) != 0) {
1491 FDH_REALLYCLOSE(fdP);
1495 offset = (vno << LINKTABLE_SHIFT) + 8; /* * 2 + sizeof stamp */
1497 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1498 if (nBytes != sizeof(row)) {
1504 /* Now find a free column in this row and claim it. */
1505 for (col = 0; col < NAMEI_MAXVOLS; col++) {
1506 coldata = 7 << (col * 3);
1507 if ((row & coldata) == 0)
1510 if (col >= NAMEI_MAXVOLS) {
1515 coldata = 1 << (col * 3);
1518 if (FDH_PWRITE(fdP, (char *)&row, sizeof(row), offset) != sizeof(row)) {
1522 FDH_UNLOCKFILE(fdP, offset);
1523 FDH_REALLYCLOSE(fdP);
1527 FDH_UNLOCKFILE(fdP, offset);
1528 FDH_REALLYCLOSE(fdP);
1534 /* namei_SetLinkCount
1535 * If locked is set, assume file is locked. Otherwise, lock file before
1536 * proceeding to modify it.
1539 namei_SetLinkCount(FdHandle_t * fdP, Inode ino, int count, int locked)
1545 ssize_t nBytes = -1;
1547 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1550 if (FDH_LOCKFILE(fdP, offset) != 0) {
1555 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1556 if (nBytes != sizeof(row)) {
1558 errno = OS_ERROR(EBADF);
1559 goto bad_SetLinkCount;
1564 bytesRead = 7 << index;
1566 row &= (unsigned short)~bytesRead;
1567 row |= (unsigned short)count;
1569 if (FDH_PWRITE(fdP, (char *)&row, sizeof(short), offset) != sizeof(short)) {
1570 errno = OS_ERROR(EBADF);
1571 goto bad_SetLinkCount;
1579 FDH_UNLOCKFILE(fdP, offset);
1581 /* disallowed above 7, so... */
1586 /* ListViceInodes - write inode data to a results file. */
1587 static int DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
1588 unsigned int volid);
1589 static int DecodeVolumeName(char *name, unsigned int *vid);
1590 static int namei_ListAFSSubDirs(IHandle_t * dirIH,
1591 int (*write_fun) (FILE *,
1592 struct ViceInodeInfo *,
1593 char *, char *), FILE * fp,
1594 int (*judgeFun) (struct ViceInodeInfo *,
1595 afs_uint32 vid, void *),
1596 afs_uint32 singleVolumeNumber, void *rock);
1601 * Write the inode data to the results file.
1603 * Returns -2 on error, 0 on success.
1605 * This is written as a callback simply so that other listing routines
1606 * can use the same inode reading code.
1609 WriteInodeInfo(FILE * fp, struct ViceInodeInfo *info, char *dir, char *name)
1612 n = fwrite(info, sizeof(*info), 1, fp);
1613 return (n == 1) ? 0 : -2;
1617 int mode_errors; /* Number of errors found in mode bits on directories. */
1619 VerifyDirPerms(char *path)
1621 struct afs_stat status;
1623 if (afs_stat(path, &status) < 0) {
1624 Log("Unable to stat %s. Please manually verify mode bits for this"
1625 " directory\n", path);
1627 if (((status.st_mode & 0777) != 0700) || (status.st_uid != 0))
1633 * Fill the results file with the requested inode information.
1635 * This code optimizes single volume salvages by just looking at that one
1636 * volume's directory.
1638 * @param[in] devname device name string
1639 * @param[in] moutnedOn vice partition mount point
1640 * @param[in] resultFile result file in which to write inode
1641 * metadata. If NULL, write routine is not
1643 * @param[in] judgeInode filter function pointer. if not NULL, only
1644 * inodes for which this routine returns non-
1645 * zero will be written to the results file.
1646 * @param[in] singleVolumeNumber volume id filter
1647 * @param[out] forcep always set to 0 for namei impl
1648 * @param[in] forceR not used by namei impl
1649 * @param[in] wpath not used by namei impl
1650 * @param[in] rock opaque pointer passed to judgeInode
1652 * @return operation status
1654 * @retval -1 complete failure, salvage should terminate.
1655 * @retval -2 not enough space on partition, salvager has error message
1659 ListViceInodes(char *devname, char *mountedOn, FILE *inodeFile,
1660 int (*judgeInode) (struct ViceInodeInfo * info, afs_uint32 vid, void *rock),
1661 afs_uint32 singleVolumeNumber, int *forcep, int forceR, char *wpath,
1665 struct afs_stat status;
1667 *forcep = 0; /* no need to salvage until further notice */
1669 /* Verify protections on directories. */
1671 VerifyDirPerms(mountedOn);
1674 namei_ListAFSFiles(mountedOn, WriteInodeInfo, inodeFile, judgeInode,
1675 singleVolumeNumber, rock);
1684 if (fflush(inodeFile) == EOF) {
1685 Log("Unable to successfully flush inode file for %s\n", mountedOn);
1688 if (fsync(fileno(inodeFile)) == -1) {
1689 Log("Unable to successfully fsync inode file for %s\n", mountedOn);
1694 * Paranoia: check that the file is really the right size
1696 if (afs_fstat(fileno(inodeFile), &status) == -1) {
1697 Log("Unable to successfully stat inode file for %s\n", mountedOn);
1700 if (status.st_size != ninodes * sizeof(struct ViceInodeInfo)) {
1701 Log("Wrong size (%d instead of %lu) in inode file for %s\n",
1702 (int) status.st_size,
1703 (long unsigned int) ninodes * sizeof(struct ViceInodeInfo),
1712 * Collect all the matching AFS files on the drive.
