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>
20 #ifdef HAVE_SYS_FILE_H
21 # include <sys/file.h>
33 #include <rx/rx_queue.h>
34 #ifdef AFS_PTHREAD_ENV
35 # include <opr/lock.h>
38 #include <afs/afsutil.h>
41 #include <afs/afsint.h>
45 #include "viceinode.h"
47 #include "partition.h"
49 #include "volume_inline.h"
51 #include <afs/errors.h>
54 #include <afs/errmap_nt.h>
58 #define LOCK_SH 1 /* shared lock */
59 #define LOCK_EX 2 /* exclusive lock */
60 #define LOCK_NB 4 /* don't block when locking */
61 #define LOCK_UN 8 /* unlock */
66 #include <afs/work_queue.h>
67 #include <afs/thread_pool.h>
68 #include <vol/vol-salvage.h>
73 static void namei_UnlockLinkCount(FdHandle_t
* fdP
, Inode ino
);
76 namei_iread(IHandle_t
* h
, afs_foff_t offset
, char *buf
, afs_fsize_t size
)
85 nBytes
= FDH_PREAD(fdP
, buf
, size
, offset
);
94 namei_iwrite(IHandle_t
* h
, afs_foff_t offset
, char *buf
, afs_fsize_t size
)
103 nBytes
= FDH_PWRITE(fdP
, buf
, size
, offset
);
105 FDH_REALLYCLOSE(fdP
);
112 /* Inode number format:
113 * low 32 bits - if a regular file or directory, the vnode; else the type.
114 * 32-36 - uniquifier tag and index into counts array for this vnode. Only
115 * two of the available bits are currently used. The rest are
116 * present in case we ever increase the number of types of volumes
117 * in the volume group.
118 * bit 37 : 1 == special, 0 == regular
120 # define NAMEI_VNODEMASK 0x00ffffffff
121 # define NAMEI_TAGSHIFT 32
122 # define NAMEI_INODESPECIAL 0x2000000000
123 # define NAMEI_SPECDIR "R"
124 # define NAMEI_SPECDIRC 'R'
125 #else /* !AFS_NT40_ENV */
126 /* Inode number format:
127 * low 26 bits - vnode number - all 1's if volume special file.
129 * next 3 bits spare (0's)
130 * high 32 bits - uniquifier (regular) or type if spare
132 # define NAMEI_VNODEMASK 0x003ffffff
133 # define NAMEI_TAGSHIFT 26
134 # define NAMEI_UNIQMASK 0xffffffff
135 # define NAMEI_UNIQSHIFT 32
136 # define NAMEI_INODESPECIAL ((Inode)NAMEI_VNODEMASK)
137 /* dir1 is the high 8 bits of the 26 bit vnode */
138 # define VNO_DIR1(vno) ((vno >> 14) & 0xff)
139 /* dir2 is the next 9 bits */
140 # define VNO_DIR2(vno) ((vno >> 9) & 0x1ff)
141 /* "name" is the low 9 bits of the vnode, the 3 bit tag and the uniq */
142 # define NAMEI_SPECDIR "special"
143 #endif /* !AFS_NT40_ENV */
144 #define NAMEI_TAGMASK 0x7
145 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
147 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
149 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
150 * group, not counting temporary (move) volumes.
151 * This is the number of separate files, all having
152 * the same vnode number, which can occur in a volume
164 static int GetFreeTag(IHandle_t
* ih
, int vno
);
166 /* namei_HandleToInodeDir
168 * Construct the path name of the directory holding the inode data.
169 * Format: /<vicepx>/INODEDIR
174 namei_HandleToInodeDir(namei_t
* name
, IHandle_t
* ih
)
176 memset(name
, '\0', sizeof(*name
));
177 nt_DevToDrive(name
->n_drive
, ih
->ih_dev
);
178 strlcpy(name
->n_path
, name
->n_drive
, sizeof(name
->n_path
));
182 /* Format: /<vicepx>/INODEDIR */
183 #define PNAME_BLEN 64
185 namei_HandleToInodeDir(namei_t
* name
, IHandle_t
* ih
)
189 memset(name
, '\0', sizeof(*name
));
192 * Add the /vicepXX string to the start of name->n_base and then calculate
193 * offset as the number of bytes we know we added.
195 * FIXME: This embeds knowledge of the vice partition naming scheme and
196 * mapping from device numbers. There needs to be an API that tells us
199 volutil_PartitionName_r(ih
->ih_dev
, name
->n_base
, sizeof(name
->n_base
));
200 offset
= VICE_PREFIX_SIZE
+ (ih
->ih_dev
> 25 ? 2 : 1);
201 name
->n_base
[offset
] = OS_DIRSEPC
;
203 strlcpy(name
->n_base
+ offset
, INODEDIR
, sizeof(name
->n_base
) - offset
);
204 strlcpy(name
->n_path
, name
->n_base
, sizeof(name
->n_path
));
208 #define addtoname(N, C) \
210 if ((N)->n_path[strlen((N)->n_path)-1] != OS_DIRSEPC) \
211 strlcat((N)->n_path, OS_DIRSEP, sizeof((N)->n_path)); \
212 strlcat((N)->n_path, (C), sizeof((N)->n_path)); \
218 namei_HandleToVolDir(namei_t
* name
, IHandle_t
* ih
)
220 /* X:\Vol_XXXXXXX.data */
224 namei_HandleToInodeDir(name
, ih
);
225 /* nt_drive added to name by namei_HandleToInodeDir() */
226 namep
= name
->n_voldir
;
227 (void)memcpy(namep
, "\\Vol_", 5);
229 (void)strcpy(namep
, int_to_base32(str1
, ih
->ih_vid
));
230 namep
+= strlen(namep
);
231 memcpy(namep
, ".data", 5);
234 addtoname(name
, name
->n_voldir
);
238 namei_HandleToVolDir(namei_t
* name
, IHandle_t
* ih
)
242 namei_HandleToInodeDir(name
, ih
);
243 (void)int32_to_flipbase64(tmp
, (int64_t) (ih
->ih_vid
& 0xff));
244 strlcpy(name
->n_voldir1
, tmp
, sizeof(name
->n_voldir1
));
245 addtoname(name
, name
->n_voldir1
);
246 (void)int32_to_flipbase64(tmp
, (int64_t) ih
->ih_vid
);
247 strlcpy(name
->n_voldir2
, tmp
, sizeof(name
->n_voldir2
));
248 addtoname(name
, name
->n_voldir2
);
252 /* namei_HandleToName
254 * Constructs a file name for the fully qualified handle.
257 /* Note that special files end up in X:\Vol_XXXXXXX.data\R */
259 namei_HandleToName(namei_t
* name
, IHandle_t
* ih
)
261 int vno
= (int)(ih
->ih_ino
& NAMEI_VNODEMASK
);
262 int special
= (ih
->ih_ino
& NAMEI_INODESPECIAL
)?1:0;
263 int tag
= (int)((ih
->ih_ino
>> NAMEI_TAGSHIFT
) & NAMEI_TAGMASK
);
266 namei_HandleToVolDir(name
, ih
);
269 name
->n_dir
[0] = NAMEI_SPECDIRC
;
272 name
->n_dir
[0] = 'Q';
274 name
->n_dir
[0] = ((vno
& 0x1f) >> 1) + 'A';
277 name
->n_dir
[1] = '\0';
278 addtoname(name
, name
->n_dir
);
279 /* X:\Vol_XXXXXXX.data\X\V_XXXXXXX.XXX */
280 namep
= name
->n_inode
;
281 (void)memcpy(namep
, "\\V_", 3);
283 (void)strcpy(namep
, int_to_base32(str1
, vno
));
284 namep
+= strlen(namep
);
286 (void)strcpy(namep
, int_to_base32(str1
, tag
));
287 namep
+= strlen(namep
);
288 addtoname(name
, name
->n_inode
);
291 /* Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special */
293 namei_HandleToName(namei_t
* name
, IHandle_t
* ih
)
295 int vno
= (int)(ih
->ih_ino
& NAMEI_VNODEMASK
);
298 namei_HandleToVolDir(name
, ih
);
300 if (vno
== NAMEI_VNODESPECIAL
) {
301 strlcpy(name
->n_dir1
, NAMEI_SPECDIR
, sizeof(name
->n_dir1
));
302 addtoname(name
, name
->n_dir1
);
303 name
->n_dir2
[0] = '\0';
305 (void)int32_to_flipbase64(str
, VNO_DIR1(vno
));
306 strlcpy(name
->n_dir1
, str
, sizeof(name
->n_dir1
));
307 addtoname(name
, name
->n_dir1
);
308 (void)int32_to_flipbase64(str
, VNO_DIR2(vno
));
309 strlcpy(name
->n_dir2
, str
, sizeof(name
->n_dir2
));
310 addtoname(name
, name
->n_dir2
);
312 (void)int64_to_flipbase64(str
, (int64_t) ih
->ih_ino
);
313 strlcpy(name
->n_inode
, str
, sizeof(name
->n_inode
));
314 addtoname(name
, name
->n_inode
);
319 /* The following is a warning to tell sys-admins to not muck about in this
322 #define VICE_README "These files and directories are a part of the AFS \
323 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
324 ownership and permissions included.\n"
326 namei_ViceREADME(char *partition
)
331 /* Create the inode directory if we're starting for the first time */
332 snprintf(filename
, sizeof filename
, "%s" OS_DIRSEP
"%s", partition
,
334 mkdir(filename
, 0700);
336 snprintf(filename
, sizeof filename
,
337 "%s" OS_DIRSEP
"%s" OS_DIRSEP
"README",
338 partition
, INODEDIR
);
339 fd
= OS_OPEN(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
, 0444);
340 if (fd
!= INVALID_FD
) {
341 len
= strlen(VICE_README
);
342 if (OS_WRITE(fd
, VICE_README
, len
) != len
)
352 /* namei_CreateDataDirectories
354 * If creating the file failed because of ENOENT or ENOTDIR, try
355 * creating all the directories first.
359 namei_CreateDataDirectories(namei_t
* name
, int *created
)
366 snprintf(tmp
, 256, "%s" OS_DIRSEP
"%s", name
->n_drive
, name
->n_voldir
);
368 if (mkdir(tmp
) < 0) {
379 for (i
= 'A'; i
<= NAMEI_SPECDIRC
; i
++) {
381 if (mkdir(tmp
) < 0 && errno
!= EEXIST
)
387 #define create_dir() \
389 if (mkdir(tmp, 0700)<0) { \
390 if (errno != EEXIST) \
398 #define create_nextdir(A) \
400 strcat(tmp, OS_DIRSEP); strcat(tmp, A); create_dir(); \
404 namei_CreateDataDirectories(namei_t
* name
, int *created
)
410 strlcpy(tmp
, name
->n_base
, sizeof(tmp
));
413 create_nextdir(name
->n_voldir1
);
414 create_nextdir(name
->n_voldir2
);
415 create_nextdir(name
->n_dir1
);
416 if (name
->n_dir2
[0]) {
417 create_nextdir(name
->n_dir2
);
424 /* delTree(): Deletes an entire tree of directories (no files)
426 * root : Full path to the subtree. Should be big enough for PATH_MAX
427 * tree : the subtree to be deleted is rooted here. Specifies only the
428 * subtree beginning at tree (not the entire path). It should be
429 * a pointer into the "root" buffer.
431 * errp : errno of the first error encountered during the directory cleanup.
432 * *errp should have been initialized to 0.
