12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730 |
- /* vi: set sw=4 ts=4: */
- /*
- * Mini mount implementation for busybox
- *
- * Copyright (C) 1995, 1996 by Bruce Perens <bruce@pixar.com>.
- * Copyright (C) 1999-2004 by Erik Andersen <andersen@codepoet.org>
- * Copyright (C) 2005-2006 by Rob Landley <rob@landley.net>
- *
- * Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
- */
- /* Design notes: There is no spec for mount. Remind me to write one.
- mount_main() calls singlemount() which calls mount_it_now().
- mount_main() can loop through /etc/fstab for mount -a
- singlemount() can loop through /etc/filesystems for fstype detection.
- mount_it_now() does the actual mount.
- */
- #include "busybox.h"
- #include <mntent.h>
- /* Needed for nfs support only... */
- #include <syslog.h>
- #include <sys/utsname.h>
- #undef TRUE
- #undef FALSE
- #include <rpc/rpc.h>
- #include <rpc/pmap_prot.h>
- #include <rpc/pmap_clnt.h>
- #if defined(__dietlibc__)
- /* 16.12.2006, Sampo Kellomaki (sampo@iki.fi)
- * dietlibc-0.30 does not have implementation of getmntent_r() */
- /* OTOH: why we use getmntent_r instead of getmntent? TODO... */
- struct mntent *getmntent_r(FILE* stream, struct mntent* result, char* buffer, int bufsize)
- {
- /* *** XXX FIXME WARNING: This hack is NOT thread safe. --Sampo */
- struct mntent* ment = getmntent(stream);
- memcpy(result, ment, sizeof(struct mntent));
- return result;
- }
- #endif
- // Not real flags, but we want to be able to check for this.
- enum {
- MOUNT_USERS = (1<<28)*ENABLE_DESKTOP,
- MOUNT_NOAUTO = (1<<29),
- MOUNT_SWAP = (1<<30),
- };
- // TODO: more "user" flag compatibility.
- // "user" option (from mount manpage):
- // Only the user that mounted a filesystem can unmount it again.
- // If any user should be able to unmount, then use users instead of user
- // in the fstab line. The owner option is similar to the user option,
- // with the restriction that the user must be the owner of the special file.
- // This may be useful e.g. for /dev/fd if a login script makes
- // the console user owner of this device.
- /* Standard mount options (from -o options or --options), with corresponding
- * flags */
- struct {
- char *name;
- long flags;
- } static mount_options[] = {
- // MS_FLAGS set a bit. ~MS_FLAGS disable that bit. 0 flags are NOPs.
- USE_FEATURE_MOUNT_LOOP(
- {"loop", 0},
- )
- USE_FEATURE_MOUNT_FSTAB(
- {"defaults", 0},
- /* {"quiet", 0}, - do not filter out, vfat wants to see it */
- {"noauto", MOUNT_NOAUTO},
- {"swap", MOUNT_SWAP},
- USE_DESKTOP({"user", MOUNT_USERS},)
- USE_DESKTOP({"users", MOUNT_USERS},)
- )
- USE_FEATURE_MOUNT_FLAGS(
- // vfs flags
- {"nosuid", MS_NOSUID},
- {"suid", ~MS_NOSUID},
- {"dev", ~MS_NODEV},
- {"nodev", MS_NODEV},
- {"exec", ~MS_NOEXEC},
- {"noexec", MS_NOEXEC},
- {"sync", MS_SYNCHRONOUS},
- {"async", ~MS_SYNCHRONOUS},
- {"atime", ~MS_NOATIME},
- {"noatime", MS_NOATIME},
- {"diratime", ~MS_NODIRATIME},
- {"nodiratime", MS_NODIRATIME},
- {"loud", ~MS_SILENT},
- // action flags
- {"bind", MS_BIND},
- {"move", MS_MOVE},
- {"shared", MS_SHARED},
- {"slave", MS_SLAVE},
- {"private", MS_PRIVATE},
- {"unbindable", MS_UNBINDABLE},
- {"rshared", MS_SHARED|MS_RECURSIVE},
- {"rslave", MS_SLAVE|MS_RECURSIVE},
- {"rprivate", MS_SLAVE|MS_RECURSIVE},
- {"runbindable", MS_UNBINDABLE|MS_RECURSIVE},
- )
- // Always understood.
- {"ro", MS_RDONLY}, // vfs flag
- {"rw", ~MS_RDONLY}, // vfs flag
- {"remount", MS_REMOUNT}, // action flag
- };
- #define VECTOR_SIZE(v) (sizeof(v) / sizeof((v)[0]))
- /* Append mount options to string */
- static void append_mount_options(char **oldopts, char *newopts)
- {
- if (*oldopts && **oldopts) {
- /* do not insert options which are already there */
- while (newopts[0]) {
- char *p;
- int len = strlen(newopts);
- p = strchr(newopts, ',');
- if (p) len = p - newopts;
- p = *oldopts;
- while (1) {
- if (!strncmp(p, newopts, len)
- && (p[len]==',' || p[len]==0))
- goto skip;
- p = strchr(p,',');
- if(!p) break;
- p++;
- }
- p = xasprintf("%s,%.*s", *oldopts, len, newopts);
- free(*oldopts);
- *oldopts = p;
- skip:
- newopts += len;
- while (newopts[0] == ',') newopts++;
- }
- } else {
- if (ENABLE_FEATURE_CLEAN_UP) free(*oldopts);
- *oldopts = xstrdup(newopts);
- }
- }
- /* Use the mount_options list to parse options into flags.