1713 * If singleVolumeNumber is non-zero, just return files for that volume.
1715 * @param[in] dev vice partition path
1716 * @param[in] writeFun function pointer to a function which
1717 * writes inode information to FILE fp
1718 * @param[in] fp file stream where inode metadata is sent
1719 * @param[in] judgeFun filter function pointer. if not NULL,
1720 * only entries for which a non-zero value
1721 * is returned are written to fp
1722 * @param[in] singleVolumeNumber volume id filter. if nonzero, only
1723 * process files for that specific volume id
1724 * @param[in] rock opaque pointer passed into writeFun and
1727 * @return operation status
1729 * @retval >=0 number of matching files found
1732 namei_ListAFSFiles(char *dev,
1733 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1736 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1737 afs_uint32 singleVolumeNumber, void *rock)
1744 #ifndef AFS_NT40_ENV
1750 static void FreeZLCList(void);
1753 memset((void *)&ih, 0, sizeof(IHandle_t));
1755 ih.ih_dev = nt_DriveToDev(dev);
1757 ih.ih_dev = volutil_GetPartitionID(dev);
1760 if (singleVolumeNumber) {
1761 ih.ih_vid = singleVolumeNumber;
1762 namei_HandleToVolDir(&name, &ih);
1764 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1765 singleVolumeNumber, rock);
1769 /* Find all volume data directories and descend through them. */
1770 namei_HandleToInodeDir(&name, &ih);
1772 dirp1 = opendir(name.n_path);
1775 while ((dp1 = readdir(dirp1))) {
1777 /* Heirarchy is one level on Windows */
1778 if (!DecodeVolumeName(dp1->d_name, &ih.ih_vid)) {
1780 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1784 if (*dp1->d_name == '.')
1786 afs_snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", name.n_path,
1788 dirp2 = opendir(path2);
1790 while ((dp2 = readdir(dirp2))) {
1791 if (*dp2->d_name == '.')
1793 if (!DecodeVolumeName(dp2->d_name, &ih.ih_vid)) {
1795 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1812 static void AddToZLCDeleteList(char dir, char *name);
1813 static void DeleteZLCFiles(char *path);
1817 * examine a namei volume special file.
1819 * @param[in] path1 volume special directory path
1820 * @param[in] dname directory entry name
1821 * @param[in] myIH inode handle to volume directory
1822 * @param[out] linkHandle namei link count fd handle. if
1823 * the inode in question is the link
1824 * table, then the FdHandle is populated
1825 * @param[in] writeFun metadata write function pointer
1826 * @param[in] fp file pointer where inode metadata
1827 * is written by (*writeFun)()
1828 * @param[in] judgeFun inode filter function pointer. if
1829 * not NULL, only inodes for which this
1830 * function returns non-zero are recorded
1831 * into fp by writeFun
1832 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1833 * inodes associated with this volume id
1834 * are recorded by writeFun
1835 * @param[in] rock opaque pointer passed to writeFun and
1838 * @return operation status
1839 * @retval 1 count this inode
1840 * @retval 0 don't count this inode
1841 * @retval -1 failure
1846 _namei_examine_special(char * path1,
1849 FdHandle_t * linkHandle,
1850 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1853 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1854 int singleVolumeNumber,
1858 struct ViceInodeInfo info;
1859 afs_uint32 inode_vgid;
1861 if (DecodeInode(path1, dname, &info, myIH->ih_vid) < 0) {
1867 inode_vgid = myIH->ih_vid;
1869 inode_vgid = (info.inodeNumber >> NAMEI_UNIQSHIFT) & NAMEI_UNIQMASK;
1872 if (info.u.param[2] != VI_LINKTABLE) {
1874 } else if ((info.u.param[0] != myIH->ih_vid) ||
1875 (inode_vgid != myIH->ih_vid)) {
1876 /* VGID encoded in linktable filename and/or OGM data isn't
1877 * consistent with VGID encoded in namei path */
1878 Log("namei_ListAFSSubDirs: warning: inconsistent linktable "
1879 "filename \"%s" OS_DIRSEP "%s\"; salvager will delete it "
1880 "(dir_vgid=%u, inode_vgid=%u, ogm_vgid=%u)\n",
1881 path1, dname, myIH->ih_vid,
1882 (unsigned int)inode_vgid,
1886 /* Open this handle */
1887 (void)afs_snprintf(path2, sizeof(path2),
1888 "%s" OS_DIRSEP "%s", path1, dname);
1889 linkHandle->fd_fd = OS_OPEN(path2, Testing ? O_RDONLY : O_RDWR, 0666);
1891 namei_GetLinkCount(linkHandle, (Inode) 0, 1, 1, Testing);
1895 (*judgeFun) (&info, singleVolumeNumber, rock)) {
1897 if ((*writeFun) (fp, &info, path1, dname) < 0) {
1907 * examine a namei file.