435 * -1 : If errors were encountered during cleanup and error is set to
439 * If there are errors, we try to work around them and delete as many
440 * directories as possible. We don't attempt to remove directories that still
441 * have non-dir entries in them.
444 delTree(char *root
, char *tree
, int *errp
)
452 /* delete the children first */
453 cp
= strchr(tree
, OS_DIRSEPC
);
455 delTree(root
, cp
+ 1, errp
);
458 cp
= tree
+ strlen(tree
); /* move cp to the end of string tree */
460 /* now delete all entries in this dir */
461 if ((ds
= opendir(root
)) != NULL
) {
463 while ((dirp
= readdir(ds
))) {
464 /* ignore . and .. */
465 if (!strcmp(dirp
->d_name
, ".") || !strcmp(dirp
->d_name
, ".."))
467 /* since root is big enough, we reuse the space to
468 * concatenate the dirname to the current tree
470 strcat(root
, OS_DIRSEP
);
471 strcat(root
, dirp
->d_name
);
472 if (afs_stat(root
, &st
) == 0 && S_ISDIR(st
.st_mode
)) {
473 /* delete this subtree */
474 delTree(root
, cp
+ 1, errp
);
476 *errp
= *errp
? *errp
: errno
;
478 /* recover path to our cur tree by truncating it to
483 /* if (!errno) -- closedir not implicit if we got an error */
487 /* finally axe the current dir */
489 *errp
= *errp
? *errp
: errno
;
491 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
493 #endif /* !AFS_PTHREAD_ENV */
498 /* if we encountered errors during cleanup, we return a -1 */
507 /* namei_RemoveDataDirectories
509 * Returns 0 on success.
510 * Returns -1 on error. Typically, callers ignore this error because we
511 * can continue running if the removes fail. The salvage process will
512 * finish tidying up for us.
517 namei_RemoveDataDirectories(namei_t
* name
)
524 snprintf(tmp
, 256, "%s" OS_DIRSEP
"%s", name
->n_drive
, name
->n_voldir
);
527 path
+= strlen(path
);
528 *path
++ = OS_DIRSEPC
;
530 for (i
= 'A'; i
<= NAMEI_SPECDIRC
; i
++) {
532 if (rmdir(name
->n_path
) < 0 && errno
!= ENOENT
)
537 /* Delete the Vol_NNNNNN.data directory. */
540 if (rmdir(name
->n_path
) < 0 && errno
!= ENOENT
) {
548 * We only use the n_base and n_voldir1 entries
549 * and only do rmdir's.
552 namei_RemoveDataDirectories(namei_t
* name
)
556 int prefixlen
= strlen(name
->n_base
), err
= 0;
557 int vollen
= strlen(name
->n_voldir1
);
558 char pbuf
[MAXPATHLEN
];
562 strlcpy(path
, name
->n_path
, sizeof(pbuf
));
564 /* move past the prefix and n_voldir1 */
565 path
= path
+ prefixlen
+ 1 + vollen
+ 1; /* skip over the trailing / */
567 /* now delete all dirs upto path */
568 code
= delTree(pbuf
, path
, &err
);
570 /* We've now deleted everything under /n_base/n_voldir1/n_voldir2 that
571 * we could. Do not delete /n_base/n_voldir1, since doing such might
572 * interrupt another thread trying to create a volume. We could introduce
573 * some locking to make this safe (or only remove it for whole-partition
574 * salvages), but by not deleting it we only leave behind a maximum of
575 * 256 empty directories. So at least for now, don't bother. */
580 /* Create the file in the name space.
582 * Parameters stored as follows:
584 * p1 - volid - implied in containing directory.
585 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
586 * p3 - uniq -- bits 4-0 are in mode bits 4-0
587 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
589 * p1 - volid - creation time - dwHighDateTime
590 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
591 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
592 * p4 - parid - parent volume id - implied in containing directory.
594 * Return value is the inode number or (Inode)-1 if error.
595 * We "know" there is only one link table, so return EEXIST if there already
596 * is a link table. It's up to the calling code to test errno and increment
602 * This function is called by VCreateVolume to hide the implementation
603 * details of the inode numbers. This only allows for 7 volume special
604 * types, but if we get that far, this could should be dead by then.
607 namei_MakeSpecIno(VolumeId volid
, int type
)
610 ino
= NAMEI_INODESPECIAL
;
613 /* tag is always 0 for special */
615 type
&= NAMEI_TAGMASK
;
616 ino
|= ((Inode
) type
) << NAMEI_TAGSHIFT
;
617 ino
|= ((Inode
) volid
) << NAMEI_UNIQSHIFT
;
625 SetOGM(FD_t fd
, int parm
, int tag
)
631 SetWinOGM(FD_t fd
, int p1
, int p2
)
636 ftime
.dwHighDateTime
= p1
;
637 ftime
.dwLowDateTime
= p2
;
639 code
= SetFileTime(fd
, &ftime
, NULL
/*access*/, NULL
/*write*/);
646 GetWinOGM(FD_t fd
, int *p1
, int *p2
)
651 code
= GetFileTime(fd
, &ftime
, NULL
/*access*/, NULL
/*write*/);
655 *p1
= ftime
.dwHighDateTime
;
656 *p2
= ftime
.dwLowDateTime
;
662 CheckOGM(FdHandle_t
*fdP
, int p1
)
666 if (GetWinOGM(fdP
->fd_fd
, &ogm_p1
, &ogm_p2
)) {
678 FixSpecialOGM(FdHandle_t
*fdP
, int check
)
680 Inode ino
= fdP
->fd_ih
->ih_ino
;
681 VnodeId vno
= NAMEI_VNODESPECIAL
, ogm_vno
;
684 if (GetWinOGM(fdP
->fd_fd
, &ogm_volid
, &ogm_vno
)) {
688 /* the only thing we can check is the vnode number; for the volid we have
689 * nothing else to compare against */
690 if (vno
!= ogm_vno
) {
694 if (SetWinOGM(fdP
->fd_fd
, ogm_volid
, vno
)) {
701 #else /* AFS_NT40_ENV */
702 /* SetOGM - set owner group and mode bits from parm and tag */
704 SetOGM(FD_t fd
, int parm
, int tag
)
707 * owner - low 15 bits of parm.
708 * group - next 15 bits of parm.
709 * mode - 2 bits of parm, then lowest = 3 bits of tag.
711 int owner
, group
, mode
;
713 owner
= parm
& 0x7fff;
714 group
= (parm
>> 15) & 0x7fff;
715 if (fchown(fd
, owner
, group
) < 0)
718 mode
= (parm
>> 27) & 0x18;
720 if (fchmod(fd
, mode
) < 0)
725 /* GetOGM - get parm and tag from owner, group and mode bits. */
727 GetOGMFromStat(struct afs_stat_st
*status
, int *parm
, int *tag
)
729 *parm
= status
->st_uid
| (status
->st_gid
<< 15);
730 *parm
|= (status
->st_mode
& 0x18) << 27;
731 *tag
= status
->st_mode
& 0x7;
735 GetOGM(FdHandle_t
*fdP
, int *parm
, int *tag
)
737 struct afs_stat_st status
;
738 if (afs_fstat(fdP
->fd_fd
, &status
) < 0)
740 GetOGMFromStat(&status
, parm
, tag
);
745 CheckOGM(FdHandle_t
*fdP
, int p1
)
749 if (GetOGM(fdP
, &parm
, &tag
) < 0)
758 FixSpecialOGM(FdHandle_t
*fdP
, int check
)
760 int inode_volid
, ogm_volid
;
761 int inode_type
, ogm_type
;
762 Inode ino
= fdP
->fd_ih
->ih_ino
;
764 inode_volid
= ((ino
>> NAMEI_UNIQSHIFT
) & NAMEI_UNIQMASK
);
765 inode_type
= (int)((ino
>> NAMEI_TAGSHIFT
) & NAMEI_TAGMASK
);
767 if (GetOGM(fdP
, &ogm_volid
, &ogm_type
) < 0) {
768 Log("Error retrieving OGM info\n");
772 if (inode_volid
!= ogm_volid
|| inode_type
!= ogm_type
) {
773 Log("%sIncorrect OGM data (ino: vol %u type %d) (ogm: vol %u type %d)\n",
774 check
?"":"Fixing ", inode_volid
, inode_type
, ogm_volid
, ogm_type
);
780 if (SetOGM(fdP
->fd_fd
, inode_volid
, inode_type
) < 0) {
781 Log("Error setting OGM data\n");
788 #endif /* !AFS_NT40_ENV */
791 * Check/fix the OGM data for an inode
793 * @param[in] fdP Open file handle for the inode to check
794 * @param[in] check 1 to just check the OGM data, and return an error if it
795 * is incorrect. 0 to fix the OGM data if it is incorrect.
797 * @pre fdP must be for a special inode
804 namei_FixSpecialOGM(FdHandle_t
*fdP
, int check
)
807 Inode ino
= fdP
->fd_ih
->ih_ino
;
809 vnode
= (int)(ino
& NAMEI_VNODEMASK
);
810 if (vnode
!= NAMEI_VNODESPECIAL
) {
811 Log("FixSpecialOGM: non-special vnode %u\n", vnode
);
815 return FixSpecialOGM(fdP
, check
);
818 int big_vno
= 0; /* Just in case we ever do 64 bit vnodes. */
820 /* Derive the name and create it O_EXCL. If that fails we have an error.
821 * Get the tag from a free column in the link table.
