SEARCHFS(2) BSD System Calls Manual SEARCHFS(2)
NAME
searchfs -- search a volume quickly
SYNOPSIS
#include <sys/attr.h>
#include <unistd.h>
int
searchfs(const char* path, struct fssearchblock* searchBlock, unsigned int* numMatches, unsigned int scriptCode, unsigned int options,
struct searchstate* state);
DESCRIPTION
The searchfs() function searches the volume (that is, mounted file system) specified by path for file system objects matching the criteria
specified by searchBlock, scriptCode, and options. The numMatches parameter returns the number of matching file system objects found. The
function also returns attributes of those file system objects in a buffer specified by searchBlock. The searchState parameter allows you
search the volume using multiple calls to searchfs(), resuming the search where it left off. The routine will only return objects to which
you have access (that is, you have execute permissions on the directories leading to this object from the root).
The path parameter must reference a valid file system object on the volume to be searched. Typically the path is to the volume's root direc-
tory. The entire volume is always searched. All directories listed in the path name leading to this object must be searchable.
The searchBlock parameter is a pointer to an fssearchblock structure, as defined by <sys/attr.h> (shown below). You are responsible for
filling out all fields of this structure before calling the function.
struct fssearchblock {
struct attrlist * returnattrs;
void * returnbuffer;
size_t returnbuffersize;
unsigned int maxmatches;
struct timeval timelimit;
void * searchparams1;
size_t sizeofsearchparams1;
void * searchparams2;
size_t sizeofsearchparams2;
struct attrlist searchattrs;
};
For information about the attrlist structure, see the discussion of getattrlist(2).
The fields of the fssearchblock structure are defined as follows.
returnattrs searchfs() can return arbitrary attributes of the file system objects that meet the designated search criteria passed in
via searchparams1 and searchparams2. This field must point to an attrlist structure that specifies the attributes that
you want returned. To request an attribute you must set the corresponding bit in the appropriate attrgroup_t field of
the attrlist structure. You are responsible for filling out all fields of this structure before calling the function.
You must not request volume attributes.
returnbuffer searchfs() places attributes of the matching file system objects into this returned attributes buffer. The attributes
for any given object are grouped together and packed in exactly the same way as they would be returned from
getdirentriesattr(2). The initial contents of this buffer are ignored.
returnbuffersize Set this field to the size, in bytes, of the buffer pointed to by returnbuffer.
maxmatches Specifies the maximum number of matches that you want this call to searchfs() to return.
timelimit Specifies the maximum time that you want this call to searchfs() to run.
If you're implementing a volume format, you should impose your own internal limit on the duration of this call to pre-
vent a malicious user program from monopolizing kernel resources.
searchparams1 Specifies the lower bound of the search criteria. This is discussed in detail below. You must place attribute values
into the buffer in the same way as they would be returned by getattrlist(2), where the searchattrs field determines the
exact layout of the attribute values.
sizeofsearchparams1 Set this field to the size, in bytes, of the buffer pointed to by searchparams1.
searchparams2 Specifies the upper bound of the search criteria. This is discussed in detail below. You must place attribute values
into the buffer in the same way as they would be returned by getattrlist(2), where the searchattrs field determines the
exact layout of the attribute values.
sizeofsearchparams2 Set this field to the size, in bytes, of the buffer pointed to by searchparams2.
searchattrs Specifies the attributes that you want you use for your search criteria. You are responsible for filling out all fields
of this structure before calling the function. To search for an attribute you must set the corresponding bit in the
appropriate attrgroup_t field of the attrlist structure, and place the appropriate values into the searchparam1 and
searchparam2 buffers. The attributes specified here determine the format of those buffers. This is discussed in detail
below.
The numMatches parameter points to an unsigned int variable. The initial value of this variable is ignored. On return, this variable con-
tains the number of matching file system objects found. The is always less than or equal to the maxmatches field of the searchBlock parame-
ter. The attributes for the matching objects have been placed into the returned attributes buffer.
The scriptCode parameter is currently ignored. You should always pass in the value 0x08000103, which corresponds to the UTF-8 text encoding
value defined by <CarbonCore/TextCommon.h>.
The options parameter is a bit set that controls the behaviour of searchfs(). The following option bits are defined.
SRCHFS_START If this bit is set, searchfs() will ignore the state parameter and start a new search. Otherwise searchfs()
assumes that searchstate is valid and attempts to resume a previous search based on that state.