1909 * @param[in] path3 volume special directory path
1910 * @param[in] dname directory entry name
1911 * @param[in] myIH inode handle to volume directory
1912 * @param[in] linkHandle namei link count fd handle.
1913 * @param[in] writeFun metadata write function pointer
1914 * @param[in] fp file pointer where inode metadata
1915 * is written by (*writeFun)()
1916 * @param[in] judgeFun inode filter function pointer. if
1917 * not NULL, only inodes for which this
1918 * function returns non-zero are recorded
1919 * into fp by writeFun
1920 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1921 * inodes associated with this volume id
1922 * are recorded by writeFun
1923 * @param[in] rock opaque pointer passed to writeFun and
1926 * @return operation status
1927 * @retval 1 count this inode
1928 * @retval 0 don't count this inode
1929 * @retval -1 failure
1930 * @retval -2 request ZLC delete
1935 _namei_examine_reg(char * path3,
1938 FdHandle_t * linkHandle,
1939 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
1942 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
1943 int singleVolumeNumber,
1947 struct ViceInodeInfo info;
1949 int dirl; /* Windows-only (one level hash dir) */
1952 if (DecodeInode(path3, dname, &info, myIH->ih_vid) < 0) {
1957 namei_GetLinkCount(linkHandle,
1958 info.inodeNumber, 1, 1, Testing);
1959 if (info.linkCount == 0) {
1961 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n", path3, dname);
1962 #ifdef AFS_SALSRV_ENV
1963 /* defer -- the AddToZLCDeleteList() interface is not MT-safe */
1965 #else /* !AFS_SALSRV_ENV */
1966 dirl = path3[strlen(path3)-1];
1967 AddToZLCDeleteList((char)dirl, dname);
1968 #endif /* !AFS_SALSRV_ENV */
1969 #else /* !DELETE_ZLC */
1970 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path3,
1977 (*judgeFun) (&info, singleVolumeNumber, rock)) {
1979 if ((*writeFun) (fp, &info, path3, dname) < 0) {
1989 * listsubdirs work queue node.
1991 struct listsubdirs_work_node {
1992 #ifdef AFS_SALSRV_ENV
1993 int *error; /**< *error set if an error was
1994 * encountered in any listsubdirs
1998 IHandle_t * IH; /**< volume directory handle */
1999 FdHandle_t *linkHandle; /**< namei link count fd handle. when
2000 * examinining the link table special
2001 * inode, this will be pointed at the
2004 FILE * fp; /**< file pointer for writeFun */
2006 /** function which will write inode metadata to fp */
2007 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *, char *);
2009 /** inode filter function */
2010 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *);
2011 int singleVolumeNumber; /**< volume id filter */
2012 void * rock; /**< pointer passed to writeFun and judgeFun */
2013 int code; /**< return code from examine function */
2014 int special; /**< asserted when this is a volume
2019 * simple wrapper around _namei_examine_special and _namei_examine_reg.
2021 * @param[in] work the struct listsubdirs_work_node for the associated
2022 * "list subdirs" job
2023 * @param[in] dir the directory to examine
2024 * @param[in] filename the filename in 'dir' to examine
2026 * @return operation status
2027 * @retval 1 count this inode
2028 * @retval 0 don't count this inode
2029 * @retval -1 failure
2032 _namei_examine_file(const struct listsubdirs_work_node *work, char *dir,
2035 if (work->special) {
2036 return _namei_examine_special(dir, filename, work->IH,
2037 work->linkHandle, work->writeFun, work->fp,
2038 work->judgeFun, work->singleVolumeNumber,
2041 return _namei_examine_reg(dir, filename, work->IH,
2042 work->linkHandle, work->writeFun, work->fp,
2043 work->judgeFun, work->singleVolumeNumber,
2049 #ifdef AFS_SALSRV_ENV
2050 /** @addtogroup afs_vol_salsrv_pario */
2054 * arguments for the _namei_examine_file_cbk callback function.
2056 struct listsubdirs_args {
2057 const struct listsubdirs_work_node *work; /**< arguments that are the same
2058 * for all invocations of
2059 * _namei_examine_file_cbk, in
2061 int *result; /**< where we can store the return code of _namei_examine_file */
2063 char dir[512]; /**< directory to examine */
2064 char filename[256]; /**< filename in 'dir' to examine */
2068 * a node in the list of results of listsubdir jobs.
2070 struct listsubdirs_result {
2072 int inodes; /**< return value from _namei_examine_file */
2076 * clean up a list of 'struct listsubdirs_result's and interpret the results.