825 namei_icreate(IHandle_t
* lh
, char *part
, afs_uint32 p1
, afs_uint32 p2
, afs_uint32 p3
, afs_uint32 p4
)
828 FD_t fd
= INVALID_FD
;
839 memset((void *)&tmp
, 0, sizeof(IHandle_t
));
840 memset(&tfd
, 0, sizeof(FdHandle_t
));
842 tmp
.ih_dev
= nt_DriveToDev(part
);
843 if (tmp
.ih_dev
== -1) {
848 if (p2
== INODESPECIAL
) {
849 /* Parameters for special file:
850 * p1 - volume id - goes into owner/group/mode
851 * p2 - vnode == INODESPECIAL
853 * p4 - parent volume id
858 tmp
.ih_vid
= p4
; /* Use parent volume id, where this file will be. */
859 tmp
.ih_ino
= namei_MakeSpecIno(p1
, p3
);
861 int vno
= p2
& NAMEI_VNODEMASK
;
862 /* Parameters for regular file:
876 tmp
.ih_ino
= (Inode
) p2
;
881 namei_HandleToName(&name
, &tmp
);
882 p
= strrchr((char *)&name
.n_path
, '.');
884 for (tag
= 0; tag
< NAMEI_MAXVOLS
; tag
++) {
885 *p
= *int_to_base32(str1
, tag
);
886 fd
= OS_OPEN((char *)&name
.n_path
, O_CREAT
| O_RDWR
| O_EXCL
, 0666);
887 if (fd
== INVALID_FD
) {
888 if (errno
== ENOTDIR
|| errno
== ENOENT
) {
889 if (namei_CreateDataDirectories(&name
, &created_dir
) == 0)
890 fd
= OS_OPEN((char *)&name
.n_path
, O_CREAT
| O_RDWR
| O_EXCL
, 0666);
894 if (fd
!= INVALID_FD
)
896 if (p2
== INODESPECIAL
&& p3
== VI_LINKTABLE
)
899 if (fd
== INVALID_FD
) {
903 tmp
.ih_ino
&= ~(((Inode
) NAMEI_TAGMASK
) << NAMEI_TAGSHIFT
);
904 tmp
.ih_ino
|= (((Inode
) tag
) << NAMEI_TAGSHIFT
);
907 if (SetWinOGM(fd
, ogm_p1
, ogm_p2
)) {
908 errno
= OS_ERROR(EBADF
);
914 if (p2
!= INODESPECIAL
) {
915 if (fd
== INVALID_FD
) {
917 code
= nt_unlink((char *)&name
.n_path
);
926 code
= namei_SetLinkCount(fdP
, tmp
.ih_ino
, 1, 0);
928 } else if (p2
== INODESPECIAL
&& p3
== VI_LINKTABLE
) {
929 if (fd
== INVALID_FD
)
931 /* hack at tmp to setup for set link count call. */
933 code
= namei_SetLinkCount(&tfd
, (Inode
) 0, 1, 0);
938 if (fd
!= INVALID_FD
)
941 if (code
|| (fd
== INVALID_FD
)) {
942 if (p2
!= INODESPECIAL
) {
945 namei_SetLinkCount(fdP
, tmp
.ih_ino
, 0, 0);
951 int save_errno
= errno
;
952 namei_RemoveDataDirectories(&name
);
956 return (code
|| (fd
== INVALID_FD
)) ? (Inode
) -1 : tmp
.ih_ino
;
958 #else /* !AFS_NT40_ENV */
960 icreate(IHandle_t
* lh
, char *part
, afs_uint32 p1
, afs_uint32 p2
, afs_uint32 p3
, afs_uint32 p4
,
968 IHandle_t
*realh
= NULL
;
973 memset((void *)&tmp
, 0, sizeof(IHandle_t
));
975 tmp
.ih_dev
= volutil_GetPartitionID(part
);
976 if (tmp
.ih_dev
== -1) {
982 /* Parameters for special file:
983 * p1 - volume id - goes into owner/group/mode
986 * p4 - parent volume id
991 tmp
.ih_vid
= p4
; /* Use parent volume id, where this file will be. */
992 tmp
.ih_ino
= namei_MakeSpecIno(p1
, p3
);
994 int vno
= p2
& NAMEI_VNODEMASK
;
995 /* Parameters for regular file:
1007 /* If GetFreeTag succeeds, it atomically sets link count to 1. */
1008 tag
= GetFreeTag(lh
, p2
);
1013 tmp
.ih_ino
= (Inode
) p2
;
1014 /* name is <uniq(p3)><tag><vno(p2)> */
1015 tmp
.ih_ino
|= ((Inode
) tag
) << NAMEI_TAGSHIFT
;
1016 tmp
.ih_ino
|= ((Inode
) p3
) << NAMEI_UNIQSHIFT
;
1021 namei_HandleToName(&name
, &tmp
);
1022 fd
= OS_OPEN(name
.n_path
, O_CREAT
| O_EXCL
| O_RDWR
, 0);
1023 if (fd
== INVALID_FD
) {
1024 if (errno
== ENOTDIR
|| errno
== ENOENT
) {
1025 if (namei_CreateDataDirectories(&name
, &created_dir
) < 0)
1027 fd
= OS_OPEN(name
.n_path
, O_CREAT
| O_EXCL
| O_RDWR
,
1029 if (fd
== INVALID_FD
)
1035 if (SetOGM(fd
, ogm_parm
, tag
) < 0) {
1041 IH_INIT(realh
, tmp
.ih_dev
, tmp
.ih_vid
, tmp
.ih_ino
);
1042 fdP
= ih_attachfd(realh
, fd
);
1044 /* ih_attachfd can only return NULL if we give it an invalid fd; our fd
1045 * must be valid by this point. */
1048 if (p2
== (afs_uint32
)-1 && p3
== VI_LINKTABLE
) {
1049 code
= namei_SetLinkCount(fdP
, (Inode
) 0, 1, 0);
1055 if (code
|| (fd
== INVALID_FD
)) {
1059 namei_SetLinkCount(fdP
, tmp
.ih_ino
, 0, 0);
1072 namei_icreate(IHandle_t
* lh
, char *part
,
1073 afs_uint32 p1
, afs_uint32 p2
, afs_uint32 p3
, afs_uint32 p4
)
1076 IHandle_t
*ihP
= NULL
;
1079 code
= icreate(lh
, part
, p1
, p2
, p3
, p4
, &ihP
);
1091 namei_icreate_init(IHandle_t
* lh
, int dev
, char *part
,
1092 afs_uint32 p1
, afs_uint32 p2
, afs_uint32 p3
, afs_uint32 p4
)
1095 IHandle_t
*ihP
= NULL
;
1097 code
= icreate(lh
, part
, p1
, p2
, p3
, p4
, &ihP
);
1107 namei_iopen(IHandle_t
* h
)
1112 /* Convert handle to file name. */
1113 namei_HandleToName(&name
, h
);
1114 fd
= OS_OPEN((char *)&name
.n_path
, O_RDWR
, 0666);
1118 /* Need to detect vol special file and just unlink. In those cases, the
1119 * handle passed in _is_ for the inode. We only check p1 for the special
1123 namei_dec(IHandle_t
* ih
, Inode ino
, int p1
)
1130 if ((ino
& NAMEI_INODESPECIAL
) == NAMEI_INODESPECIAL
) {
1132 int type
= (int)((ino
>> NAMEI_TAGSHIFT
) & NAMEI_TAGMASK
);
1134 /* Verify this is the right file. */
1135 IH_INIT(tmp
, ih
->ih_dev
, ih
->ih_vid
, ino
);
1137 namei_HandleToName(&name
, tmp
);
1142 errno
= OS_ERROR(ENOENT
);
1146 if (CheckOGM(fdP
, p1
) < 0) {
1147 FDH_REALLYCLOSE(fdP
);
1149 errno
= OS_ERROR(EINVAL
);
1153 /* If it's the link table itself, decrement the link count. */
1154 if (type
== VI_LINKTABLE
) {
1155 if ((count
= namei_GetLinkCount(fdP
, (Inode
) 0, 1, 0, 1)) < 0) {
1156 FDH_REALLYCLOSE(fdP
);
1163 /* if our count is non-zero, we just set our new linkcount and
1164 * return. But if our count is 0, don't bother updating the
1165 * linktable, since we're about to delete the link table,
1167 if (namei_SetLinkCount(fdP
, (Inode
) 0, count
< 0 ? 0 : count
, 1) < 0) {
1168 FDH_REALLYCLOSE(fdP
);
1178 namei_UnlockLinkCount(fdP
, (Inode
) 0);
1181 /* We should IH_REALLYCLOSE right before deleting the special file from
1182 * disk, to ensure that somebody else cannot create a special inode,
1183 * then IH_OPEN that special inode and get back a cached fd for the
1184 * file we are deleting here (instead of an fd for the file they just
1186 IH_REALLYCLOSE(tmp
);
1187 FDH_REALLYCLOSE(fdP
);
1190 if ((code
= OS_UNLINK(name
.n_path
)) == 0) {
1191 if (type
== VI_LINKTABLE
) {
1192 /* Try to remove directory. If it fails, that's ok.
1193 * Salvage will clean up.
1195 char *slash
= strrchr(name
.n_path
, OS_DIRSEPC
);
1197 /* avoid an rmdir() on the file we just unlinked */
1200 (void)namei_RemoveDataDirectories(&name
);
1204 /* Get a file descriptor handle for this Inode */
1210 if ((count
= namei_GetLinkCount(fdP
, ino
, 1, 0, 1)) < 0) {
1211 FDH_REALLYCLOSE(fdP
);
1217 if (namei_SetLinkCount(fdP
, ino
, count
, 1) < 0) {
1218 FDH_REALLYCLOSE(fdP
);
1223 IH_INIT(th
, ih
->ih_dev
, ih
->ih_vid
, ino
);
1224 Log("Warning: Lost ref on ihandle dev %d vid %" AFS_VOLID_FMT
" ino %lld\n",
1225 th
->ih_dev
, afs_printable_VolumeId_lu(th
->ih_vid
), (afs_int64
)th
->ih_ino
);
1228 /* If we're less than 0, someone presumably unlinked;
1229 don't bother setting count to 0, but we need to drop a lock */
1230 if (namei_SetLinkCount(fdP
, ino
, 0, 1) < 0) {
1231 FDH_REALLYCLOSE(fdP
);
1237 IH_INIT(th
, ih
->ih_dev
, ih
->ih_vid
, ino
);
1239 namei_HandleToName(&name
, th
);
1241 code
= OS_UNLINK(name
.n_path
);
1250 namei_inc(IHandle_t
* h
, Inode ino
, int p1
)
1256 if ((ino
& NAMEI_INODESPECIAL
) == NAMEI_INODESPECIAL
) {
1257 int type
= (int)((ino
>> NAMEI_TAGSHIFT
) & NAMEI_TAGMASK
);
1258 if (type
!= VI_LINKTABLE
)
1263 /* Get a file descriptor handle for this Inode */
1269 if ((count
= namei_GetLinkCount(fdP
, ino
, 1, 0, 1)) < 0)
1274 errno
= OS_ERROR(EINVAL
);
1278 if (namei_SetLinkCount(fdP
, ino
, count
, 1) < 0)
1282 FDH_REALLYCLOSE(fdP
);
1289 #ifndef AFS_NT40_ENV
1291 namei_replace_file_by_hardlink(IHandle_t
*hLink
, IHandle_t
*hTarget
)
1297 /* Convert handle to file name. */
1298 namei_HandleToName(&nameLink
, hLink
);
1299 namei_HandleToName(&nameTarget
, hTarget
);
1301 OS_UNLINK(nameLink
.n_path
);
1302 code
= link(nameTarget
.n_path
, nameLink
.n_path
);
1307 namei_copy_on_write(IHandle_t
*h
)
1313 struct afs_stat_st tstat
;
1316 namei_HandleToName(&name
, h
);
1317 if (afs_stat(name
.n_path
, &tstat
) < 0)
1319 if (tstat
.st_nlink
> 1) { /* do a copy on write */
1320 char path
[NAMEI_PATH_LEN
+ 4];
1328 snprintf(path
, sizeof(path
), "%s-tmp", name
.n_path
);
1329 fd
= OS_OPEN(path
, O_CREAT
| O_EXCL
| O_RDWR
, 0);
1330 if (fd
== INVALID_FD
) {
1341 size
= tstat
.st_size
;
1344 tlen
= size
> 8192 ? 8192 : size
;
1345 if (FDH_PREAD(fdP
, buf
, tlen
, offset
) != tlen
)
1347 if (OS_WRITE(fd
, buf
, tlen
) != tlen
)
1353 FDH_REALLYCLOSE(fdP
);
1358 OS_UNLINK(name
.n_path
);
1359 code
= rename(path
, name
.n_path
);
1366 /************************************************************************
1367 * File Name Structure
1368 ************************************************************************
1370 * Each AFS file needs a unique name and it needs to be findable with
1371 * minimal lookup time. Note that the constraint on the number of files and
1372 * directories in a volume is the size of the vnode index files and the
1373 * max file size AFS supports (for internal files) of 2^31. Since a record
1374 * in the small vnode index file is 64 bytes long, we can have at most
1375 * (2^31)/64 or 33554432 files. A record in the large index file is
1376 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
1377 * Another layout parameter is that there is roughly a 16 to 1 ratio between
1378 * the number of files and the number of directories.