- * Also return list of unrecognized options if unrecognized!=NULL */
- static int parse_mount_options(char *options, char **unrecognized)
- {
- int flags = MS_SILENT;
- // Loop through options
- for (;;) {
- int i;
- char *comma = strchr(options, ',');
- if (comma) *comma = 0;
- // Find this option in mount_options
- for (i = 0; i < VECTOR_SIZE(mount_options); i++) {
- if (!strcasecmp(mount_options[i].name, options)) {
- long fl = mount_options[i].flags;
- if (fl < 0) flags &= fl;
- else flags |= fl;
- break;
- }
- }
- // If unrecognized not NULL, append unrecognized mount options */
- if (unrecognized && i == VECTOR_SIZE(mount_options)) {
- // Add it to strflags, to pass on to kernel
- i = *unrecognized ? strlen(*unrecognized) : 0;
- *unrecognized = xrealloc(*unrecognized, i+strlen(options)+2);
- // Comma separated if it's not the first one
- if (i) (*unrecognized)[i++] = ',';
- strcpy((*unrecognized)+i, options);
- }
- // Advance to next option, or finish
- if (comma) {
- *comma = ',';
- options = ++comma;
- } else break;
- }
- return flags;
- }
- // Return a list of all block device backed filesystems
- static llist_t *get_block_backed_filesystems(void)
- {
- static const char *const filesystems[] = {
- "/etc/filesystems",
- "/proc/filesystems",
- 0
- };
- char *fs, *buf;
- llist_t *list = 0;
- int i;
- FILE *f;
- for (i = 0; filesystems[i]; i++) {
- f = fopen(filesystems[i], "r");
- if (!f) continue;
- while ((buf = xmalloc_getline(f)) != 0) {
- if (!strncmp(buf, "nodev", 5) && isspace(buf[5]))
- continue;
- fs = skip_whitespace(buf);
- if (*fs=='#' || *fs=='*' || !*fs) continue;
- llist_add_to_end(&list, xstrdup(fs));
- free(buf);
- }
- if (ENABLE_FEATURE_CLEAN_UP) fclose(f);
- }
- return list;
- }
- llist_t *fslist = 0;
- #if ENABLE_FEATURE_CLEAN_UP
- static void delete_block_backed_filesystems(void)
- {
- llist_free(fslist, free);
- }
- #else
- void delete_block_backed_filesystems(void);
- #endif
- #if ENABLE_FEATURE_MTAB_SUPPORT
- static int useMtab = 1;
- static int fakeIt;
- #else
- #define useMtab 0
- #define fakeIt 0
- #endif
- // Perform actual mount of specific filesystem at specific location.
- // NB: mp->xxx fields may be trashed on exit
- static int mount_it_now(struct mntent *mp, int vfsflags, char *filteropts)
- {
- int rc = 0;
- if (fakeIt) goto mtab;
- // Mount, with fallback to read-only if necessary.
- for (;;) {
- rc = mount(mp->mnt_fsname, mp->mnt_dir, mp->mnt_type,
- vfsflags, filteropts);
- if (!rc || (vfsflags&MS_RDONLY) || (errno!=EACCES && errno!=EROFS))
- break;
- bb_error_msg("%s is write-protected, mounting read-only",
- mp->mnt_fsname);
- vfsflags |= MS_RDONLY;
- }
- // Abort entirely if permission denied.
- if (rc && errno == EPERM)
- bb_error_msg_and_die(bb_msg_perm_denied_are_you_root);
- /* If the mount was successful, and we're maintaining an old-style
- * mtab file by hand, add the new entry to it now. */
- mtab:
- if (ENABLE_FEATURE_MTAB_SUPPORT && useMtab && !rc && !(vfsflags & MS_REMOUNT)) {
- char *fsname;
- FILE *mountTable = setmntent(bb_path_mtab_file, "a+");
- int i;
- if (!mountTable) {
- bb_error_msg("no %s",bb_path_mtab_file);
- goto ret;
- }
- // Add vfs string flags
- for (i=0; mount_options[i].flags != MS_REMOUNT; i++)
- if (mount_options[i].flags > 0 && (mount_options[i].flags & vfsflags))
- append_mount_options(&(mp->mnt_opts), mount_options[i].name);
- // Remove trailing / (if any) from directory we mounted on
- i = strlen(mp->mnt_dir) - 1;
- if (i > 0 && mp->mnt_dir[i] == '/') mp->mnt_dir[i] = 0;
- // Convert to canonical pathnames as needed
- mp->mnt_dir = bb_simplify_path(mp->mnt_dir);
- fsname = 0;
- if (!mp->mnt_type || !*mp->mnt_type) { /* bind mount */
- mp->mnt_fsname = fsname = bb_simplify_path(mp->mnt_fsname);
- mp->mnt_type = "bind";
- }
- mp->mnt_freq = mp->mnt_passno = 0;
- // Write and close.
- addmntent(mountTable, mp);
- endmntent(mountTable);
- if (ENABLE_FEATURE_CLEAN_UP) {
- free(mp->mnt_dir);
- free(fsname);
- }
- }
- ret:
- return rc;
- }
- #if ENABLE_FEATURE_MOUNT_NFS
- /*
- * Linux NFS mount
- * Copyright (C) 1993 Rick Sladkey <jrs@world.std.com>
- *
- * Licensed under GPLv2, see file LICENSE in this tarball for details.
- *
- * Wed Feb 8 12:51:48 1995, biro@yggdrasil.com (Ross Biro): allow all port
- * numbers to be specified on the command line.
- *
- * Fri, 8 Mar 1996 18:01:39, Swen Thuemmler <swen@uni-paderborn.de>:
- * Omit the call to connect() for Linux version 1.3.11 or later.
- *
- * Wed Oct 1 23:55:28 1997: Dick Streefland <dick_streefland@tasking.com>
- * Implemented the "bg", "fg" and "retry" mount options for NFS.
- *
- * 1999-02-22 Arkadiusz Mi¶kiewicz <misiek@misiek.eu.org>
- * - added Native Language Support
- *
- * Modified by Olaf Kirch and Trond Myklebust for new NFS code,
- * plus NFSv3 stuff.
- */
- /* This is just a warning of a common mistake. Possibly this should be a
- * uclibc faq entry rather than in busybox... */
- #if defined(__UCLIBC__) && ! defined(__UCLIBC_HAS_RPC__)
- #error "You need to build uClibc with UCLIBC_HAS_RPC for NFS support."