SRCHFS_MATCHPARTIALNAMES If this bit is set, searchfs() will consider substrings to be successful matches when evaluating the ATTR_CMN_NAME
attribute.
SRCHFS_MATCHDIRS If this bit is set, searchfs() will search for directories that match the search criteria. To get meaningful
results you must specify either this bit or SRCHFS_MATCHFILES, or both.
SRCHFS_MATCHFILES If this bit is set, searchfs() will search for files that match the search criteria. To get meaningful results you
must specify either this bit or SRCHFS_MATCHDIRS, or both.
SRCHFS_SKIPLINKS If this bit is set, searchfs() will only return one reference for a hard linked file, rather than a reference for
each hard link to the file.
This option is not recommended for general development. Its primary client is the quotacheck(2) utility. Note
that not all filesystems that support searchfs() support this option and may return EINVAL if it is requested.
This option is privileged (the caller's effective UID must be 0) and cannot be used if you request the
ATTR_CMN_NAME or ATTR_CMN_PAROBJID attributes.
Introduced with Darwin 7.0 (Mac OS X version 10.3).
SRCHFS_SKIPINVISIBLE If this bit is set, searchfs() will not match any invisible file system objects (that is, objects whose
ATTR_CMN_FNDRINFO attribute has bit 6 set in the ninth byte) or any objects within invisible directories.
Introduced with Darwin 7.0 (Mac OS X version 10.3).
SRCHFS_SKIPPACKAGES If this bit is set, searchfs() will not match any file system objects that are inside a package. A package is
defined as a directory whose extension matches one of the extensions that are configured into the kernel by Launch
Services.
Introduced with Darwin 7.0 (Mac OS X version 10.3).
SRCHFS_SKIPINAPPROPRIATE If this bit is set, searchfs() will not match any file system objects that are within an inappropriate directory.
The current list of inappropriate directories contains one item: /System.
Introduced with Darwin 7.0 (Mac OS X version 10.3).
SRCHFS_NEGATEPARAMS If this bit is set, searchfs() will return all the file system objects that do not match the search criteria.
Introduced with Darwin 7.0 (Mac OS X version 10.3).
The state parameter is a pointer to an opaque data structure that searchfs() uses to maintain the state of a search between successive calls.
In your first call to searchfs(), you specify the SRCHFS_START flag in the options parameter. This tells searchfs() that the search state is
invalid and that it should start a new search. When this call completes, it may have only returned partial results; in that case, it will
have updated the structure pointed to by state. If you call searchfs() again, this time without specifying the SRCHFS_START flag in the
options parameter, it will resume the search where it left off, using the search state that it previously stored in the state structure. You
do not need to explicitly dispose of this state.
The searchfs() function returns significant errors in the followings cases.
o If it has found as many objects as you requested in the maxmatches field of the searchBlock parameter, it will return EAGAIN.
o If there is not enough space in the returned attributes buffer for the first match, it will return ENOBUFS. You should allocate a larger
returned attributes buffer and try again. numMatches will be zero in this case.
o If the timeout expires it will return EAGAIN.
o If you attempt to resume a search (that is, SRCHFS_START is not specified in the options parameter) and the catalog has changed since the
last search, the function will return EBUSY. You must start your search again from the beginning.
If searchfs() returns EAGAIN, the value in numMatches may be greater than zero. This is known as a partial result. You should be sure to
process these matches before calling searchfs() again.
SEARCH CRITERIA
You specify the search criteria using a combination of the searchattrs, searchparams1, sizeofsearchparams1, searchparams2, and
sizeofsearchparams2 fields of the searchBlock parameter, and various flags in the options parameter. The searchattrs field determines the
attributes considered when comparing a file system object to the search criteria. You can specify that an attribute should be considered by
setting the corresponding bit in the appropriate attrgroup_t field of the attrlist structure. See the discussion of getattrlist(2) for a
detailed description of this structure.
The searchparams1, sizeofsearchparams1, searchparams2, and sizeofsearchparams2 fields specify the attribute values that must be matched. The
format of each of these buffers is determined by the attributes that you're searching for. The values are packed in exactly the same way as
they would be returned from getattrlist(2), including the leading u_int32_t length value. Note that the size of these buffers must be
bounded by SEARCHFS_MAX_SEARCHPARMS bytes, which is defined in <sys/attr.h>.