1380 * Using this information we can see that a layout of 256 directories, each
1381 * with 512 subdirectories and each of those having 512 files gives us
1382 * 256*512*512 = 67108864 AFS files and directories.
1384 * The volume, vnode, uniquifier and data version, as well as the tag
1385 * are required, either for finding the file or for salvaging. It's best to
1386 * restrict the name to something that can be mapped into 64 bits so the
1387 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
1388 * is used to distinguish between different versions of the same file
1389 * which are currently in the RW and clones of a volume. See "Link Table
1390 * Organization" below for more information on the tag. The tag is
1391 * required in the name of the file to ensure a unique name.
1393 * ifdef AFS_NT40_ENV
1394 * The data for each volume group is in a separate directory. The name of the
1395 * volume is of the form: Vol_NNNNNN.data, where NNNNNN is a base 32
1396 * representation of the RW volume ID (even where the RO is the only volume
1397 * on the partition). Below that are separate subdirectories for the
1398 * AFS directories and special files. There are also 16 directories for files,
1399 * hashed on the low 5 bits (recall bit0 is always 0) of the vnode number.
1400 * These directories are named:
1401 * A - P - 16 file directories.
1402 * Q ----- data directory
1403 * R ----- special files directory
1405 * The vnode is hashed into the directory using the low bits of the
1408 * The format of a file name for a regular file is:
1409 * Y:\Vol_NNNNNN.data\X\V_IIIIII.J
1410 * Y - partition encoded as drive letter, starting with D
1411 * NNNNNN - base 32 encoded volume number of RW volume
1412 * X - hash directory, as above
1413 * IIIIII - base 32 encoded vnode number
1414 * J - base 32 encoded tag
1416 * uniq is stored in the dwHighDateTime creation time field
1417 * dv is stored in the dwLowDateTime creation time field
1419 * Special inodes are always in the R directory, as above, and are
1421 * True child volid is stored in the dwHighDateTime creation time field
1422 * vnode number is always -1 (Special)
1423 * type is the IIIIII part of the filename
1424 * uniq is the J part of the filename
1425 * parent volume id is implied in the containing directory
1428 * We can store data in the uid, gid and mode bits of the files, provided
1429 * the directories have root only access. This gives us 15 bits for each
1430 * of uid and gid (GNU chown considers 65535 to mean "don't change").
1431 * There are 9 available mode bits. Adn we need to store a total of
1432 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
1433 * or 131 bits somewhere.
1435 * The format of a file name for a regular file is:
1436 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
1437 * V1 - low 8 bits of RW volume id
1438 * V2 - all bits of RW volume id
1439 * AA - high 8 bits of vnode number.
1440 * BB - next 9 bits of vnode number.
1441 * <tag><uniq><vno> - file name
1443 * Volume special files are stored in a separate directory:
1444 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
1447 * The vnode is hashed into the directory using the high bits of the
1448 * vnode number. This is so that consecutively created vnodes are in
1449 * roughly the same area on the disk. This will at least be optimal if
1450 * the user is creating many files in the same AFS directory. The name
1451 * should be formed so that the leading characters are different as quickly
1452 * as possible, leading to faster discards of incorrect matches in the
1460 /************************************************************************
1461 * Link Table Organization
1462 ************************************************************************
1464 * The link table volume special file is used to hold the link counts that
1465 * are held in the inodes in inode based AFS vice filesystems. For user
1466 * space access, the link counts are being kept in a separate
1467 * volume special file. The file begins with the usual version stamp
1468 * information and is then followed by one row per vnode number. vnode 0
1469 * is used to hold the link count of the link table itself. That is because
1470 * the same link table is shared among all the volumes of the volume group
1471 * and is deleted only when the last volume of a volume group is deleted.
1473 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
1474 * link count from 0 through 7. Each colume represents a unique instance of
1475 * that vnode. Say we have a file shared between the RW and a RO and a
1476 * different version of the file (or a different uniquifer) for the BU volume.
1477 * Then one column would be holding the link count of 2 for the RW and RO
1478 * and a different column would hold the link count of 1 for the BU volume.
1479 * # ifdef AFS_NT40_ENV
1480 * The column used is determined for NT by the uniquifier tag applied to
1481 * generate a unique file name in the NTFS namespace. The file name is
1482 * of the form "V_<vno>.<tag>" . And the <tag> is also the column number
1483 * in the link table.
1485 * Note that we allow only 5 volumes per file, giving 15 bits used in the
1489 #define LINKTABLE_WIDTH 2
1490 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
1493 * compute namei link table file and bit offset from inode number.
1495 * @param[in] ino inode number
1496 * @param[out] offset link table file offset
1497 * @param[out] index bit offset within 2-byte record
1502 namei_GetLCOffsetAndIndexFromIno(Inode ino
, afs_foff_t
* offset
, int *index
)
1504 int toff
= (int)(ino
& NAMEI_VNODEMASK
);
1505 int tindex
= (int)((ino
>> NAMEI_TAGSHIFT
) & NAMEI_TAGMASK
);
1507 *offset
= (afs_foff_t
) ((toff
<< LINKTABLE_SHIFT
) + 8); /* * 2 + sizeof stamp */
1508 *index
= (tindex
<< 1) + tindex
;
1511 #ifdef AFS_PTHREAD_ENV
1512 /* XXX do static initializers work for WINNT/pthread? */
1513 pthread_mutex_t _namei_glc_lock
= PTHREAD_MUTEX_INITIALIZER
;
1514 #define NAMEI_GLC_LOCK opr_mutex_enter(&_namei_glc_lock)
1515 #define NAMEI_GLC_UNLOCK opr_mutex_exit(&_namei_glc_lock)
1516 #else /* !AFS_PTHREAD_ENV */
1517 #define NAMEI_GLC_LOCK
1518 #define NAMEI_GLC_UNLOCK
1519 #endif /* !AFS_PTHREAD_ENV */
1522 * get the link count of an inode.
1524 * @param[in] h namei link count table file handle
1525 * @param[in] ino inode number for which we are requesting a link count
1526 * @param[in] lockit if asserted, return with lock held on link table file
1527 * @param[in] fixup if asserted, write 1 to link count when read() returns
1529 * @param[in] nowrite return success on zero byte read or ZLC
1531 * @post if lockit asserted and lookup was successful, will return with write
1532 * lock on link table file descriptor
1534 * @return link count
1535 * @retval -1 namei link table i/o error
1540 namei_GetLinkCount(FdHandle_t
* h
, Inode ino
, int lockit
, int fixup
, int nowrite
)
1542 unsigned short row
= 0;
1547 /* there's no linktable yet. the salvager will create one later */
1548 if (h
->fd_fd
== INVALID_FD
&& fixup
)
1550 namei_GetLCOffsetAndIndexFromIno(ino
, &offset
, &index
);
1553 if (FDH_LOCKFILE(h
, offset
) != 0)
1557 rc
= FDH_PREAD(h
, (char*)&row
, sizeof(row
), offset
);
1559 goto bad_getLinkByte
;
1561 if ((rc
== 0 || !((row
>> index
) & NAMEI_TAGMASK
)) && fixup
&& nowrite
)
1563 if (rc
== 0 && fixup
) {
1565 * extend link table and write a link count of 1 for ino
1567 * in order to make MT-safe, truncation (extension really)
1568 * must happen under a mutex
1571 if (FDH_SIZE(h
) >= offset
+sizeof(row
)) {
1573 goto bad_getLinkByte
;
1575 if (FDH_TRUNC(h
, offset
+sizeof(row
))) {
1577 goto bad_getLinkByte
;
1580 rc
= FDH_PWRITE(h
, (char *)&row
, sizeof(row
), offset
);
1583 if (rc
!= sizeof(row
)) {
1584 goto bad_getLinkByte
;
1587 if (fixup
&& !((row
>> index
) & NAMEI_TAGMASK
)) {
1591 * in order to make this mt-safe, we need to do the read-modify-write
1592 * under a mutex. thus, we repeat the read inside the lock.
1595 rc
= FDH_PREAD(h
, (char *)&row
, sizeof(row
), offset
);
1596 if (rc
== sizeof(row
)) {
1598 rc
= FDH_PWRITE(h
, (char *)&row
, sizeof(row
), offset
);
1601 if (rc
!= sizeof(row
))
1602 goto bad_getLinkByte
;
1605 return (int)((row
>> index
) & NAMEI_TAGMASK
);
1609 FDH_UNLOCKFILE(h
, offset
);
1613 /* Return a free column index for this vnode. */
1615 GetFreeTag(IHandle_t
* ih
, int vno
)
1629 /* Only one manipulates at a time. */
1630 if (FDH_LOCKFILE(fdP
, offset
) != 0) {
1631 FDH_REALLYCLOSE(fdP
);
1635 offset
= (vno
<< LINKTABLE_SHIFT
) + 8; /* * 2 + sizeof stamp */
1637 nBytes
= FDH_PREAD(fdP
, (char *)&row
, sizeof(row
), offset
);
1638 if (nBytes
!= sizeof(row
)) {
1644 /* Now find a free column in this row and claim it. */
1645 for (col
= 0; col
< NAMEI_MAXVOLS
; col
++) {
1646 coldata
= 7 << (col
* 3);
1647 if ((row
& coldata
) == 0)
1650 if (col
>= NAMEI_MAXVOLS
) {
1655 coldata
= 1 << (col
* 3);
1658 if (FDH_PWRITE(fdP
, (char *)&row
, sizeof(row
), offset
) != sizeof(row
)) {
1661 (void)FDH_SYNC(fdP
);
1662 FDH_UNLOCKFILE(fdP
, offset
);
1667 FDH_UNLOCKFILE(fdP
, offset
);
1668 FDH_REALLYCLOSE(fdP
);
1674 /* namei_SetLinkCount
1675 * If locked is set, assume file is locked. Otherwise, lock file before
1676 * proceeding to modify it.
1679 namei_SetLinkCount(FdHandle_t
* fdP
, Inode ino
, int count
, int locked
)
1685 ssize_t nBytes
= -1;
1687 namei_GetLCOffsetAndIndexFromIno(ino
, &offset
, &index
);
1690 if (FDH_LOCKFILE(fdP
, offset
) != 0) {
1695 nBytes
= FDH_PREAD(fdP
, (char *)&row
, sizeof(row
), offset
);
1696 if (nBytes
!= sizeof(row
)) {
1698 errno
= OS_ERROR(EBADF
);
1699 goto bad_SetLinkCount
;
1704 bytesRead
= 7 << index
;
1706 row
&= (unsigned short)~bytesRead
;
1707 row
|= (unsigned short)count
;
1709 if (FDH_PWRITE(fdP
, (char *)&row
, sizeof(short), offset
) != sizeof(short)) {
1710 errno
= OS_ERROR(EBADF
);
1711 goto bad_SetLinkCount
;
1713 (void)FDH_SYNC(fdP
);
1719 FDH_UNLOCKFILE(fdP
, offset
);
1721 /* disallowed above 7, so... */
1726 namei_UnlockLinkCount(FdHandle_t
* fdP
, Inode ino
)
1731 namei_GetLCOffsetAndIndexFromIno(ino
, &offset
, &index
);
1733 FDH_UNLOCKFILE(fdP
, offset
);
1737 /* ListViceInodes - write inode data to a results file. */
1738 static int DecodeInode(char *dpath
, char *name
, struct ViceInodeInfo
*info
,
1740 static int DecodeVolumeName(char *name
, VolumeId
*vid
);
1741 static int namei_ListAFSSubDirs(IHandle_t
* dirIH
,
1742 int (*write_fun
) (FD_t
,
1743 struct ViceInodeInfo
*,
1744 char *, char *), FD_t fp
,
1745 int (*judgeFun
) (struct ViceInodeInfo
*,
1746 VolumeId vid
, void *),
1747 VolumeId singleVolumeNumber
, void *rock
);
1752 * Write the inode data to the results file.