- #endif
- #define MOUNTPORT 635
- #define MNTPATHLEN 1024
- #define MNTNAMLEN 255
- #define FHSIZE 32
- #define FHSIZE3 64
- typedef char fhandle[FHSIZE];
- typedef struct {
- unsigned int fhandle3_len;
- char *fhandle3_val;
- } fhandle3;
- enum mountstat3 {
- MNT_OK = 0,
- MNT3ERR_PERM = 1,
- MNT3ERR_NOENT = 2,
- MNT3ERR_IO = 5,
- MNT3ERR_ACCES = 13,
- MNT3ERR_NOTDIR = 20,
- MNT3ERR_INVAL = 22,
- MNT3ERR_NAMETOOLONG = 63,
- MNT3ERR_NOTSUPP = 10004,
- MNT3ERR_SERVERFAULT = 10006,
- };
- typedef enum mountstat3 mountstat3;
- struct fhstatus {
- unsigned int fhs_status;
- union {
- fhandle fhs_fhandle;
- } fhstatus_u;
- };
- typedef struct fhstatus fhstatus;
- struct mountres3_ok {
- fhandle3 fhandle;
- struct {
- unsigned int auth_flavours_len;
- char *auth_flavours_val;
- } auth_flavours;
- };
- typedef struct mountres3_ok mountres3_ok;
- struct mountres3 {
- mountstat3 fhs_status;
- union {
- mountres3_ok mountinfo;
- } mountres3_u;
- };
- typedef struct mountres3 mountres3;
- typedef char *dirpath;
- typedef char *name;
- typedef struct mountbody *mountlist;
- struct mountbody {
- name ml_hostname;
- dirpath ml_directory;
- mountlist ml_next;
- };
- typedef struct mountbody mountbody;
- typedef struct groupnode *groups;
- struct groupnode {
- name gr_name;
- groups gr_next;
- };
- typedef struct groupnode groupnode;
- typedef struct exportnode *exports;
- struct exportnode {
- dirpath ex_dir;
- groups ex_groups;
- exports ex_next;
- };
- typedef struct exportnode exportnode;
- struct ppathcnf {
- int pc_link_max;
- short pc_max_canon;
- short pc_max_input;
- short pc_name_max;
- short pc_path_max;
- short pc_pipe_buf;
- uint8_t pc_vdisable;
- char pc_xxx;
- short pc_mask[2];
- };
- typedef struct ppathcnf ppathcnf;
- #define MOUNTPROG 100005
- #define MOUNTVERS 1
- #define MOUNTPROC_NULL 0
- #define MOUNTPROC_MNT 1
- #define MOUNTPROC_DUMP 2
- #define MOUNTPROC_UMNT 3
- #define MOUNTPROC_UMNTALL 4
- #define MOUNTPROC_EXPORT 5
- #define MOUNTPROC_EXPORTALL 6
- #define MOUNTVERS_POSIX 2
- #define MOUNTPROC_PATHCONF 7
- #define MOUNT_V3 3
- #define MOUNTPROC3_NULL 0
- #define MOUNTPROC3_MNT 1
- #define MOUNTPROC3_DUMP 2
- #define MOUNTPROC3_UMNT 3
- #define MOUNTPROC3_UMNTALL 4
- #define MOUNTPROC3_EXPORT 5
- enum {
- #ifndef NFS_FHSIZE
- NFS_FHSIZE = 32,
- #endif
- #ifndef NFS_PORT
- NFS_PORT = 2049
- #endif
- };
- /*
- * We want to be able to compile mount on old kernels in such a way
- * that the binary will work well on more recent kernels.
- * Thus, if necessary we teach nfsmount.c the structure of new fields
- * that will come later.
- *
- * Moreover, the new kernel includes conflict with glibc includes
- * so it is easiest to ignore the kernel altogether (at compile time).
- */
- struct nfs2_fh {
- char data[32];
- };
- struct nfs3_fh {
- unsigned short size;
- unsigned char data[64];
- };
- struct nfs_mount_data {
- int version; /* 1 */
- int fd; /* 1 */
- struct nfs2_fh old_root; /* 1 */
- int flags; /* 1 */
- int rsize; /* 1 */
- int wsize; /* 1 */
- int timeo; /* 1 */
- int retrans; /* 1 */
- int acregmin; /* 1 */
- int acregmax; /* 1 */
- int acdirmin; /* 1 */
- int acdirmax; /* 1 */
- struct sockaddr_in addr; /* 1 */
- char hostname[256]; /* 1 */
- int namlen; /* 2 */
- unsigned int bsize; /* 3 */
- struct nfs3_fh root; /* 4 */
- };
- /* bits in the flags field */
- enum {
- NFS_MOUNT_SOFT = 0x0001, /* 1 */
- NFS_MOUNT_INTR = 0x0002, /* 1 */
- NFS_MOUNT_SECURE = 0x0004, /* 1 */
- NFS_MOUNT_POSIX = 0x0008, /* 1 */
- NFS_MOUNT_NOCTO = 0x0010, /* 1 */
- NFS_MOUNT_NOAC = 0x0020, /* 1 */
- NFS_MOUNT_TCP = 0x0040, /* 2 */
- NFS_MOUNT_VER3 = 0x0080, /* 3 */
- NFS_MOUNT_KERBEROS = 0x0100, /* 3 */
- NFS_MOUNT_NONLM = 0x0200 /* 3 */
- };
- /*
- * We need to translate between nfs status return values and
- * the local errno values which may not be the same.
- *
- * Andreas Schwab <schwab@LS5.informatik.uni-dortmund.de>: change errno:
- * "after #include <errno.h> the symbol errno is reserved for any use,
- * it cannot even be used as a struct tag or field name".
- */
- #ifndef EDQUOT
- #define EDQUOT ENOSPC
- #endif
- // Convert each NFSERR_BLAH into EBLAH
- static const struct {
- int stat;
- int errnum;
- } nfs_errtbl[] = {
- {0,0}, {1,EPERM}, {2,ENOENT}, {5,EIO}, {6,ENXIO}, {13,EACCES}, {17,EEXIST},
- {19,ENODEV}, {20,ENOTDIR}, {21,EISDIR}, {22,EINVAL}, {27,EFBIG},
- {28,ENOSPC}, {30,EROFS}, {63,ENAMETOOLONG}, {66,ENOTEMPTY}, {69,EDQUOT},
- {70,ESTALE}, {71,EREMOTE}, {-1,EIO}
- };
- static char *nfs_strerror(int status)
- {
- int i;
- static char buf[sizeof("unknown nfs status return value: ") + sizeof(int)*3];
- for (i = 0; nfs_errtbl[i].stat != -1; i++) {
- if (nfs_errtbl[i].stat == status)
- return strerror(nfs_errtbl[i].errnum);
- }
- sprintf(buf, "unknown nfs status return value: %d", status);
- return buf;
- }
- static bool_t xdr_fhandle(XDR *xdrs, fhandle objp)
- {
- if (!xdr_opaque(xdrs, objp, FHSIZE))
- return FALSE;
- return TRUE;
- }
- static bool_t xdr_fhstatus(XDR *xdrs, fhstatus *objp)
- {
- if (!xdr_u_int(xdrs, &objp->fhs_status))
- return FALSE;
- switch (objp->fhs_status) {
- case 0:
- if (!xdr_fhandle(xdrs, objp->fhstatus_u.fhs_fhandle))
- return FALSE;
- break;
- default:
- break;
- }
- return TRUE;
- }
- static bool_t xdr_dirpath(XDR *xdrs, dirpath *objp)
- {
- if (!xdr_string(xdrs, objp, MNTPATHLEN))
- return FALSE;
- return TRUE;
- }
- static bool_t xdr_fhandle3(XDR *xdrs, fhandle3 *objp)
- {
- if (!xdr_bytes(xdrs, (char **)&objp->fhandle3_val, (unsigned int *) &objp->fhandle3_len, FHSIZE3))
- return FALSE;
- return TRUE;
- }
- static bool_t xdr_mountres3_ok(XDR *xdrs, mountres3_ok *objp)
- {
- if (!xdr_fhandle3(xdrs, &objp->fhandle))
- return FALSE;
- if (!xdr_array(xdrs, &(objp->auth_flavours.auth_flavours_val), &(objp->auth_flavours.auth_flavours_len), ~0,
- sizeof (int), (xdrproc_t) xdr_int))
- return FALSE;
- return TRUE;
- }
- static bool_t xdr_mountstat3(XDR *xdrs, mountstat3 *objp)
- {
- if (!xdr_enum(xdrs, (enum_t *) objp))
- return FALSE;
- return TRUE;
- }
- static bool_t xdr_mountres3(XDR *xdrs, mountres3 *objp)
- {
- if (!xdr_mountstat3(xdrs, &objp->fhs_status))
- return FALSE;
- switch (objp->fhs_status) {
- case MNT_OK:
- if (!xdr_mountres3_ok(xdrs, &objp->mountres3_u.mountinfo))
- return FALSE;
- break;
- default:
- break;
- }
- return TRUE;
- }
- #define MAX_NFSPROT ((nfs_mount_version >= 4) ? 3 : 2)
- /*
- * nfs_mount_version according to the sources seen at compile time.