2078 * @param[in] resultlist a list of 'struct listsubdirs_result's
2080 * @return number of inodes found
2084 _namei_listsubdirs_cleanup_results(struct rx_queue *resultlist)
2086 struct listsubdirs_result *res, *nres;
2089 for(queue_Scan(resultlist, res, nres, listsubdirs_result)) {
2091 /* noop, retain erroneous error code */
2092 } else if (res->inodes < 0) {
2107 * wait for the spawned listsubdirs jobs to finish, and return how many inodes
2110 * @param[in] queue queue to wait to finish
2111 * @param[in] resultlist list of 'struct listsubdirs_result's that the queued
2112 * jobs are storing their results in
2114 * @return number of inodes found
2118 _namei_listsubdirs_wait(struct afs_work_queue *queue, struct rx_queue *resultlist)
2122 code = afs_wq_wait_all(queue);
2127 return _namei_listsubdirs_cleanup_results(resultlist);
2131 * work queue entry point for examining namei files.
2133 * @param[in] queue pointer to struct Vwork_queue
2134 * @param[in] node pointer to struct Vwork_queue_node
2135 * @param[in] queue_rock opaque pointer to struct salsrv_pool_state
2136 * @param[in] node_rock opaque pointer to struct listsubdirs_work_node
2137 * @param[in] caller_rock opaque pointer to struct salsrv_worker_thread_state
2139 * @return operation status
2141 * @see Vwork_queue_callback_func_t
2146 _namei_examine_file_cbk(struct afs_work_queue *queue,
2147 struct afs_work_queue_node *node,
2153 struct listsubdirs_args *args = node_rock;
2154 const struct listsubdirs_work_node * work = args->work;
2155 char *dir = args->dir;
2156 char *filename = args->filename;
2158 code = _namei_examine_file(work, dir, filename);
2160 *(args->result) = code;
2164 /* we've errored, so no point in letting more jobs continue */
2165 afs_wq_shutdown(queue);
2171 static pthread_once_t wq_once = PTHREAD_ONCE_INIT;
2172 static pthread_key_t wq_key;
2175 * create the wq_key key for storing a work queue.
2178 _namei_wq_keycreate(void)
2180 osi_Assert(pthread_key_create(&wq_key, NULL) == 0);
2184 * set the work queue for this thread to use for backgrounding namei ops.
2186 * The work queue will be used in ListAFSSubdirs (called indirectly by
2187 * ListViceInodes) to examine files in parallel.
2189 * @param[in] wq the work queue to use
2192 namei_SetWorkQueue(struct afs_work_queue *wq)
2194 osi_Assert(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2196 osi_Assert(pthread_setspecific(wq_key, wq) == 0);
2200 * enqueue an examine file work unit.
2202 * @param[in] work the _namei_examine_file arguments that are common to
2203 * all callers within the same ListAFSFiles operation
2204 * @param[in] dir the specific directory to look at (string will be
2205 * copied; can be stack/temporary memory)
2206 * @param[in] filename the filename to look at (string will be copied; can be
2207 * stack/temporary memory)
2208 * @param[in] wq work queue to enqueue this work unit to
2209 * @param[in] resultlist the list to append the 'struct listsubdirs_result' to
2210 * for the enqueued work unit
2212 * @return operation status
2214 * @retval -1 fatal error
2216 * @note errors MUST be indicated by a -1 error code and nothing else, to be
2217 * compatible with _namei_examine_reg and _namei_examine_special
2222 _namei_examine_file_spawn(const struct listsubdirs_work_node *work,
2223 const char *dir, const char *filename,
2224 struct afs_work_queue *wq,
2225 struct rx_queue *resultlist)
2228 struct listsubdirs_args *args = NULL;
2229 struct listsubdirs_result *result = NULL;
2230 struct afs_work_queue_node *node = NULL;
2231 struct afs_work_queue_add_opts opts;
2233 args = malloc(sizeof(*args));
2239 result = malloc(sizeof(*result));
2240 if (result == NULL) {
2245 code = afs_wq_node_alloc(&node);
2250 code = afs_wq_node_set_detached(node);
2257 args->result = &result->inodes;
2258 strlcpy(args->dir, dir, sizeof(args->dir));
2259 strlcpy(args->filename, filename, sizeof(args->filename));
2261 code = afs_wq_node_set_callback(node,
2262 &_namei_examine_file_cbk,
2270 afs_wq_add_opts_init(&opts);
2273 code = afs_wq_add(wq, node, &opts);
2280 queue_Append(resultlist, result);
2285 afs_wq_node_put(node);
2301 #else /* !AFS_SALSRV_ENV */
2302 # define _namei_examine_file_spawn(work, dir, file, wq, resultlist) \
2303 _namei_examine_file(work, dir, file)
2304 #endif /* !AFS_SALSRV_ENV */
2307 * traverse and check inodes.