1754 * Returns -2 on error, 0 on success.
1756 * This is written as a callback simply so that other listing routines
1757 * can use the same inode reading code.
1760 WriteInodeInfo(FD_t fp
, struct ViceInodeInfo
*info
, char *dir
, char *name
)
1763 n
= OS_WRITE(fp
, info
, sizeof(*info
));
1764 return (n
== sizeof(*info
)) ? 0 : -2;
1768 int mode_errors
; /* Number of errors found in mode bits on directories. */
1770 VerifyDirPerms(char *path
)
1772 struct afs_stat_st status
;
1774 if (afs_stat(path
, &status
) < 0) {
1775 Log("Unable to stat %s. Please manually verify mode bits for this"
1776 " directory\n", path
);
1778 if (((status
.st_mode
& 0777) != 0700) || (status
.st_uid
!= 0))
1784 * Fill the results file with the requested inode information.
1786 * This code optimizes single volume salvages by just looking at that one
1787 * volume's directory.
1789 * @param[in] devname device name string
1790 * @param[in] moutnedOn vice partition mount point
1791 * @param[in] resultFile result file in which to write inode
1792 * metadata. If NULL, write routine is not
1794 * @param[in] judgeInode filter function pointer. if not NULL, only
1795 * inodes for which this routine returns non-
1796 * zero will be written to the results file.
1797 * @param[in] singleVolumeNumber volume id filter
1798 * @param[out] forcep always set to 0 for namei impl
1799 * @param[in] forceR not used by namei impl
1800 * @param[in] wpath not used by namei impl
1801 * @param[in] rock opaque pointer passed to judgeInode
1803 * @return operation status
1805 * @retval -1 complete failure, salvage should terminate.
1806 * @retval -2 not enough space on partition, salvager has error message
1810 ListViceInodes(char *devname
, char *mountedOn
, FD_t inodeFile
,
1811 int (*judgeInode
) (struct ViceInodeInfo
* info
, VolumeId vid
, void *rock
),
1812 VolumeId singleVolumeNumber
, int *forcep
, int forceR
, char *wpath
,
1817 *forcep
= 0; /* no need to salvage until further notice */
1819 /* Verify protections on directories. */
1821 VerifyDirPerms(mountedOn
);
1824 namei_ListAFSFiles(mountedOn
, WriteInodeInfo
, inodeFile
, judgeInode
,
1825 singleVolumeNumber
, rock
);
1827 if (inodeFile
== INVALID_FD
)
1834 if (OS_SYNC(inodeFile
) == -1) {
1835 Log("Unable to successfully fsync inode file for %s\n", mountedOn
);
1840 * Paranoia: check that the file is really the right size
1842 if (OS_SIZE(inodeFile
) != ninodes
* sizeof(struct ViceInodeInfo
)) {
1843 Log("Wrong size (%d instead of %lu) in inode file for %s\n",
1844 (int) OS_SIZE(inodeFile
),
1845 (long unsigned int) ninodes
* sizeof(struct ViceInodeInfo
),
1854 * Collect all the matching AFS files on the drive.
1855 * If singleVolumeNumber is non-zero, just return files for that volume.
1857 * @param[in] dev vice partition path
1858 * @param[in] writeFun function pointer to a function which
1859 * writes inode information to FILE fp
1860 * @param[in] fp file stream where inode metadata is sent
1861 * @param[in] judgeFun filter function pointer. if not NULL,
1862 * only entries for which a non-zero value
1863 * is returned are written to fp
1864 * @param[in] singleVolumeNumber volume id filter. if nonzero, only
1865 * process files for that specific volume id
1866 * @param[in] rock opaque pointer passed into writeFun and
1869 * @return operation status
1871 * @retval >=0 number of matching files found
1874 namei_ListAFSFiles(char *dev
,
1875 int (*writeFun
) (FD_t
, struct ViceInodeInfo
*, char *,
1878 int (*judgeFun
) (struct ViceInodeInfo
*, VolumeId
, void *),
1879 VolumeId singleVolumeNumber
, void *rock
)
1886 #ifndef AFS_NT40_ENV
1892 static void FreeZLCList(void);
1895 memset((void *)&ih
, 0, sizeof(IHandle_t
));
1897 ih
.ih_dev
= nt_DriveToDev(dev
);
1899 ih
.ih_dev
= volutil_GetPartitionID(dev
);
1902 if (singleVolumeNumber
) {
1903 ih
.ih_vid
= singleVolumeNumber
;
1904 namei_HandleToVolDir(&name
, &ih
);
1906 namei_ListAFSSubDirs(&ih
, writeFun
, fp
, judgeFun
,
1907 singleVolumeNumber
, rock
);
1911 /* Find all volume data directories and descend through them. */
1912 namei_HandleToInodeDir(&name
, &ih
);
1914 dirp1
= opendir(name
.n_path
);
1917 while ((dp1
= readdir(dirp1
))) {
1919 /* Heirarchy is one level on Windows */
1920 if (!DecodeVolumeName(dp1
->d_name
, &ih
.ih_vid
)) {
1922 namei_ListAFSSubDirs(&ih
, writeFun
, fp
, judgeFun
,
1926 if (*dp1
->d_name
== '.')
1928 snprintf(path2
, sizeof(path2
), "%s" OS_DIRSEP
"%s", name
.n_path
,
1930 dirp2
= opendir(path2
);
1932 while ((dp2
= readdir(dirp2
))) {
1933 if (*dp2
->d_name
== '.')
1935 if (!DecodeVolumeName(dp2
->d_name
, &ih
.ih_vid
)) {
1937 namei_ListAFSSubDirs(&ih
, writeFun
, fp
, judgeFun
,
1954 static void AddToZLCDeleteList(char dir
, char *name
);
1955 static void DeleteZLCFiles(char *path
);
1959 * examine a namei volume special file.
1961 * @param[in] path1 volume special directory path
1962 * @param[in] dname directory entry name
1963 * @param[in] myIH inode handle to volume directory
1964 * @param[out] linkHandle namei link count fd handle. if
1965 * the inode in question is the link
1966 * table, then the FdHandle is populated
1967 * @param[in] writeFun metadata write function pointer
1968 * @param[in] fp file pointer where inode metadata
1969 * is written by (*writeFun)()
1970 * @param[in] judgeFun inode filter function pointer. if
1971 * not NULL, only inodes for which this
1972 * function returns non-zero are recorded
1973 * into fp by writeFun
1974 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1975 * inodes associated with this volume id
1976 * are recorded by writeFun
1977 * @param[in] rock opaque pointer passed to writeFun and
1980 * @return operation status
1981 * @retval 1 count this inode
1982 * @retval 0 don't count this inode
1983 * @retval -1 failure
1988 _namei_examine_special(char * path1
,
1991 FdHandle_t
* linkHandle
,
1992 int (*writeFun
) (FD_t
, struct ViceInodeInfo
*, char *,
1995 int (*judgeFun
) (struct ViceInodeInfo
*, VolumeId
, void *),
1996 VolumeId singleVolumeNumber
,
2000 struct ViceInodeInfo info
;
2002 if (DecodeInode(path1
, dname
, &info
, myIH
) < 0) {
2007 if (info
.u
.param
[2] != VI_LINKTABLE
) {
2009 } else if (info
.u
.param
[0] != myIH
->ih_vid
) {
2010 /* VGID encoded in linktable filename and/or OGM data isn't
2011 * consistent with VGID encoded in namei path */
2012 Log("namei_ListAFSSubDirs: warning: inconsistent linktable "
2013 "filename \"%s" OS_DIRSEP
"%s\"; salvager will delete it "
2014 "(dir_vgid=%" AFS_VOLID_FMT
", inode_vgid=%" AFS_VOLID_FMT
")\n",
2015 path1
, dname
, afs_printable_VolumeId_lu(myIH
->ih_vid
),
2016 afs_printable_VolumeId_lu(info
.u
.param
[0]));
2017 /* We need to set the linkCount to _something_, so linkCount
2018 * doesn't just contain stack garbage. Set it to 0, so in case
2019 * the salvager or whatever our caller is does try to process
2020 * this like a normal file, we won't try to INC or DEC it. */
2024 /* Open this handle */
2025 snprintf(path2
, sizeof(path2
),
2026 "%s" OS_DIRSEP
"%s", path1
, dname
);
2027 linkHandle
->fd_fd
= OS_OPEN(path2
, Testing
? O_RDONLY
: O_RDWR
, 0666);
2029 namei_GetLinkCount(linkHandle
, (Inode
) 0, 1, 1, Testing
);
2033 (*judgeFun
) (&info
, singleVolumeNumber
, rock
)) {
2034 ret
= (*writeFun
) (fp
, &info
, path1
, dname
);
2036 Log("_namei_examine_special: writeFun returned %d\n", ret
);
2048 * examine a namei file.
2050 * @param[in] path3 volume special directory path
2051 * @param[in] dname directory entry name
2052 * @param[in] myIH inode handle to volume directory
2053 * @param[in] linkHandle namei link count fd handle.
2054 * @param[in] writeFun metadata write function pointer
2055 * @param[in] fp file pointer where inode metadata
2056 * is written by (*writeFun)()
2057 * @param[in] judgeFun inode filter function pointer. if
2058 * not NULL, only inodes for which this
2059 * function returns non-zero are recorded
2060 * into fp by writeFun
2061 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
2062 * inodes associated with this volume id
2063 * are recorded by writeFun
2064 * @param[in] rock opaque pointer passed to writeFun and
2067 * @return operation status
2068 * @retval 1 count this inode
2069 * @retval 0 don't count this inode
2070 * @retval -1 failure
2071 * @retval -2 request ZLC delete
2076 _namei_examine_reg(char * path3
,
2079 FdHandle_t
* linkHandle
,
2080 int (*writeFun
) (FD_t
, struct ViceInodeInfo
*, char *,
2083 int (*judgeFun
) (struct ViceInodeInfo
*, VolumeId
, void *),
2084 VolumeId singleVolumeNumber
,
2088 struct ViceInodeInfo info
;
2090 int dirl
; /* Windows-only (one level hash dir) */
2093 if (DecodeInode(path3
, dname
, &info
, myIH
) < 0) {
2098 namei_GetLinkCount(linkHandle
,
2099 info
.inodeNumber
, 1, 1, Testing
);
2100 if (info
.linkCount
== 0) {
2102 Log("Found 0 link count file %s" OS_DIRSEP
"%s, deleting it.\n", path3
, dname
);
2103 #ifdef AFS_SALSRV_ENV
2104 /* defer -- the AddToZLCDeleteList() interface is not MT-safe */
2106 #else /* !AFS_SALSRV_ENV */
2107 dirl
= path3
[strlen(path3
)-1];
2108 AddToZLCDeleteList((char)dirl
, dname
);
2109 #endif /* !AFS_SALSRV_ENV */
2110 #else /* !DELETE_ZLC */
2111 Log("Found 0 link count file %s" OS_DIRSEP
"%s.\n", path3
,
2118 (*judgeFun
) (&info
, singleVolumeNumber
, rock
)) {
2119 ret
= (*writeFun
) (fp
, &info
, path3
, dname
);
2121 Log("_namei_examine_reg: writeFun returned %d\n", ret
);
2133 * listsubdirs work queue node.