- */
- static int nfs_mount_version;
- static int kernel_version;
- /*
- * Unfortunately, the kernel prints annoying console messages
- * in case of an unexpected nfs mount version (instead of
- * just returning some error). Therefore we'll have to try
- * and figure out what version the kernel expects.
- *
- * Variables:
- * KERNEL_NFS_MOUNT_VERSION: kernel sources at compile time
- * NFS_MOUNT_VERSION: these nfsmount sources at compile time
- * nfs_mount_version: version this source and running kernel can handle
- */
- static void
- find_kernel_nfs_mount_version(void)
- {
- if (kernel_version)
- return;
- nfs_mount_version = 4; /* default */
- kernel_version = get_linux_version_code();
- if (kernel_version) {
- if (kernel_version < KERNEL_VERSION(2,1,32))
- nfs_mount_version = 1;
- else if (kernel_version < KERNEL_VERSION(2,2,18) ||
- (kernel_version >= KERNEL_VERSION(2,3,0) &&
- kernel_version < KERNEL_VERSION(2,3,99)))
- nfs_mount_version = 3;
- /* else v4 since 2.3.99pre4 */
- }
- }
- static struct pmap *
- get_mountport(struct sockaddr_in *server_addr,
- long unsigned prog,
- long unsigned version,
- long unsigned proto,
- long unsigned port)
- {
- struct pmaplist *pmap;
- static struct pmap p = {0, 0, 0, 0};
- server_addr->sin_port = PMAPPORT;
- pmap = pmap_getmaps(server_addr);
- if (version > MAX_NFSPROT)
- version = MAX_NFSPROT;
- if (!prog)
- prog = MOUNTPROG;
- p.pm_prog = prog;
- p.pm_vers = version;
- p.pm_prot = proto;
- p.pm_port = port;
- while (pmap) {
- if (pmap->pml_map.pm_prog != prog)
- goto next;
- if (!version && p.pm_vers > pmap->pml_map.pm_vers)
- goto next;
- if (version > 2 && pmap->pml_map.pm_vers != version)
- goto next;
- if (version && version <= 2 && pmap->pml_map.pm_vers > 2)
- goto next;
- if (pmap->pml_map.pm_vers > MAX_NFSPROT ||
- (proto && p.pm_prot && pmap->pml_map.pm_prot != proto) ||
- (port && pmap->pml_map.pm_port != port))
- goto next;
- memcpy(&p, &pmap->pml_map, sizeof(p));
- next:
- pmap = pmap->pml_next;
- }
- if (!p.pm_vers)
- p.pm_vers = MOUNTVERS;
- if (!p.pm_port)
- p.pm_port = MOUNTPORT;
- if (!p.pm_prot)
- p.pm_prot = IPPROTO_TCP;
- return &p;
- }
- static int daemonize(void)
- {
- int fd;
- int pid = fork();
- if (pid < 0) /* error */
- return -errno;
- if (pid > 0) /* parent */
- return 0;
- /* child */
- fd = xopen(bb_dev_null, O_RDWR);
- dup2(fd, 0);
- dup2(fd, 1);
- dup2(fd, 2);
- while (fd > 2) close(fd--);
- setsid();
- openlog(applet_name, LOG_PID, LOG_DAEMON);
- logmode = LOGMODE_SYSLOG;
- return 1;
- }
- // TODO
- static inline int we_saw_this_host_before(const char *hostname)
- {
- return 0;
- }
- /* RPC strerror analogs are terminally idiotic:
- * *mandatory* prefix and \n at end.
- * This hopefully helps. Usage:
- * error_msg_rpc(clnt_*error*(" ")) */
- static void error_msg_rpc(const char *msg)
- {
- int len;
- while (msg[0] == ' ' || msg[0] == ':') msg++;
- len = strlen(msg);
- while (len && msg[len-1] == '\n') len--;
- bb_error_msg("%.*s", len, msg);
- }
- // NB: mp->xxx fields may be trashed on exit
- static int nfsmount(struct mntent *mp, int vfsflags, char *filteropts)
- {
- CLIENT *mclient;
- char *hostname;
- char *pathname;
- char *mounthost;
- struct nfs_mount_data data;
- char *opt;
- struct hostent *hp;
- struct sockaddr_in server_addr;
- struct sockaddr_in mount_server_addr;
- int msock, fsock;
- union {
- struct fhstatus nfsv2;
- struct mountres3 nfsv3;
- } status;
- int daemonized;
- char *s;
- int port;
- int mountport;
- int proto;
- int bg;
- int soft;
- int intr;
- int posix;
- int nocto;
- int noac;
- int nolock;
- int retry;
- int tcp;
- int mountprog;
- int mountvers;
- int nfsprog;
- int nfsvers;
- int retval;
- find_kernel_nfs_mount_version();
- daemonized = 0;
- mounthost = NULL;
- retval = ETIMEDOUT;
- msock = fsock = -1;
- mclient = NULL;
- /* NB: hostname, mounthost, filteropts must be free()d prior to return */
- filteropts = xstrdup(filteropts); /* going to trash it later... */
- hostname = xstrdup(mp->mnt_fsname);
- /* mount_main() guarantees that ':' is there */
- s = strchr(hostname, ':');
- pathname = s + 1;
- *s = '\0';
- /* Ignore all but first hostname in replicated mounts
- until they can be fully supported. (mack@sgi.com) */
- s = strchr(hostname, ',');
- if (s) {
- *s = '\0';
- bb_error_msg("warning: multiple hostnames not supported");
- }
- server_addr.sin_family = AF_INET;
- if (!inet_aton(hostname, &server_addr.sin_addr)) {
- hp = gethostbyname(hostname);
- if (hp == NULL) {
- bb_herror_msg("%s", hostname);
- goto fail;
- }
- if (hp->h_length > sizeof(struct in_addr)) {
- bb_error_msg("got bad hp->h_length");
- hp->h_length = sizeof(struct in_addr);
- }
- memcpy(&server_addr.sin_addr,
- hp->h_addr, hp->h_length);
- }
- memcpy(&mount_server_addr, &server_addr, sizeof(mount_server_addr));
- /* add IP address to mtab options for use when unmounting */
- if (!mp->mnt_opts) { /* TODO: actually mp->mnt_opts is never NULL */
- mp->mnt_opts = xasprintf("addr=%s", inet_ntoa(server_addr.sin_addr));
- } else {
- char *tmp = xasprintf("%s%saddr=%s", mp->mnt_opts,
- mp->mnt_opts[0] ? "," : "",
- inet_ntoa(server_addr.sin_addr));
- free(mp->mnt_opts);
- mp->mnt_opts = tmp;
- }
- /* Set default options.