The attribute values in the first and second search buffers form a lower and upper bound for the search, respectively. These have different
meanings depending on the type of attribute.
o For string attributes (specifically ATTR_CMN_NAME, the object name), the value in the first search buffer is significant and the value in
the second search buffer is ignored. The string comparison is either an exact match or a substring match depending on the
SRCHFS_MATCHPARTIALNAMES flag in the options parameter.
o For structured attributes (specifically ATTR_CMN_FNDRINFO, the Finder information), the value from the file system object is masked (log-
ical AND) with the value in the second search buffer and then compared, byte for byte, against the value in the first search buffer. If
it is equal, the object is a match.
o For scalar attributes (all other attributes, for example, ATTR_CMN_MODTIME, the modification date), the values in the first and second
search buffers are literally a lower and upper bound. An object matches the criteria if its value is greater than or equal to the value
in the first buffer and less than or equal to the value in the second.
RETURN VALUES
Upon successful completion, a value of 0 is returned. This means that the entire volume has been searched and all matches returned. Other-
wise, a value of -1 is returned and errno is set to indicate the error.
See the discussion of the EAGAIN, ENOBUFS, and EBUSY error codes above.
COMPATIBILITY
Not all volumes support searchfs(). You can test whether a volume supports searchfs() by using getattrlist(2) to get the volume capabilities
attribute ATTR_VOL_CAPABILITIES, and then testing the VOL_CAP_INT_SEARCHFS flag.
The searchfs() function has been undocumented for more than two years. In that time a number of volume format implementations have been cre-
ated without a proper specification for the behaviour of this routine. You may encounter volume format implementations with slightly differ-
ent behaviour than what is described here. Your program is expected to be tolerant of this variant behaviour.
If you're implementing a volume format that supports searchfs(), you should be careful to support the behaviour specified by this document.
A bug in systems prior to Darwin 7.0 (Mac OS X version 10.3) makes searching for the ATTR_CMN_BKUPTIME attribute tricky. The bug causes the
attribute to consume two items in the search attribute buffers, the first in the proper place and the second between ATTR_CMN_FNDRINFO and
ATTR_CMN_OWNERID.
ERRORS
searchfs() will fail if:
[ENOTSUP] The volume does not support searchfs().
[ENOTDIR] A component of the path prefix is not a directory.
[ENAMETOOLONG] A component of a path name exceeded NAME_MAX characters, or an entire path name exceeded PATH_MAX characters.
[ENOENT] The file system object does not exist.
[EACCES] Search permission is denied for a component of the path prefix.
[ELOOP] Too many symbolic links were encountered in translating the pathname.
[EFAULT] One of the pointer parameters points to an invalid address.
[EINVAL] The options parameter contains an invalid flag or sizeofsearchparams1/2 is greater than SEARCHFS_MAX_SEARCHPARMS (see
attr.h). Additionally, filesystems that do not support SRCHFS_SKIPLINKS may return EINVAL if this search option is
requested.
[EAGAIN] The search terminated with partial results, either because numMatches has hit the limit specified by maxmatches or because
the timeout expired. Process the matches returned so far and then call searchfs() again to look for more.
[ENOBUFS] The returned attributes buffer is too small for the first match. You should allocate a larger returned attributes buffer
and try again. numMatches will be zero in this case.
[EBUSY] The search could not be resumed because the volume has changed.
[EIO] An I/O error occurred while reading from or writing to the file system.
CAVEATS
The list of attributes that are valid as search criteria currently includes the following list of attributes. Note that this list is sub-
stantially smaller than what is available via getattrlist(2) for a particular filesystem object. In general, a filesystem that supports
searchfs() will typically supply per-item attributes for matched objects that are also supported by the getdirentries(2) system call. This
varies from filesystem to filesystem.
ATTR_CMN_NAME
ATTR_CMN_OBJID
ATTR_CMN_PAROBJID
ATTR_CMN_CRTIME
ATTR_CMN_MODTIME
ATTR_CMN_CHGTIME
ATTR_CMN_ACCTIME
ATTR_CMN_BKUPTIME
ATTR_CMN_FNDRINFO
ATTR_CMN_BKUPTIME
ATTR_CMN_OWNERID
ATTR_CMN_GRPID
ATTR_CMN_ACCESSMASK
ATTR_CMN_FILEID
ATTR_CMN_PARENTID
ATTR_DIR_ENTRYCOUNT
ATTR_FILE_DATALENGTH
ATTR_FILE_DATAALLOCSIZE
ATTR_FILE_RSRCLENGTH
ATTR_FILE_RSRCALLOCSIZE
EXAMPLES
The following code searches a volume for files of the specified type and creator.