2135 struct listsubdirs_work_node
{
2136 #ifdef AFS_SALSRV_ENV
2137 int *error
; /**< *error set if an error was
2138 * encountered in any listsubdirs
2142 IHandle_t
* IH
; /**< volume directory handle */
2143 FdHandle_t
*linkHandle
; /**< namei link count fd handle. when
2144 * examinining the link table special
2145 * inode, this will be pointed at the
2148 FD_t fp
; /**< file pointer for writeFun */
2150 /** function which will write inode metadata to fp */
2151 int (*writeFun
) (FD_t
, struct ViceInodeInfo
*, char *, char *);
2153 /** inode filter function */
2154 int (*judgeFun
) (struct ViceInodeInfo
*, VolumeId
, void *);
2155 VolumeId singleVolumeNumber
; /**< volume id filter */
2156 void * rock
; /**< pointer passed to writeFun and judgeFun */
2157 int code
; /**< return code from examine function */
2158 int special
; /**< asserted when this is a volume
2163 * simple wrapper around _namei_examine_special and _namei_examine_reg.
2165 * @param[in] work the struct listsubdirs_work_node for the associated
2166 * "list subdirs" job
2167 * @param[in] dir the directory to examine
2168 * @param[in] filename the filename in 'dir' to examine
2170 * @return operation status
2171 * @retval 1 count this inode
2172 * @retval 0 don't count this inode
2173 * @retval -1 failure
2176 _namei_examine_file(const struct listsubdirs_work_node
*work
, char *dir
,
2179 if (work
->special
) {
2180 return _namei_examine_special(dir
, filename
, work
->IH
,
2181 work
->linkHandle
, work
->writeFun
, work
->fp
,
2182 work
->judgeFun
, work
->singleVolumeNumber
,
2185 return _namei_examine_reg(dir
, filename
, work
->IH
,
2186 work
->linkHandle
, work
->writeFun
, work
->fp
,
2187 work
->judgeFun
, work
->singleVolumeNumber
,
2193 #ifdef AFS_SALSRV_ENV
2194 /** @addtogroup afs_vol_salsrv_pario */
2198 * arguments for the _namei_examine_file_cbk callback function.
2200 struct listsubdirs_args
{
2201 const struct listsubdirs_work_node
*work
; /**< arguments that are the same
2202 * for all invocations of
2203 * _namei_examine_file_cbk, in
2205 int *result
; /**< where we can store the return code of _namei_examine_file */
2207 char dir
[512]; /**< directory to examine */
2208 char filename
[256]; /**< filename in 'dir' to examine */
2212 * a node in the list of results of listsubdir jobs.
2214 struct listsubdirs_result
{
2216 int inodes
; /**< return value from _namei_examine_file */
2220 * clean up a list of 'struct listsubdirs_result's and interpret the results.
2222 * @param[in] resultlist a list of 'struct listsubdirs_result's
2224 * @return number of inodes found
2228 _namei_listsubdirs_cleanup_results(struct rx_queue
*resultlist
)
2230 struct listsubdirs_result
*res
, *nres
;
2233 for(queue_Scan(resultlist
, res
, nres
, listsubdirs_result
)) {
2235 /* noop, retain erroneous error code */
2236 } else if (res
->inodes
< 0) {
2251 * wait for the spawned listsubdirs jobs to finish, and return how many inodes
2254 * @param[in] queue queue to wait to finish
2255 * @param[in] resultlist list of 'struct listsubdirs_result's that the queued
2256 * jobs are storing their results in
2258 * @return number of inodes found
2262 _namei_listsubdirs_wait(struct afs_work_queue
*queue
, struct rx_queue
*resultlist
)
2266 code
= afs_wq_wait_all(queue
);
2271 return _namei_listsubdirs_cleanup_results(resultlist
);
2275 * work queue entry point for examining namei files.
2277 * @param[in] queue pointer to struct Vwork_queue
2278 * @param[in] node pointer to struct Vwork_queue_node
2279 * @param[in] queue_rock opaque pointer to struct salsrv_pool_state
2280 * @param[in] node_rock opaque pointer to struct listsubdirs_work_node
2281 * @param[in] caller_rock opaque pointer to struct salsrv_worker_thread_state
2283 * @return operation status
2285 * @see Vwork_queue_callback_func_t
2290 _namei_examine_file_cbk(struct afs_work_queue
*queue
,
2291 struct afs_work_queue_node
*node
,
2297 struct listsubdirs_args
*args
= node_rock
;
2298 const struct listsubdirs_work_node
* work
= args
->work
;
2299 char *dir
= args
->dir
;
2300 char *filename
= args
->filename
;
2302 code
= _namei_examine_file(work
, dir
, filename
);
2304 *(args
->result
) = code
;
2308 /* we've errored, so no point in letting more jobs continue */
2309 afs_wq_shutdown(queue
);
2315 static pthread_once_t wq_once
= PTHREAD_ONCE_INIT
;
2316 static pthread_key_t wq_key
;
2319 * create the wq_key key for storing a work queue.
2322 _namei_wq_keycreate(void)
2324 opr_Verify(pthread_key_create(&wq_key
, NULL
) == 0);
2328 * set the work queue for this thread to use for backgrounding namei ops.
2330 * The work queue will be used in ListAFSSubdirs (called indirectly by
2331 * ListViceInodes) to examine files in parallel.
2333 * @param[in] wq the work queue to use
2336 namei_SetWorkQueue(struct afs_work_queue
*wq
)
2338 opr_Verify(pthread_once(&wq_once
, _namei_wq_keycreate
) == 0);
2340 opr_Verify(pthread_setspecific(wq_key
, wq
) == 0);
2344 * enqueue an examine file work unit.
2346 * @param[in] work the _namei_examine_file arguments that are common to
2347 * all callers within the same ListAFSFiles operation
2348 * @param[in] dir the specific directory to look at (string will be
2349 * copied; can be stack/temporary memory)
2350 * @param[in] filename the filename to look at (string will be copied; can be
2351 * stack/temporary memory)
2352 * @param[in] wq work queue to enqueue this work unit to
2353 * @param[in] resultlist the list to append the 'struct listsubdirs_result' to
2354 * for the enqueued work unit
2356 * @return operation status
2358 * @retval -1 fatal error
2360 * @note errors MUST be indicated by a -1 error code and nothing else, to be
2361 * compatible with _namei_examine_reg and _namei_examine_special
2366 _namei_examine_file_spawn(const struct listsubdirs_work_node
*work
,
2367 const char *dir
, const char *filename
,
2368 struct afs_work_queue
*wq
,
2369 struct rx_queue
*resultlist
)
2372 struct listsubdirs_args
*args
= NULL
;
2373 struct listsubdirs_result
*result
= NULL
;
2374 struct afs_work_queue_node
*node
= NULL
;
2375 struct afs_work_queue_add_opts opts
;
2377 args
= malloc(sizeof(*args
));
2383 result
= malloc(sizeof(*result
));
2384 if (result
== NULL
) {
2389 code
= afs_wq_node_alloc(&node
);
2394 code
= afs_wq_node_set_detached(node
);
2401 args
->result
= &result
->inodes
;
2402 strlcpy(args
->dir
, dir
, sizeof(args
->dir
));
2403 strlcpy(args
->filename
, filename
, sizeof(args
->filename
));
2405 code
= afs_wq_node_set_callback(node
,
2406 &_namei_examine_file_cbk
,
2414 afs_wq_add_opts_init(&opts
);
2417 code
= afs_wq_add(wq
, node
, &opts
);
2424 queue_Append(resultlist
, result
);
2429 afs_wq_node_put(node
);
2445 #else /* !AFS_SALSRV_ENV */
2446 # define _namei_examine_file_spawn(work, dir, file, wq, resultlist) \
2447 _namei_examine_file(work, dir, file)
2448 #endif /* !AFS_SALSRV_ENV */
2451 * traverse and check inodes.
2453 * @param[in] dirIH volume group directory handle
2454 * @param[in] writeFun function pointer which will write inode
2455 * metadata to FILE stream fp
2456 * @param[in] fp file stream where inode metadata gets
2458 * @param[in] judgeFun inode filter function. if not NULL, only
2459 * inodes for which the filter returns non-zero
2460 * will be written out by writeFun
2461 * @param[in] singleVolumeNumber volume id filter. only inodes matching this
2462 * filter are written out by writeFun
2463 * @param[in] rock opaque pointer passed to judgeFun and writeFun
2465 * @return operation status
2467 * @retval >=0 number of matching inodes found
2469 * @todo the salsrv implementation may consume a lot of
2470 * memory for a large volume. at some point we should
2471 * probably write a background thread to asynchronously
2472 * clean up the resultlist nodes to reduce memory footprint
2477 namei_ListAFSSubDirs(IHandle_t
* dirIH
,
2478 int (*writeFun
) (FD_t
, struct ViceInodeInfo
*, char *,
2481 int (*judgeFun
) (struct ViceInodeInfo
*, VolumeId
, void *),
2482 VolumeId singleVolumeNumber
, void *rock
)
2484 int code
= 0, ret
= 0;
2485 IHandle_t myIH
= *dirIH
;
2487 char path1
[512], path3
[512];
2489 #ifndef AFS_NT40_ENV
2494 struct dirent
*dp1
, *dp3
;
2495 FdHandle_t linkHandle
;
2497 struct listsubdirs_work_node work
;
2498 #ifdef AFS_SALSRV_ENV
2500 struct afs_work_queue
*wq
;
2502 struct rx_queue resultlist
;
2505 namei_HandleToVolDir(&name
, &myIH
);
2506 strlcpy(path1
, name
.n_path
, sizeof(path1
));
2508 /* Do the directory containing the special files first to pick up link
2511 (void)strcat(path1
, OS_DIRSEP
);
2512 (void)strcat(path1
, NAMEI_SPECDIR
);
2514 linkHandle
.fd_fd
= INVALID_FD
;
2515 #ifdef AFS_SALSRV_ENV
2516 opr_Verify(pthread_once(&wq_once
, _namei_wq_keycreate
) == 0);
2518 wq
= pthread_getspecific(wq_key
);
2524 queue_Init(&resultlist
);
2527 memset(&work
, 0, sizeof(work
));
2528 work
.linkHandle
= &linkHandle
;
2531 work
.writeFun
= writeFun
;
2532 work
.judgeFun
= judgeFun
;
2533 work
.singleVolumeNumber
= singleVolumeNumber
;
2536 #ifdef AFS_SALSRV_ENV
2537 work
.error
= &error
;
2540 dirp1
= opendir(path1
);
2542 while ((dp1
= readdir(dirp1
))) {
2543 if (*dp1
->d_name
== '.')