- * rsize/wsize (and bsize, for ver >= 3) are left 0 in order to
- * let the kernel decide.
- * timeo is filled in after we know whether it'll be TCP or UDP. */
- memset(&data, 0, sizeof(data));
- data.retrans = 3;
- data.acregmin = 3;
- data.acregmax = 60;
- data.acdirmin = 30;
- data.acdirmax = 60;
- data.namlen = NAME_MAX;
- bg = 0;
- soft = 0;
- intr = 0;
- posix = 0;
- nocto = 0;
- nolock = 0;
- noac = 0;
- retry = 10000; /* 10000 minutes ~ 1 week */
- tcp = 0;
- mountprog = MOUNTPROG;
- mountvers = 0;
- port = 0;
- mountport = 0;
- nfsprog = 100003;
- nfsvers = 0;
- /* parse options */
- for (opt = strtok(filteropts, ","); opt; opt = strtok(NULL, ",")) {
- char *opteq = strchr(opt, '=');
- if (opteq) {
- const char *const options[] = {
- /* 0 */ "rsize",
- /* 1 */ "wsize",
- /* 2 */ "timeo",
- /* 3 */ "retrans",
- /* 4 */ "acregmin",
- /* 5 */ "acregmax",
- /* 6 */ "acdirmin",
- /* 7 */ "acdirmax",
- /* 8 */ "actimeo",
- /* 9 */ "retry",
- /* 10 */ "port",
- /* 11 */ "mountport",
- /* 12 */ "mounthost",
- /* 13 */ "mountprog",
- /* 14 */ "mountvers",
- /* 15 */ "nfsprog",
- /* 16 */ "nfsvers",
- /* 17 */ "vers",
- /* 18 */ "proto",
- /* 19 */ "namlen",
- /* 20 */ "addr",
- NULL
- };
- int val = xatoi_u(opteq + 1);
- *opteq = '\0';
- switch (index_in_str_array(options, opt)) {
- case 0: // "rsize"
- data.rsize = val;
- break;
- case 1: // "wsize"
- data.wsize = val;
- break;
- case 2: // "timeo"
- data.timeo = val;
- break;
- case 3: // "retrans"
- data.retrans = val;
- break;
- case 4: // "acregmin"
- data.acregmin = val;
- break;
- case 5: // "acregmax"
- data.acregmax = val;
- break;
- case 6: // "acdirmin"
- data.acdirmin = val;
- break;
- case 7: // "acdirmax"
- data.acdirmax = val;
- break;
- case 8: // "actimeo"
- data.acregmin = val;
- data.acregmax = val;
- data.acdirmin = val;
- data.acdirmax = val;
- break;
- case 9: // "retry"
- retry = val;
- break;
- case 10: // "port"
- port = val;
- break;
- case 11: // "mountport"
- mountport = val;
- break;
- case 12: // "mounthost"
- mounthost = xstrndup(opteq+1,
- strcspn(opteq+1," \t\n\r,"));
- break;
- case 13: // "mountprog"
- mountprog = val;
- break;
- case 14: // "mountvers"
- mountvers = val;
- break;
- case 15: // "nfsprog"
- nfsprog = val;
- break;
- case 16: // "nfsvers"
- case 17: // "vers"
- nfsvers = val;
- break;
- case 18: // "proto"
- if (!strncmp(opteq+1, "tcp", 3))
- tcp = 1;
- else if (!strncmp(opteq+1, "udp", 3))
- tcp = 0;
- else
- bb_error_msg("warning: unrecognized proto= option");
- break;
- case 19: // "namlen"
- if (nfs_mount_version >= 2)
- data.namlen = val;
- else
- bb_error_msg("warning: option namlen is not supported\n");
- break;
- case 20: // "addr" - ignore
- break;
- default:
- bb_error_msg("unknown nfs mount parameter: %s=%d", opt, val);
- goto fail;
- }
- }
- else {
- const char *const options[] = {
- "bg",
- "fg",
- "soft",
- "hard",
- "intr",
- "posix",
- "cto",
- "ac",
- "tcp",
- "udp",
- "lock",
- NULL
- };
- int val = 1;
- if (!strncmp(opt, "no", 2)) {
- val = 0;
- opt += 2;
- }
- switch (index_in_str_array(options, opt)) {
- case 0: // "bg"
- bg = val;
- break;
- case 1: // "fg"
- bg = !val;
- break;
- case 2: // "soft"
- soft = val;
- break;
- case 3: // "hard"
- soft = !val;
- break;
- case 4: // "intr"
- intr = val;
- break;
- case 5: // "posix"
- posix = val;
- break;
- case 6: // "cto"
- nocto = !val;
- break;
- case 7: // "ac"
- noac = !val;
- break;
- case 8: // "tcp"
- tcp = val;
- break;
- case 9: // "udp"
- tcp = !val;
- break;
- case 10: // "lock"
- if (nfs_mount_version >= 3)
- nolock = !val;
- else
- bb_error_msg("warning: option nolock is not supported");
- break;
- default:
- bb_error_msg("unknown nfs mount option: %s%s", val ? "" : "no", opt);
- goto fail;
- }
- }
- }
- proto = (tcp) ? IPPROTO_TCP : IPPROTO_UDP;
- data.flags = (soft ? NFS_MOUNT_SOFT : 0)
- | (intr ? NFS_MOUNT_INTR : 0)
- | (posix ? NFS_MOUNT_POSIX : 0)
- | (nocto ? NFS_MOUNT_NOCTO : 0)
- | (noac ? NFS_MOUNT_NOAC : 0);
- if (nfs_mount_version >= 2)
- data.flags |= (tcp ? NFS_MOUNT_TCP : 0);
- if (nfs_mount_version >= 3)
- data.flags |= (nolock ? NFS_MOUNT_NONLM : 0);
- if (nfsvers > MAX_NFSPROT || mountvers > MAX_NFSPROT) {
- bb_error_msg("NFSv%d not supported", nfsvers);
- goto fail;
- }
- if (nfsvers && !mountvers)
- mountvers = (nfsvers < 3) ? 1 : nfsvers;
- if (nfsvers && nfsvers < mountvers) {
- mountvers = nfsvers;
- }
- /* Adjust options if none specified */
- if (!data.timeo)
- data.timeo = tcp ? 70 : 7;
- data.version = nfs_mount_version;
- if (vfsflags & MS_REMOUNT)
- goto do_mount;
- /*
- * If the previous mount operation on the same host was
- * backgrounded, and the "bg" for this mount is also set,
- * give up immediately, to avoid the initial timeout.