2309 * @param[in] dirIH volume group directory handle
2310 * @param[in] writeFun function pointer which will write inode
2311 * metadata to FILE stream fp
2312 * @param[in] fp file stream where inode metadata gets
2314 * @param[in] judgeFun inode filter function. if not NULL, only
2315 * inodes for which the filter returns non-zero
2316 * will be written out by writeFun
2317 * @param[in] singleVolumeNumber volume id filter. only inodes matching this
2318 * filter are written out by writeFun
2319 * @param[in] rock opaque pointer passed to judgeFun and writeFun
2321 * @return operation status
2323 * @retval >=0 number of matching inodes found
2325 * @todo the salsrv implementation may consume a lot of
2326 * memory for a large volume. at some point we should
2327 * probably write a background thread to asynchronously
2328 * clean up the resultlist nodes to reduce memory footprint
2333 namei_ListAFSSubDirs(IHandle_t * dirIH,
2334 int (*writeFun) (FILE *, struct ViceInodeInfo *, char *,
2337 int (*judgeFun) (struct ViceInodeInfo *, afs_uint32, void *),
2338 afs_uint32 singleVolumeNumber, void *rock)
2340 int code = 0, ret = 0;
2341 IHandle_t myIH = *dirIH;
2343 char path1[512], path3[512];
2345 #ifndef AFS_NT40_ENV
2350 struct dirent *dp1, *dp3;
2351 FdHandle_t linkHandle;
2353 struct listsubdirs_work_node work;
2354 #ifdef AFS_SALSRV_ENV
2356 struct afs_work_queue *wq;
2358 struct rx_queue resultlist;
2361 namei_HandleToVolDir(&name, &myIH);
2362 strlcpy(path1, name.n_path, sizeof(path1));
2364 /* Do the directory containing the special files first to pick up link
2367 (void)strcat(path1, OS_DIRSEP);
2368 (void)strcat(path1, NAMEI_SPECDIR);
2370 linkHandle.fd_fd = INVALID_FD;
2371 #ifdef AFS_SALSRV_ENV
2372 osi_Assert(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2374 wq = pthread_getspecific(wq_key);
2380 queue_Init(&resultlist);
2383 memset(&work, 0, sizeof(work));
2384 work.linkHandle = &linkHandle;
2387 work.writeFun = writeFun;
2388 work.judgeFun = judgeFun;
2389 work.singleVolumeNumber = singleVolumeNumber;
2392 #ifdef AFS_SALSRV_ENV
2393 work.error = &error;
2396 dirp1 = opendir(path1);
2398 while ((dp1 = readdir(dirp1))) {
2399 if (*dp1->d_name == '.')
2402 #ifdef AFS_SALSRV_ENV
2408 #endif /* AFS_SALSRV_ENV */
2410 code = _namei_examine_file_spawn(&work, path1, dp1->d_name, wq, &resultlist);
2420 /* count this inode */
2421 #ifndef AFS_SALSRV_ENV
2430 #ifdef AFS_SALSRV_ENV
2431 /* effectively a barrier */
2432 code = _namei_listsubdirs_wait(wq, &resultlist);
2433 if (code < 0 || error) {
2441 if (linkHandle.fd_fd == INVALID_FD) {
2442 Log("namei_ListAFSSubDirs: warning: VG %u does not have a link table; "
2443 "salvager will recreate it.\n", dirIH->ih_vid);
2446 /* Now run through all the other subdirs */
2447 namei_HandleToVolDir(&name, &myIH);
2448 strlcpy(path1, name.n_path, sizeof(path1));
2452 dirp1 = opendir(path1);
2454 while ((dp1 = readdir(dirp1))) {
2455 #ifndef AFS_NT40_ENV
2456 if (*dp1->d_name == '.')
2459 if (!strcmp(dp1->d_name, NAMEI_SPECDIR))
2462 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2463 /* Now we've got a next level subdir. */
2464 afs_snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", path1, dp1->d_name);
2465 dirp2 = opendir(path2);
2467 while ((dp2 = readdir(dirp2))) {
2468 if (*dp2->d_name == '.')
2471 /* Now we've got to the actual data */
2472 afs_snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s", path2,
2475 /* Now we've got to the actual data */
2476 afs_snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s", path1,
2479 dirp3 = opendir(path3);
2481 while ((dp3 = readdir(dirp3))) {
2482 #ifndef AFS_NT40_ENV
2483 if (*dp3->d_name == '.')