2546 #ifdef AFS_SALSRV_ENV
2552 #endif /* AFS_SALSRV_ENV */
2554 code
= _namei_examine_file_spawn(&work
, path1
, dp1
->d_name
, wq
, &resultlist
);
2564 /* count this inode */
2565 #ifndef AFS_SALSRV_ENV
2574 #ifdef AFS_SALSRV_ENV
2575 /* effectively a barrier */
2576 code
= _namei_listsubdirs_wait(wq
, &resultlist
);
2577 if (code
< 0 || error
) {
2585 if (linkHandle
.fd_fd
== INVALID_FD
) {
2586 Log("namei_ListAFSSubDirs: warning: VG %" AFS_VOLID_FMT
" does not have a link table; "
2587 "salvager will recreate it.\n", afs_printable_VolumeId_lu(dirIH
->ih_vid
));
2590 /* Now run through all the other subdirs */
2591 namei_HandleToVolDir(&name
, &myIH
);
2592 strlcpy(path1
, name
.n_path
, sizeof(path1
));
2596 dirp1
= opendir(path1
);
2598 while ((dp1
= readdir(dirp1
))) {
2599 #ifndef AFS_NT40_ENV
2600 if (*dp1
->d_name
== '.')
2603 if (!strcmp(dp1
->d_name
, NAMEI_SPECDIR
))
2606 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2607 /* Now we've got a next level subdir. */
2608 snprintf(path2
, sizeof(path2
), "%s" OS_DIRSEP
"%s",
2609 path1
, dp1
->d_name
);
2610 dirp2
= opendir(path2
);
2612 while ((dp2
= readdir(dirp2
))) {
2613 if (*dp2
->d_name
== '.')
2616 /* Now we've got to the actual data */
2617 snprintf(path3
, sizeof(path3
), "%s" OS_DIRSEP
"%s",
2618 path2
, dp2
->d_name
);
2620 /* Now we've got to the actual data */
2621 snprintf(path3
, sizeof(path3
), "%s" OS_DIRSEP
"%s",
2622 path1
, dp1
->d_name
);
2624 dirp3
= opendir(path3
);
2626 while ((dp3
= readdir(dirp3
))) {
2627 #ifndef AFS_NT40_ENV
2628 if (*dp3
->d_name
== '.')
2632 #ifdef AFS_SALSRV_ENV
2635 #ifndef AFS_NT40_ENV
2642 #endif /* AFS_SALSRV_ENV */
2644 code
= _namei_examine_file_spawn(&work
, path3
,
2651 #ifndef AFS_NT40_ENV
2659 #ifndef AFS_SALSRV_ENV
2667 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2676 #ifdef AFS_SALSRV_ENV
2677 /* effectively a barrier */
2678 code
= _namei_listsubdirs_wait(wq
, &resultlist
);
2679 if (code
< 0 || error
) {
2689 /* Then why does this directory exist? Blow it away. */
2690 namei_HandleToVolDir(&name
, dirIH
);
2691 namei_RemoveDataDirectories(&name
);
2695 #ifdef AFS_SALSRV_ENV
2697 afs_wq_wait_all(wq
);
2699 _namei_listsubdirs_cleanup_results(&resultlist
);
2701 if (linkHandle
.fd_fd
!= INVALID_FD
)
2702 OS_CLOSE(linkHandle
.fd_fd
);
2714 DecodeVolumeName(char *name
, VolumeId
*vid
)
2716 /* Name begins with "Vol_" and ends with .data. See nt_HandleToVolDir() */
2723 if (strncmp(name
, "Vol_", 4))
2725 if (strcmp(name
+ len
- 5, ".data"))
2728 stmp
[len
- 5] = '\0';
2729 *vid
= base32_to_int(stmp
+ 4);
2734 DecodeVolumeName(char *name
, VolumeId
*vid
)
2736 if (strlen(name
) < 1)
2738 *vid
= (unsigned int)flipbase64_to_int64(name
);
2746 * Get the inode number from the name.
2751 DecodeInode(char *dpath
, char *name
, struct ViceInodeInfo
*info
,
2756 WIN32_FIND_DATA data
;
2760 FdHandle_t linkHandle
;
2762 VolumeId volid
= myIH
->ih_vid
;
2764 snprintf(fpath
, sizeof(fpath
), "%s" OS_DIRSEP
"%s", dpath
, name
);
2766 dirH
= FindFirstFileEx(fpath
, FindExInfoStandard
, &data
,
2767 FindExSearchNameMatch
, NULL
,
2768 FIND_FIRST_EX_CASE_SENSITIVE
);
2769 if (dirH
== INVALID_HANDLE_VALUE
)
2772 (void)strcpy(stmp
, name
);
2773 s
= strrchr(stmp
, '_');
2777 t
= strrchr(s
, '.');
2782 vno
= base32_to_int(s
); /* type for special files */
2783 tag
= base32_to_int(t
+1);
2784 info
->inodeNumber
= ((Inode
) tag
) << NAMEI_TAGSHIFT
;
2785 info
->inodeNumber
|= vno
;
2786 info
->byteCount
= data
.nFileSizeLow
;
2788 dirl
= dpath
[strlen(dpath
)-1];
2789 if (dirl
== NAMEI_SPECDIRC
) { /* Special inode. */
2790 info
->inodeNumber
|= NAMEI_INODESPECIAL
;
2791 info
->u
.param
[0] = data
.ftCreationTime
.dwHighDateTime
;
2792 info
->u
.param
[1] = data
.ftCreationTime
.dwLowDateTime
;
2793 info
->u
.param
[2] = vno
; /* type */
2794 info
->u
.param
[3] = volid
;
2795 if (vno
!= VI_LINKTABLE
)
2796 info
->linkCount
= 1;
2798 /* Open this handle */
2800 (void)sprintf(lpath
, "%s" OS_DIRSEP
"%s", fpath
, data
.cFileName
);
2801 linkHandle
.fd_fd
= OS_OPEN(lpath
, O_RDONLY
, 0666);
2803 namei_GetLinkCount(&linkHandle
, (Inode
) 0, 0, 0, 0);
2807 namei_GetLinkCount(&linkHandle
, info
->inodeNumber
, 0, 0, 0);
2808 if (info
->linkCount
== 0) {
2810 Log("Found 0 link count file %s" OS_DIRSEP
"%s, deleting it.\n",
2811 fpath
, data
.cFileName
);
2812 AddToZLCDeleteList(dirl
, data
.cFileName
);
2814 Log("Found 0 link count file %s" OS_DIRSEP
"%s.\n", path
,
2818 info
->u
.param
[2] = data
.ftCreationTime
.dwHighDateTime
;
2819 info
->u
.param
[3] = data
.ftCreationTime
.dwLowDateTime
;
2820 info
->u
.param
[1] = vno
;
2821 info
->u
.param
[0] = volid
;
2828 DecodeInode(char *dpath
, char *name
, struct ViceInodeInfo
*info
,
2832 struct afs_stat_st status
;
2833 struct afs_stat_st checkstatus
;
2835 lb64_string_t check
;
2836 VolumeId volid
= myIH
->ih_vid
;
2840 snprintf(fpath
, sizeof(fpath
), "%s" OS_DIRSEP
"%s", dpath
, name
);
2842 if (afs_stat(fpath
, &status
) < 0) {
2846 info
->byteCount
= status
.st_size
;
2847 info
->inodeNumber
= (Inode
) flipbase64_to_int64(name
);
2849 int64_to_flipbase64(check
, info
->inodeNumber
);
2850 if (strcmp(name
, check
))
2853 /* Check if the _full_ path is correct, to ensure we can actually open this
2854 * file later. Otherwise, the salvager can choke. */
2855 memset(&tmpih
, 0, sizeof(tmpih
));
2856 tmpih
.ih_dev
= myIH
->ih_dev
;
2857 tmpih
.ih_vid
= myIH
->ih_vid
;
2858 tmpih
.ih_ino
= info
->inodeNumber
;
2859 namei_HandleToName(&nameiname
, &tmpih
);
2860 if ((afs_stat(nameiname
.n_path
, &checkstatus
) < 0) ||
2861 checkstatus
.st_ino
!= status
.st_ino
||
2862 checkstatus
.st_size
!= status
.st_size
) {
2864 /* log something for this case, since this means the filename looks
2865 * like a valid inode, but it's just in the wrong place. That's pretty
2870 Log(" Seemingly-misplaced files have been found, which I am\n");
2871 Log(" ignoring for now. If you cannot salvage the relevant volume,\n");
2872 Log(" you may try manually moving them to their correct location.\n");
2873 Log(" If the relevant volume seems fine, and these files do not\n");
2874 Log(" appear to contain important data, you can probably manually\n");
2875 Log(" delete them, or leave them alone. Contact your local OpenAFS\n");
2876 Log(" expert if you are unsure.\n");
2878 Log("Ignoring misplaced file in volume group %u: %s (should be %s)\n",
2879 (unsigned)myIH
->ih_vid
, fpath
, nameiname
.n_path
);
2883 if ((info
->inodeNumber
& NAMEI_INODESPECIAL
) == NAMEI_INODESPECIAL
) {
2884 parm
= ((info
->inodeNumber
>> NAMEI_UNIQSHIFT
) & NAMEI_UNIQMASK
);
2885 tag
= (int)((info
->inodeNumber
>> NAMEI_TAGSHIFT
) & NAMEI_TAGMASK
);
2887 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
2888 info
->u
.param
[0] = parm
;
2889 info
->u
.param
[1] = -1;
2890 info
->u
.param
[2] = tag
;
2891 info
->u
.param
[3] = volid
;
2893 GetOGMFromStat(&status
, &parm
, &tag
);
2894 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
2895 info
->u
.param
[0] = volid
;
2896 info
->u
.param
[1] = (int)(info
->inodeNumber
& NAMEI_VNODEMASK
);
2897 info
->u
.param
[2] = (int)((info
->inodeNumber
>> NAMEI_UNIQSHIFT
)
2898 & (Inode
) NAMEI_UNIQMASK
);
2899 info
->u
.param
[3] = parm
;
2906 * Convert the VolumeInfo file from RO to RW
2907 * this routine is called by namei_convertROtoRWvolume()
2910 #ifdef FSSYNC_BUILD_CLIENT
2912 convertVolumeInfo(FD_t fdr
, FD_t fdw
, VolumeId vid
)
2914 struct VolumeDiskData vd
;
2917 if (OS_READ(fdr
, &vd
, sizeof(struct VolumeDiskData
)) !=
2918 sizeof(struct VolumeDiskData
)) {
2919 Log("1 convertVolumeInfo: read failed for %" AFS_VOLID_FMT
" with code %d\n",
2920 afs_printable_VolumeId_lu(vid
),
2924 vd
.restoredFromId
= vd
.id
; /* remember the RO volume here */
2926 vd
.id
= vd
.parentId
;
2930 vd
.uniquifier
+= 5000; /* just in case there are still file copies from
2931 * the old RW volume around */
2933 /* For ROs, the copyDate contains the time that the RO volume was actually
2934 * created, and the creationDate just contains the last time the RO was
2935 * copied from the RW data. So, make the new RW creationDate more accurate
2936 * by setting it to copyDate, if copyDate is older. Since, we know the
2937 * volume is at least as old as copyDate. */
2938 if (vd
.copyDate
< vd
.creationDate
) {
2939 vd
.creationDate
= vd
.copyDate
;
2941 /* If copyDate is newer, just make copyDate and creationDate the same,
2942 * for consistency with other RWs */
2943 vd
.copyDate
= vd
.creationDate
;
2946 p
= strrchr(vd
.name
, '.');
2947 if (p
&& !strcmp(p
, ".readonly")) {
2950 if (OS_WRITE(fdw
, &vd
, sizeof(struct VolumeDiskData
)) !=
2951 sizeof(struct VolumeDiskData
)) {
2952 Log("1 convertVolumeInfo: write failed for %lu with code %d\n",
2953 afs_printable_uint32_lu(vid
),
2962 * Convert a RO-volume into a RW-volume
2964 * This function allows to recover very fast from the loss of a partition
2965 * from RO-copies if all RO-Copies exist on another partition.