- */
- if (bg && we_saw_this_host_before(hostname)) {
- daemonized = daemonize(); /* parent or error */
- if (daemonized <= 0) { /* parent or error */
- retval = -daemonized;
- goto ret;
- }
- }
- /* create mount daemon client */
- /* See if the nfs host = mount host. */
- if (mounthost) {
- if (mounthost[0] >= '0' && mounthost[0] <= '9') {
- mount_server_addr.sin_family = AF_INET;
- mount_server_addr.sin_addr.s_addr = inet_addr(hostname);
- } else {
- hp = gethostbyname(mounthost);
- if (hp == NULL) {
- bb_herror_msg("%s", mounthost);
- goto fail;
- } else {
- if (hp->h_length > sizeof(struct in_addr)) {
- bb_error_msg("got bad hp->h_length?");
- hp->h_length = sizeof(struct in_addr);
- }
- mount_server_addr.sin_family = AF_INET;
- memcpy(&mount_server_addr.sin_addr,
- hp->h_addr, hp->h_length);
- }
- }
- }
- /*
- * The following loop implements the mount retries. When the mount
- * times out, and the "bg" option is set, we background ourself
- * and continue trying.
- *
- * The case where the mount point is not present and the "bg"
- * option is set, is treated as a timeout. This is done to
- * support nested mounts.
- *
- * The "retry" count specified by the user is the number of
- * minutes to retry before giving up.
- */
- {
- struct timeval total_timeout;
- struct timeval retry_timeout;
- struct pmap* pm_mnt;
- time_t t;
- time_t prevt;
- time_t timeout;
- retry_timeout.tv_sec = 3;
- retry_timeout.tv_usec = 0;
- total_timeout.tv_sec = 20;
- total_timeout.tv_usec = 0;
- timeout = time(NULL) + 60 * retry;
- prevt = 0;
- t = 30;
- retry:
- /* be careful not to use too many CPU cycles */
- if (t - prevt < 30)
- sleep(30);
- pm_mnt = get_mountport(&mount_server_addr,
- mountprog,
- mountvers,
- proto,
- mountport);
- nfsvers = (pm_mnt->pm_vers < 2) ? 2 : pm_mnt->pm_vers;
- /* contact the mount daemon via TCP */
- mount_server_addr.sin_port = htons(pm_mnt->pm_port);
- msock = RPC_ANYSOCK;
- switch (pm_mnt->pm_prot) {
- case IPPROTO_UDP:
- mclient = clntudp_create(&mount_server_addr,
- pm_mnt->pm_prog,
- pm_mnt->pm_vers,
- retry_timeout,
- &msock);
- if (mclient)
- break;
- mount_server_addr.sin_port = htons(pm_mnt->pm_port);
- msock = RPC_ANYSOCK;
- case IPPROTO_TCP:
- mclient = clnttcp_create(&mount_server_addr,
- pm_mnt->pm_prog,
- pm_mnt->pm_vers,
- &msock, 0, 0);
- break;
- default:
- mclient = 0;
- }
- if (!mclient) {
- if (!daemonized && prevt == 0)
- error_msg_rpc(clnt_spcreateerror(" "));
- } else {
- enum clnt_stat clnt_stat;
- /* try to mount hostname:pathname */
- mclient->cl_auth = authunix_create_default();
- /* make pointers in xdr_mountres3 NULL so
- * that xdr_array allocates memory for us
- */
- memset(&status, 0, sizeof(status));
- if (pm_mnt->pm_vers == 3)
- clnt_stat = clnt_call(mclient, MOUNTPROC3_MNT,
- (xdrproc_t) xdr_dirpath,
- (caddr_t) &pathname,
- (xdrproc_t) xdr_mountres3,
- (caddr_t) &status,
- total_timeout);
- else
- clnt_stat = clnt_call(mclient, MOUNTPROC_MNT,
- (xdrproc_t) xdr_dirpath,
- (caddr_t) &pathname,
- (xdrproc_t) xdr_fhstatus,
- (caddr_t) &status,
- total_timeout);
- if (clnt_stat == RPC_SUCCESS)
- goto prepare_kernel_data; /* we're done */
- if (errno != ECONNREFUSED) {
- error_msg_rpc(clnt_sperror(mclient, " "));
- goto fail; /* don't retry */
- }
- /* Connection refused */
- if (!daemonized && prevt == 0) /* print just once */
- error_msg_rpc(clnt_sperror(mclient, " "));
- auth_destroy(mclient->cl_auth);
- clnt_destroy(mclient);
- mclient = 0;
- close(msock);
- }
- /* Timeout. We are going to retry... maybe */
- if (!bg)
- goto fail;
- if (!daemonized) {
- daemonized = daemonize();
- if (daemonized <= 0) { /* parent or error */
- retval = -daemonized;
- goto ret;
- }
- }
- prevt = t;
- t = time(NULL);
- if (t >= timeout)
- /* TODO error message */
- goto fail;
- goto retry;
- }
- prepare_kernel_data:
- if (nfsvers == 2) {
- if (status.nfsv2.fhs_status != 0) {
- bb_error_msg("%s:%s failed, reason given by server: %s",
- hostname, pathname,
- nfs_strerror(status.nfsv2.fhs_status));
- goto fail;
- }
- memcpy(data.root.data,
- (char *) status.nfsv2.fhstatus_u.fhs_fhandle,
- NFS_FHSIZE);
- data.root.size = NFS_FHSIZE;
- memcpy(data.old_root.data,
- (char *) status.nfsv2.fhstatus_u.fhs_fhandle,
- NFS_FHSIZE);
- } else {
- fhandle3 *my_fhandle;
- if (status.nfsv3.fhs_status != 0) {
- bb_error_msg("%s:%s failed, reason given by server: %s",
- hostname, pathname,
- nfs_strerror(status.nfsv3.fhs_status));
- goto fail;
- }
- my_fhandle = &status.nfsv3.mountres3_u.mountinfo.fhandle;
- memset(data.old_root.data, 0, NFS_FHSIZE);
- memset(&data.root, 0, sizeof(data.root));
- data.root.size = my_fhandle->fhandle3_len;
- memcpy(data.root.data,
- (char *) my_fhandle->fhandle3_val,
- my_fhandle->fhandle3_len);
- data.flags |= NFS_MOUNT_VER3;
- }
- /* create nfs socket for kernel */
- if (tcp) {
- if (nfs_mount_version < 3) {
- bb_error_msg("NFS over TCP is not supported");
- goto fail;
- }
- fsock = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
- } else
- fsock = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