2487 #ifdef AFS_SALSRV_ENV
2490 #ifndef AFS_NT40_ENV
2497 #endif /* AFS_SALSRV_ENV */
2499 code = _namei_examine_file_spawn(&work, path3,
2506 #ifndef AFS_NT40_ENV
2514 #ifndef AFS_SALSRV_ENV
2522 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2531 #ifdef AFS_SALSRV_ENV
2532 /* effectively a barrier */
2533 code = _namei_listsubdirs_wait(wq, &resultlist);
2534 if (code < 0 || error) {
2544 /* Then why does this directory exist? Blow it away. */
2545 namei_HandleToVolDir(&name, dirIH);
2546 namei_RemoveDataDirectories(&name);
2550 #ifdef AFS_SALSRV_ENV
2552 afs_wq_wait_all(wq);
2554 _namei_listsubdirs_cleanup_results(&resultlist);
2556 if (linkHandle.fd_fd != INVALID_FD)
2557 OS_CLOSE(linkHandle.fd_fd);
2569 DecodeVolumeName(char *name, unsigned int *vid)
2571 /* Name begins with "Vol_" and ends with .data. See nt_HandleToVolDir() */
2578 if (strncmp(name, "Vol_", 4))
2580 if (strcmp(name + len - 5, ".data"))
2583 stmp[len - 5] = '\0';
2584 *vid = base32_to_int(stmp + 4);
2589 DecodeVolumeName(char *name, unsigned int *vid)
2591 if (strlen(name) < 1)
2593 *vid = (unsigned int)flipbase64_to_int64(name);
2601 * Get the inode number from the name.
2606 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2611 WIN32_FIND_DATA data;
2615 FdHandle_t linkHandle;
2618 afs_snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2620 dirH = FindFirstFileEx(fpath, FindExInfoStandard, &data,
2621 FindExSearchNameMatch, NULL,
2622 FIND_FIRST_EX_CASE_SENSITIVE);
2623 if (dirH == INVALID_HANDLE_VALUE)
2626 (void)strcpy(stmp, name);
2627 s = strrchr(stmp, '_');
2631 t = strrchr(s, '.');
2636 vno = base32_to_int(s); /* type for special files */
2637 tag = base32_to_int(t+1);
2638 info->inodeNumber = ((Inode) tag) << NAMEI_TAGSHIFT;
2639 info->inodeNumber |= vno;
2640 info->byteCount = data.nFileSizeLow;
2642 dirl = dpath[strlen(dpath)-1];
2643 if (dirl == NAMEI_SPECDIRC) { /* Special inode. */
2644 info->inodeNumber |= NAMEI_INODESPECIAL;
2645 info->u.param[0] = data.ftCreationTime.dwHighDateTime;
2646 info->u.param[1] = data.ftCreationTime.dwLowDateTime;
2647 info->u.param[2] = vno; /* type */
2648 info->u.param[3] = volid;
2649 if (vno != VI_LINKTABLE)
2650 info->linkCount = 1;
2652 /* Open this handle */
2654 (void)sprintf(lpath, "%s" OS_DIRSEP "%s", fpath, data.cFileName);
2655 linkHandle.fd_fd = OS_OPEN(lpath, O_RDONLY, 0666);
2657 namei_GetLinkCount(&linkHandle, (Inode) 0, 0, 0, 0);
2661 namei_GetLinkCount(&linkHandle, info->inodeNumber, 0, 0, 0);
2662 if (info->linkCount == 0) {
2664 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n",
2665 fpath, data.cFileName);
2666 AddToZLCDeleteList(dirl, data.cFileName);
2668 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path,
2672 info->u.param[2] = data.ftCreationTime.dwHighDateTime;
2673 info->u.param[3] = data.ftCreationTime.dwLowDateTime;
2674 info->u.param[1] = vno;
2675 info->u.param[0] = volid;
2682 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2686 struct afs_stat status;
2688 lb64_string_t check;
2690 afs_snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2692 if (afs_stat(fpath, &status) < 0) {
2696 info->byteCount = status.st_size;
2697 info->inodeNumber = (Inode) flipbase64_to_int64(name);
2699 int64_to_flipbase64(check, info->inodeNumber);
2700 if (strcmp(name, check))
2703 GetOGMFromStat(&status, &parm, &tag);
2704 if ((info->inodeNumber & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
2705 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
2706 info->u.param[0] = parm;
2707 info->u.param[1] = -1;
2708 info->u.param[2] = tag;
2709 info->u.param[3] = volid;
2711 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
2712 info->u.param[0] = volid;
2713 info->u.param[1] = (int)(info->inodeNumber & NAMEI_VNODEMASK);
2714 info->u.param[2] = (int)((info->inodeNumber >> NAMEI_UNIQSHIFT)
2715 & (Inode) NAMEI_UNIQMASK);
2716 info->u.param[3] = parm;
2723 * Convert the VolumeInfo file from RO to RW
2724 * this routine is called by namei_convertROtoRWvolume()
2727 #ifdef FSSYNC_BUILD_CLIENT
2729 convertVolumeInfo(FD_t fdr, FD_t fdw, afs_uint32 vid)
2731 struct VolumeDiskData vd;
2734 if (OS_READ(fdr, &vd, sizeof(struct VolumeDiskData)) !=
2735 sizeof(struct VolumeDiskData)) {
2736 Log("1 convertVolumeInfo: read failed for %lu with code %d\n",
2737 afs_printable_uint32_lu(vid),
2741 vd.restoredFromId = vd.id; /* remember the RO volume here */
2743 vd.id = vd.parentId;
2747 vd.uniquifier += 5000; /* just in case there are still file copies from
2748 * the old RW volume around */
2750 /* For ROs, the copyDate contains the time that the RO volume was actually
2751 * created, and the creationDate just contains the last time the RO was
2752 * copied from the RW data. So, make the new RW creationDate more accurate
2753 * by setting it to copyDate, if copyDate is older. Since, we know the
2754 * volume is at least as old as copyDate. */
2755 if (vd.copyDate < vd.creationDate) {
2756 vd.creationDate = vd.copyDate;
2758 /* If copyDate is newer, just make copyDate and creationDate the same,
2759 * for consistency with other RWs */
2760 vd.copyDate = vd.creationDate;
2763 p = strrchr(vd.name, '.');
2764 if (p && !strcmp(p, ".readonly")) {
2767 if (OS_WRITE(fdw, &vd, sizeof(struct VolumeDiskData)) !=
2768 sizeof(struct VolumeDiskData)) {
2769 Log("1 convertVolumeInfo: write failed for %lu with code %d\n",
2770 afs_printable_uint32_lu(vid),
2779 * Convert a RO-volume into a RW-volume
2781 * This function allows to recover very fast from the loss of a partition
2782 * from RO-copies if all RO-Copies exist on another partition.