2966 * Then these RO-volumes can be made to the new RW-volumes.
2967 * Backup of RW-volumes then consists in "vos release".
2969 * We must make sure in this partition exists only the RO-volume which
2970 * is typical for remote replicas.
2972 * Then the linktable is already ok,
2973 * the vnode files need to be renamed
2974 * the volinfo file needs to be replaced by another one with
2975 * slightly different contents and new name.
2976 * The volume header file of the RO-volume in the /vicep<x> directory
2977 * is destroyed by this call. A new header file for the RW-volume must
2978 * be created after return from this routine.
2982 namei_ConvertROtoRWvolume(char *pname
, VolumeId volumeId
)
2985 #ifdef FSSYNC_BUILD_CLIENT
2987 char dir_name
[512], oldpath
[512], newpath
[512];
3002 struct DiskPartition64
*partP
;
3003 struct ViceInodeInfo info
;
3004 struct VolumeDiskHeader h
;
3005 char *rwpart
, *rwname
;
3007 # ifdef AFS_DEMAND_ATTACH_FS
3009 # endif /* AFS_DEMAND_ATTACH_FS */
3011 for (partP
= DiskPartitionList
; partP
&& strcmp(partP
->name
, pname
);
3012 partP
= partP
->next
);
3014 Log("1 namei_ConvertROtoRWvolume: Couldn't find DiskPartition for %s\n", pname
);
3019 # ifdef AFS_DEMAND_ATTACH_FS
3020 locktype
= VVolLockType(V_VOLUPD
, 1);
3021 code
= VLockVolumeByIdNB(volumeId
, partP
, locktype
);
3027 # endif /* AFS_DEMAND_ATTACH_FS */
3029 if (VReadVolumeDiskHeader(volumeId
, partP
, &h
)) {
3030 Log("1 namei_ConvertROtoRWvolume: Couldn't read header for RO-volume %lu.\n",
3031 afs_printable_uint32_lu(volumeId
));
3036 /* check for existing RW on any partition on this server; */
3037 /* if found, return EXDEV - invalid cross-device link */
3039 VGetVolumePath(&ec
, h
.parent
, &rwpart
, &rwname
);
3041 Log("1 namei_ConvertROtoRWvolume: RW volume %lu already exists on server partition %s.\n",
3042 afs_printable_uint32_lu(h
.parent
), rwpart
);
3049 FSYNC_VolOp(volumeId
, pname
, FSYNC_VOL_BREAKCBKS
, 0, NULL
);
3051 ino
= namei_MakeSpecIno(h
.parent
, VI_LINKTABLE
);
3052 IH_INIT(ih
, partP
->device
, h
.parent
, ino
);
3054 namei_HandleToName(&n
, ih
);
3055 strlcpy(dir_name
, n
.n_path
, sizeof(dir_name
));
3056 p
= strrchr(dir_name
, OS_DIRSEPC
);
3058 dirp
= opendir(dir_name
);
3060 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name
);
3065 while ((dp
= readdir(dirp
))) {
3066 /* struct ViceInodeInfo info; */
3067 #ifndef AFS_NT40_ENV
3068 if (*dp
->d_name
== '.')
3071 if (DecodeInode(dir_name
, dp
->d_name
, &info
, ih
) < 0) {
3072 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s" OS_DIRSEP
"%s\n",
3073 dir_name
, dp
->d_name
);
3078 if (info
.u
.param
[1] != -1) {
3079 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s" OS_DIRSEP
"%s\n", dir_name
, dp
->d_name
);
3084 if (info
.u
.param
[0] != volumeId
) {
3085 if (info
.u
.param
[0] == ih
->ih_vid
) {
3086 if (info
.u
.param
[2] == VI_LINKTABLE
) { /* link table */
3091 Log("1 namei_ConvertROtoRWvolume: found special file %s" OS_DIRSEP
"%s"
3092 " for volume %lu\n", dir_name
, dp
->d_name
,
3093 afs_printable_uint32_lu(info
.u
.param
[0]));
3098 if (info
.u
.param
[2] == VI_VOLINFO
) { /* volume info file */
3099 strlcpy(infoName
, dp
->d_name
, sizeof(infoName
));
3101 } else if (info
.u
.param
[2] == VI_SMALLINDEX
) { /* small vnodes file */
3102 strlcpy(smallName
, dp
->d_name
, sizeof(smallName
));
3104 } else if (info
.u
.param
[2] == VI_LARGEINDEX
) { /* large vnodes file */
3105 strlcpy(largeName
, dp
->d_name
, sizeof(largeName
));
3109 Log("1 namei_ConvertROtoRWvolume: unknown type %u of special file found : %s" OS_DIRSEP
"%s\n", info
.u
.param
[2], dir_name
, dp
->d_name
);
3116 if (!infoSeen
|| !smallSeen
|| !largeSeen
|| !linkSeen
) {
3117 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name
);
3123 * If we come here then there was only a RO-volume and we can safely
3127 memset(&t_ih
, 0, sizeof(t_ih
));
3128 t_ih
.ih_dev
= ih
->ih_dev
;
3129 t_ih
.ih_vid
= ih
->ih_vid
;
3131 snprintf(oldpath
, sizeof oldpath
, "%s" OS_DIRSEP
"%s", dir_name
,
3133 fd
= OS_OPEN(oldpath
, O_RDWR
, 0);
3134 if (fd
== INVALID_FD
) {
3135 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
3140 t_ih
.ih_ino
= namei_MakeSpecIno(ih
->ih_vid
, VI_VOLINFO
);
3141 namei_HandleToName(&n
, &t_ih
);
3142 fd2
= OS_OPEN(n
.n_path
, O_CREAT
| O_EXCL
| O_RDWR
, 0);
3143 if (fd2
== INVALID_FD
) {
3144 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n
.n_path
);
3149 code
= convertVolumeInfo(fd
, fd2
, ih
->ih_vid
);
3153 OS_UNLINK(n
.n_path
);
3157 SetOGM(fd2
, ih
->ih_vid
, 1);
3160 t_ih
.ih_ino
= namei_MakeSpecIno(ih
->ih_vid
, VI_SMALLINDEX
);
3161 namei_HandleToName(&n
, &t_ih
);
3162 snprintf(newpath
, sizeof newpath
, "%s" OS_DIRSEP
"%s", dir_name
,
3164 fd
= OS_OPEN(newpath
, O_RDWR
, 0);
3165 if (fd
== INVALID_FD
) {
3166 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath
);
3170 SetOGM(fd
, ih
->ih_vid
, 2);
3173 MoveFileEx(n
.n_path
, newpath
, MOVEFILE_WRITE_THROUGH
);
3175 if (link(newpath
, n
.n_path
)) {
3176 Log("1 namei_ConvertROtoRWvolume: could not move SmallIndex file: %s\n", n
.n_path
);
3183 t_ih
.ih_ino
= namei_MakeSpecIno(ih
->ih_vid
, VI_LARGEINDEX
);
3184 namei_HandleToName(&n
, &t_ih
);
3185 snprintf(newpath
, sizeof newpath
, "%s" OS_DIRSEP
"%s", dir_name
,
3187 fd
= OS_OPEN(newpath
, O_RDWR
, 0);
3188 if (fd
== INVALID_FD
) {
3189 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath
);
3193 SetOGM(fd
, ih
->ih_vid
, 3);
3196 MoveFileEx(n
.n_path
, newpath
, MOVEFILE_WRITE_THROUGH
);
3198 if (link(newpath
, n
.n_path
)) {
3199 Log("1 namei_ConvertROtoRWvolume: could not move LargeIndex file: %s\n", n
.n_path
);
3209 h
.volumeInfo_hi
= h
.id
;
3210 h
.smallVnodeIndex_hi
= h
.id
;
3211 h
.largeVnodeIndex_hi
= h
.id
;
3212 h
.linkTable_hi
= h
.id
;
3214 if (VCreateVolumeDiskHeader(&h
, partP
)) {
3215 Log("1 namei_ConvertROtoRWvolume: Couldn't write header for RW-volume %lu\n",
3216 afs_printable_uint32_lu(h
.id
));
3221 if (VDestroyVolumeDiskHeader(partP
, volumeId
, h
.parent
)) {
3222 Log("1 namei_ConvertROtoRWvolume: Couldn't unlink header for RO-volume %lu\n",
3223 afs_printable_uint32_lu(volumeId
));
3226 FSYNC_VolOp(volumeId
, pname
, FSYNC_VOL_DONE
, 0, NULL
);
3227 FSYNC_VolOp(h
.id
, pname
, FSYNC_VOL_ON
, 0, NULL
);
3230 # ifdef AFS_DEMAND_ATTACH_FS
3232 VUnlockVolumeById(volumeId
, partP
);
3234 # endif /* AFS_DEMAND_ATTACH_FS */
3242 * returns a static string used to print either 32 or 64 bit inode numbers.
3245 PrintInode(char *s
, Inode ino
)
3247 static afs_ino_str_t result
;
3251 snprintf(s
, sizeof(afs_ino_str_t
), "%llu", (afs_uintmax_t
) ino
);
3258 /* Routines to facilitate removing zero link count files. */
3259 #define MAX_ZLC_NAMES 32
3260 #define MAX_ZLC_NAMELEN 16
3261 typedef struct zlcList_s
{
3262 struct zlcList_s
*zlc_next
;
3264 char zlc_names
[MAX_ZLC_NAMES
][MAX_ZLC_NAMELEN
];
3267 static zlcList_t
*zlcAnchor
= NULL
;
3268 static zlcList_t
*zlcCur
= NULL
;
3271 AddToZLCDeleteList(char dir
, char *name
)
3273 opr_Assert(strlen(name
) <= MAX_ZLC_NAMELEN
- 3);
3275 if (!zlcCur
|| zlcCur
->zlc_n
>= MAX_ZLC_NAMES
) {
3276 if (zlcCur
&& zlcCur
->zlc_next
)
3277 zlcCur
= zlcCur
->zlc_next
;
3279 zlcList_t
*tmp
= malloc(sizeof(zlcList_t
));
3285 zlcCur
->zlc_next
= tmp
;
3289 zlcCur
->zlc_next
= NULL
;
3294 (void)sprintf(zlcCur
->zlc_names
[zlcCur
->zlc_n
], "%c" OS_DIRSEP
"%s", dir
, name
);
3296 (void)sprintf(zlcCur
->zlc_names
[zlcCur
->zlc_n
], "%s", name
);
3302 DeleteZLCFiles(char *path
)
3308 for (z
= zlcAnchor
; z
; z
= z
->zlc_next
) {
3309 for (i
= 0; i
< z
->zlc_n
; i
++) {
3311 (void)sprintf(fname
, "%s" OS_DIRSEP
"%s", path
, z
->zlc_names
[i
]);
3313 (void)sprintf(fname
, "%s", z
->zlc_names
[i
]);
3314 if (namei_unlink(fname
) < 0) {
3315 Log("ZLC: Can't unlink %s, dos error = %d\n", fname
,
3319 z
->zlc_n
= 0; /* Can reuse space. */
3332 tnext
= i
->zlc_next
;
3336 zlcCur
= zlcAnchor
= NULL
;
3340 #endif /* AFS_NAMEI_ENV */