- if (fsock < 0) {
- bb_perror_msg("nfs socket");
- goto fail;
- }
- if (bindresvport(fsock, 0) < 0) {
- bb_perror_msg("nfs bindresvport");
- goto fail;
- }
- if (port == 0) {
- server_addr.sin_port = PMAPPORT;
- port = pmap_getport(&server_addr, nfsprog, nfsvers,
- tcp ? IPPROTO_TCP : IPPROTO_UDP);
- if (port == 0)
- port = NFS_PORT;
- }
- server_addr.sin_port = htons(port);
- /* prepare data structure for kernel */
- data.fd = fsock;
- memcpy((char *) &data.addr, (char *) &server_addr, sizeof(data.addr));
- strncpy(data.hostname, hostname, sizeof(data.hostname));
- /* clean up */
- auth_destroy(mclient->cl_auth);
- clnt_destroy(mclient);
- close(msock);
- if (bg) {
- /* We must wait until mount directory is available */
- struct stat statbuf;
- int delay = 1;
- while (stat(mp->mnt_dir, &statbuf) == -1) {
- if (!daemonized) {
- daemonized = daemonize();
- if (daemonized <= 0) { /* parent or error */
- retval = -daemonized;
- goto ret;
- }
- }
- sleep(delay); /* 1, 2, 4, 8, 16, 30, ... */
- delay *= 2;
- if (delay > 30)
- delay = 30;
- }
- }
- do_mount: /* perform actual mount */
- mp->mnt_type = "nfs";
- retval = mount_it_now(mp, vfsflags, (char*)&data);
- goto ret;
- fail: /* abort */
- if (msock != -1) {
- if (mclient) {
- auth_destroy(mclient->cl_auth);
- clnt_destroy(mclient);
- }
- close(msock);
- }
- if (fsock != -1)
- close(fsock);
- ret:
- free(hostname);
- free(mounthost);
- free(filteropts);
- return retval;
- }
- #else /* !ENABLE_FEATURE_MOUNT_NFS */
- /* Never called. Call should be optimized out. */
- int nfsmount(struct mntent *mp, int vfsflags, char *filteropts);
- #endif /* !ENABLE_FEATURE_MOUNT_NFS */
- // Mount one directory. Handles CIFS, NFS, loopback, autobind, and filesystem
- // type detection. Returns 0 for success, nonzero for failure.
- // NB: mp->xxx fields may be trashed on exit
- static int singlemount(struct mntent *mp, int ignore_busy)
- {
- int rc = -1, vfsflags;
- char *loopFile = 0, *filteropts = 0;
- llist_t *fl = 0;
- struct stat st;
- vfsflags = parse_mount_options(mp->mnt_opts, &filteropts);
- // Treat fstype "auto" as unspecified.
- if (mp->mnt_type && !strcmp(mp->mnt_type,"auto")) mp->mnt_type = 0;
- // Might this be an CIFS filesystem?
- if (ENABLE_FEATURE_MOUNT_CIFS &&
- (!mp->mnt_type || !strcmp(mp->mnt_type,"cifs")) &&
- (mp->mnt_fsname[0]==mp->mnt_fsname[1] && (mp->mnt_fsname[0]=='/' || mp->mnt_fsname[0]=='\\')))
- {
- struct hostent *he;
- char ip[32], *s;
- rc = 1;
- // Replace '/' with '\' and verify that unc points to "//server/share".
- for (s = mp->mnt_fsname; *s; ++s)
- if (*s == '/') *s = '\\';
- // get server IP
- s = strrchr(mp->mnt_fsname, '\\');
- if (s == mp->mnt_fsname+1) goto report_error;
- *s = 0;
- he = gethostbyname(mp->mnt_fsname+2);
- *s = '\\';
- if (!he) goto report_error;
- // Insert ip=... option into string flags. (NOTE: Add IPv6 support.)
- sprintf(ip, "ip=%d.%d.%d.%d", he->h_addr[0], he->h_addr[1],
- he->h_addr[2], he->h_addr[3]);
- parse_mount_options(ip, &filteropts);
- // compose new unc '\\server-ip\share'
- mp->mnt_fsname = xasprintf("\\\\%s%s", ip+3,
- strchr(mp->mnt_fsname+2,'\\'));
- // lock is required
- vfsflags |= MS_MANDLOCK;
- mp->mnt_type = "cifs";
- rc = mount_it_now(mp, vfsflags, filteropts);
- if (ENABLE_FEATURE_CLEAN_UP) free(mp->mnt_fsname);
- goto report_error;
- }
- // Might this be an NFS filesystem?
- if (ENABLE_FEATURE_MOUNT_NFS &&
- (!mp->mnt_type || !strcmp(mp->mnt_type,"nfs")) &&
- strchr(mp->mnt_fsname, ':') != NULL)
- {
- rc = nfsmount(mp, vfsflags, filteropts);
- goto report_error;
- }
- // Look at the file. (Not found isn't a failure for remount, or for
- // a synthetic filesystem like proc or sysfs.)
- if (!lstat(mp->mnt_fsname, &st) && !(vfsflags & (MS_REMOUNT | MS_BIND | MS_MOVE)))
- {
- // Do we need to allocate a loopback device for it?
- if (ENABLE_FEATURE_MOUNT_LOOP && S_ISREG(st.st_mode)) {
- loopFile = bb_simplify_path(mp->mnt_fsname);
- mp->mnt_fsname = 0;
- switch (set_loop(&(mp->mnt_fsname), loopFile, 0)) {
- case 0:
- case 1:
- break;
- default:
- bb_error_msg( errno == EPERM || errno == EACCES
- ? bb_msg_perm_denied_are_you_root
- : "cannot setup loop device");
- return errno;
- }
- // Autodetect bind mounts
- } else if (S_ISDIR(st.st_mode) && !mp->mnt_type)
- vfsflags |= MS_BIND;
- }
- /* If we know the fstype (or don't need to), jump straight
- * to the actual mount. */
- if (mp->mnt_type || (vfsflags & (MS_REMOUNT | MS_BIND | MS_MOVE)))
- rc = mount_it_now(mp, vfsflags, filteropts);
- // Loop through filesystem types until mount succeeds or we run out
- else {
- /* Initialize list of block backed filesystems. This has to be
- * done here so that during "mount -a", mounts after /proc shows up
- * can autodetect. */
- if (!fslist) {
- fslist = get_block_backed_filesystems();
- if (ENABLE_FEATURE_CLEAN_UP && fslist)
- atexit(delete_block_backed_filesystems);
- }
- for (fl = fslist; fl; fl = fl->link) {
- mp->mnt_type = fl->data;
- rc = mount_it_now(mp, vfsflags, filteropts);
- if (!rc) break;
- }
- }
- // If mount failed, clean up loop file (if any).