2783 * Then these RO-volumes can be made to the new RW-volumes.
2784 * Backup of RW-volumes then consists in "vos release".
2786 * We must make sure in this partition exists only the RO-volume which
2787 * is typical for remote replicas.
2789 * Then the linktable is already ok,
2790 * the vnode files need to be renamed
2791 * the volinfo file needs to be replaced by another one with
2792 * slightly different contents and new name.
2793 * The volume header file of the RO-volume in the /vicep<x> directory
2794 * is destroyed by this call. A new header file for the RW-volume must
2795 * be created after return from this routine.
2799 namei_ConvertROtoRWvolume(char *pname, afs_uint32 volumeId)
2802 #ifdef FSSYNC_BUILD_CLIENT
2804 char dir_name[512], oldpath[512], newpath[512];
2819 struct DiskPartition64 *partP;
2820 struct ViceInodeInfo info;
2821 struct VolumeDiskHeader h;
2822 # ifdef AFS_DEMAND_ATTACH_FS
2824 # endif /* AFS_DEMAND_ATTACH_FS */
2826 for (partP = DiskPartitionList; partP && strcmp(partP->name, pname);
2827 partP = partP->next);
2829 Log("1 namei_ConvertROtoRWvolume: Couldn't find DiskPartition for %s\n", pname);
2834 # ifdef AFS_DEMAND_ATTACH_FS
2835 locktype = VVolLockType(V_VOLUPD, 1);
2836 code = VLockVolumeByIdNB(volumeId, partP, locktype);
2842 # endif /* AFS_DEMAND_ATTACH_FS */
2844 if (VReadVolumeDiskHeader(volumeId, partP, &h)) {
2845 Log("1 namei_ConvertROtoRWvolume: Couldn't read header for RO-volume %lu.\n",
2846 afs_printable_uint32_lu(volumeId));
2851 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_BREAKCBKS, 0, NULL);
2853 ino = namei_MakeSpecIno(h.parent, VI_LINKTABLE);
2854 IH_INIT(ih, partP->device, h.parent, ino);
2856 namei_HandleToName(&n, ih);
2857 strlcpy(dir_name, n.n_path, sizeof(dir_name));
2858 p = strrchr(dir_name, OS_DIRSEPC);
2860 dirp = opendir(dir_name);
2862 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name);
2867 while ((dp = readdir(dirp))) {
2868 /* struct ViceInodeInfo info; */
2869 #ifndef AFS_NT40_ENV
2870 if (*dp->d_name == '.')