#include <assert.h>
#include <stdio.h>
#include <stddef.h>
#include <string.h>
#include <sys/attr.h>
#include <sys/errno.h>
#include <unistd.h>
typedef struct attrlist attrlist_t;
typedef struct fssearchblock fssearchblock_t;
typedef struct searchstate searchstate_t;
struct SearchAttrBuf {
u_int32_t length;
char finderInfo[32];
};
typedef struct SearchAttrBuf SearchAttrBuf;
struct ResultAttrBuf {
u_int32_t length;
attrreference_t name;
fsobj_id_t parObjID;
};
typedef struct ResultAttrBuf ResultAttrBuf;
enum {
kMatchesPerCall = 16
};
static int SearchFSDemo(
const char *volPath,
const char *type,
const char *creator
)
{
int err;
fssearchblock_t searchBlock;
SearchAttrBuf lower;
SearchAttrBuf upper;
static const unsigned char kAllOnes[4] = { 0xFF, 0xFF, 0xFF, 0xFF };
unsigned int matchCount;
unsigned int matchIndex;
unsigned int options;
searchstate_t state;
ResultAttrBuf * thisEntry;
attrlist_t returnAttrList;
char resultAttrBuf[ kMatchesPerCall
* (sizeof(ResultAttrBuf) + 64)];
// resultAttrBuf is big enough for kMatchesPerCall entries,
// assuming that the average name length is less than 64.
assert(strlen(type) == 4);
assert(strlen(creator) == 4);
memset(&searchBlock, 0, sizeof(searchBlock));
searchBlock.searchattrs.bitmapcount = ATTR_BIT_MAP_COUNT;
searchBlock.searchattrs.commonattr = ATTR_CMN_FNDRINFO;
memset(&lower, 0, sizeof(lower));
memset(&upper, 0, sizeof(upper));
lower.length = sizeof(lower);
upper.length = sizeof(upper);
memcpy(&lower.finderInfo[0], type, 4);
memcpy(&lower.finderInfo[4], creator, 4);
memcpy(&upper.finderInfo[0], kAllOnes, 4);
memcpy(&upper.finderInfo[4], kAllOnes, 4);
searchBlock.searchparams1 = &lower;
searchBlock.sizeofsearchparams1 = sizeof(lower);
searchBlock.searchparams2 = &upper;
searchBlock.sizeofsearchparams2 = sizeof(lower);
searchBlock.timelimit.tv_sec = 0;
searchBlock.timelimit.tv_usec = 100 * 1000;
searchBlock.maxmatches = kMatchesPerCall;
memset(&returnAttrList, 0, sizeof(returnAttrList));
returnAttrList.bitmapcount = ATTR_BIT_MAP_COUNT;
returnAttrList.commonattr = ATTR_CMN_NAME | ATTR_CMN_PAROBJID;
searchBlock.returnattrs = &returnAttrList;
searchBlock.returnbuffer = resultAttrBuf;
searchBlock.returnbuffersize = sizeof(resultAttrBuf);
options = SRCHFS_START | SRCHFS_MATCHFILES;
do {
err = searchfs(
volPath,
&searchBlock,
&matchCount,
0x08000103,
options,
&state
);
if (err != 0) {
err = errno;
}
if ( (err == 0) || (err == EAGAIN) ) {
thisEntry = (ResultAttrBuf *) resultAttrBuf;
for (matchIndex = 0; matchIndex < matchCount; matchIndex++) {
printf("%08x ", thisEntry->parObjID.fid_objno);
printf(
"%s
",
((char *) &thisEntry->name)
+ thisEntry->name.attr_dataoffset
);
// Advance to the next entry.
((char *) thisEntry) += thisEntry->length;
}
}
options &= ~SRCHFS_START;
} while (err == EAGAIN);
return err;
}
SEE ALSO
getattrlist(2)
HISTORY
A searchfs() function call appeared in Darwin 1.3.1 (Mac OS X version 10.0).
Darwin October 13, 2008 Darwin