2873 if (DecodeInode(dir_name, dp->d_name, &info, ih->ih_vid) < 0) {
2874 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s" OS_DIRSEP "%s\n",
2875 dir_name, dp->d_name);
2880 if (info.u.param[1] != -1) {
2881 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s" OS_DIRSEP "%s\n", dir_name, dp->d_name);
2886 if (info.u.param[0] != volumeId) {
2887 if (info.u.param[0] == ih->ih_vid) {
2888 if (info.u.param[2] == VI_LINKTABLE) { /* link table */
2893 Log("1 namei_ConvertROtoRWvolume: found special file %s" OS_DIRSEP "%s"
2894 " for volume %lu\n", dir_name, dp->d_name,
2895 afs_printable_uint32_lu(info.u.param[0]));
2900 if (info.u.param[2] == VI_VOLINFO) { /* volume info file */
2901 strlcpy(infoName, dp->d_name, sizeof(infoName));
2903 } else if (info.u.param[2] == VI_SMALLINDEX) { /* small vnodes file */
2904 strlcpy(smallName, dp->d_name, sizeof(smallName));
2906 } else if (info.u.param[2] == VI_LARGEINDEX) { /* large vnodes file */
2907 strlcpy(largeName, dp->d_name, sizeof(largeName));
2911 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);
2918 if (!infoSeen || !smallSeen || !largeSeen || !linkSeen) {
2919 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name);
2925 * If we come here then there was only a RO-volume and we can safely
2929 memset(&t_ih, 0, sizeof(t_ih));
2930 t_ih.ih_dev = ih->ih_dev;
2931 t_ih.ih_vid = ih->ih_vid;
2933 (void)afs_snprintf(oldpath, sizeof oldpath, "%s" OS_DIRSEP "%s", dir_name,
2935 fd = OS_OPEN(oldpath, O_RDWR, 0);
2936 if (fd == INVALID_FD) {
2937 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
2942 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_VOLINFO);
2943 namei_HandleToName(&n, &t_ih);
2944 fd2 = OS_OPEN(n.n_path, O_CREAT | O_EXCL | O_TRUNC | O_RDWR, 0);
2945 if (fd2 == INVALID_FD) {
2946 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n.n_path);
2951 code = convertVolumeInfo(fd, fd2, ih->ih_vid);
2955 OS_UNLINK(n.n_path);
2959 SetOGM(fd2, ih->ih_vid, 1);
2962 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_SMALLINDEX);
2963 namei_HandleToName(&n, &t_ih);
2964 (void)afs_snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
2966 fd = OS_OPEN(newpath, O_RDWR, 0);
2967 if (fd == INVALID_FD) {
2968 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath);
2972 SetOGM(fd, ih->ih_vid, 2);
2975 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
2977 link(newpath, n.n_path);
2981 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_LARGEINDEX);
2982 namei_HandleToName(&n, &t_ih);
2983 (void)afs_snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
2985 fd = OS_OPEN(newpath, O_RDWR, 0);
2986 if (fd == INVALID_FD) {
2987 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath);
2991 SetOGM(fd, ih->ih_vid, 3);
2994 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
2996 link(newpath, n.n_path);
3003 h.volumeInfo_hi = h.id;
3004 h.smallVnodeIndex_hi = h.id;
3005 h.largeVnodeIndex_hi = h.id;
3006 h.linkTable_hi = h.id;
3008 if (VCreateVolumeDiskHeader(&h, partP)) {
3009 Log("1 namei_ConvertROtoRWvolume: Couldn't write header for RW-volume %lu\n",
3010 afs_printable_uint32_lu(h.id));
3015 if (VDestroyVolumeDiskHeader(partP, volumeId, h.parent)) {
3016 Log("1 namei_ConvertROtoRWvolume: Couldn't unlink header for RO-volume %lu\n",
3017 afs_printable_uint32_lu(volumeId));
3020 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_DONE, 0, NULL);
3021 FSYNC_VolOp(h.id, pname, FSYNC_VOL_ON, 0, NULL);
3024 # ifdef AFS_DEMAND_ATTACH_FS
3026 VUnlockVolumeById(volumeId, partP);
3028 # endif /* AFS_DEMAND_ATTACH_FS */
3036 * returns a static string used to print either 32 or 64 bit inode numbers.
3039 PrintInode(char *s, Inode ino)
3041 static afs_ino_str_t result;
3045 (void)afs_snprintf(s, sizeof(afs_ino_str_t), "%" AFS_UINT64_FMT, (afs_uintmax_t) ino);
3052 /* Routines to facilitate removing zero link count files. */
3053 #define MAX_ZLC_NAMES 32
3054 #define MAX_ZLC_NAMELEN 16
3055 typedef struct zlcList_s {
3056 struct zlcList_s *zlc_next;
3058 char zlc_names[MAX_ZLC_NAMES][MAX_ZLC_NAMELEN];
3061 static zlcList_t *zlcAnchor = NULL;
3062 static zlcList_t *zlcCur = NULL;
3065 AddToZLCDeleteList(char dir, char *name)
3067 osi_Assert(strlen(name) <= MAX_ZLC_NAMELEN - 3);
3069 if (!zlcCur || zlcCur->zlc_n >= MAX_ZLC_NAMES) {
3070 if (zlcCur && zlcCur->zlc_next)
3071 zlcCur = zlcCur->zlc_next;
3073 zlcList_t *tmp = (zlcList_t *) malloc(sizeof(zlcList_t));
3079 zlcCur->zlc_next = tmp;
3083 zlcCur->zlc_next = NULL;
3088 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%c" OS_DIRSEP "%s", dir, name);
3090 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%s", name);
3096 DeleteZLCFiles(char *path)
3102 for (z = zlcAnchor; z; z = z->zlc_next) {
3103 for (i = 0; i < z->zlc_n; i++) {
3105 (void)sprintf(fname, "%s" OS_DIRSEP "%s", path, z->zlc_names[i]);
3107 (void)sprintf(fname, "%s", z->zlc_names[i]);
3108 if (namei_unlink(fname) < 0) {
3109 Log("ZLC: Can't unlink %s, dos error = %d\n", fname,
3113 z->zlc_n = 0; /* Can reuse space. */
3126 tnext = i->zlc_next;
3130 zlcCur = zlcAnchor = NULL;
3134 #endif /* AFS_NAMEI_ENV */