- if (ENABLE_FEATURE_MOUNT_LOOP && rc && loopFile) {
- del_loop(mp->mnt_fsname);
- if (ENABLE_FEATURE_CLEAN_UP) {
- free(loopFile);
- free(mp->mnt_fsname);
- }
- }
- report_error:
- if (ENABLE_FEATURE_CLEAN_UP) free(filteropts);
- if (rc && errno == EBUSY && ignore_busy) rc = 0;
- if (rc < 0)
- /* perror here sometimes says "mounting ... on ... failed: Success" */
- bb_error_msg("mounting %s on %s failed", mp->mnt_fsname, mp->mnt_dir);
- return rc;
- }
- // Parse options, if necessary parse fstab/mtab, and call singlemount for
- // each directory to be mounted.
- const char must_be_root[] = "you must be root";
- int mount_main(int argc, char **argv)
- {
- enum { OPT_ALL = 0x10 };
- char *cmdopts = xstrdup(""), *fstype=0, *storage_path=0;
- char *opt_o;
- const char *fstabname;
- FILE *fstab;
- int i, j, rc = 0;
- unsigned opt;
- struct mntent mtpair[2], *mtcur = mtpair;
- SKIP_DESKTOP(const int nonroot = 0;)
- USE_DESKTOP( int nonroot = (getuid() != 0);)
- /* parse long options, like --bind and --move. Note that -o option
- * and --option are synonymous. Yes, this means --remount,rw works. */
- for (i = j = 0; i < argc; i++) {
- if (argv[i][0] == '-' && argv[i][1] == '-') {
- append_mount_options(&cmdopts, argv[i]+2);
- } else argv[j++] = argv[i];
- }
- argv[j] = 0;
- argc = j;
- // Parse remaining options
- opt = getopt32(argc, argv, "o:t:rwanfvs", &opt_o, &fstype);
- if (opt & 0x1) append_mount_options(&cmdopts, opt_o); // -o
- //if (opt & 0x2) // -t
- if (opt & 0x4) append_mount_options(&cmdopts, "ro"); // -r
- if (opt & 0x8) append_mount_options(&cmdopts, "rw"); // -w
- //if (opt & 0x10) // -a
- if (opt & 0x20) USE_FEATURE_MTAB_SUPPORT(useMtab = 0); // -n
- if (opt & 0x40) USE_FEATURE_MTAB_SUPPORT(fakeIt = 1); // -f
- //if (opt & 0x80) // -v: verbose (ignore)
- //if (opt & 0x100) // -s: sloppy (ignore)
- argv += optind;
- argc -= optind;
- // Three or more non-option arguments? Die with a usage message.
- if (argc > 2) bb_show_usage();
- // If we have no arguments, show currently mounted filesystems
- if (!argc) {
- if (!(opt & OPT_ALL)) {
- FILE *mountTable = setmntent(bb_path_mtab_file, "r");
- if (!mountTable) bb_error_msg_and_die("no %s", bb_path_mtab_file);
- while (getmntent_r(mountTable, mtpair, bb_common_bufsiz1,
- sizeof(bb_common_bufsiz1)))
- {
- // Don't show rootfs. FIXME: why??
- // util-linux 2.12a happily shows rootfs...
- //if (!strcmp(mtpair->mnt_fsname, "rootfs")) continue;
- if (!fstype || !strcmp(mtpair->mnt_type, fstype))
- printf("%s on %s type %s (%s)\n", mtpair->mnt_fsname,
- mtpair->mnt_dir, mtpair->mnt_type,
- mtpair->mnt_opts);
- }
- if (ENABLE_FEATURE_CLEAN_UP) endmntent(mountTable);
- return EXIT_SUCCESS;
- }
- } else storage_path = bb_simplify_path(argv[0]);
- // When we have two arguments, the second is the directory and we can
- // skip looking at fstab entirely. We can always abspath() the directory
- // argument when we get it.
- if (argc == 2) {
- if (nonroot)
- bb_error_msg_and_die(must_be_root);
- mtpair->mnt_fsname = argv[0];
- mtpair->mnt_dir = argv[1];
- mtpair->mnt_type = fstype;
- mtpair->mnt_opts = cmdopts;
- rc = singlemount(mtpair, 0);
- goto clean_up;
- }
- i = parse_mount_options(cmdopts, 0);
- if (nonroot && (i & ~MS_SILENT)) // Non-root users cannot specify flags
- bb_error_msg_and_die(must_be_root);
- // If we have a shared subtree flag, don't worry about fstab or mtab.
- if (ENABLE_FEATURE_MOUNT_FLAGS &&
- (i & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)))
- {
- rc = mount("", argv[0], "", i, "");
- if (rc) bb_perror_msg_and_die("%s", argv[0]);
- goto clean_up;
- }
- // Open either fstab or mtab
- fstabname = "/etc/fstab";
- if (i & MS_REMOUNT) {
- fstabname = bb_path_mtab_file;
- }
- fstab = setmntent(fstabname, "r");
- if (!fstab)
- bb_perror_msg_and_die("cannot read %s", fstabname);
- // Loop through entries until we find what we're looking for.
- memset(mtpair, 0, sizeof(mtpair));
- for (;;) {
- struct mntent *mtnext = (mtcur==mtpair ? mtpair+1 : mtpair);
- // Get next fstab entry
- if (!getmntent_r(fstab, mtcur, bb_common_bufsiz1
- + (mtcur==mtpair ? sizeof(bb_common_bufsiz1)/2 : 0),
- sizeof(bb_common_bufsiz1)/2))
- {
- // Were we looking for something specific?
- if (argc) {
- // If we didn't find anything, complain.
- if (!mtnext->mnt_fsname)
- bb_error_msg_and_die("can't find %s in %s",
- argv[0], fstabname);
- mtcur = mtnext;
- if (nonroot) {
- // fstab must have "users" or "user"
- if (!(parse_mount_options(mtcur->mnt_opts, 0) & MOUNT_USERS))
- bb_error_msg_and_die(must_be_root);
- }
- // Mount the last thing we found.
- mtcur->mnt_opts = xstrdup(mtcur->mnt_opts);
- append_mount_options(&(mtcur->mnt_opts), cmdopts);
- rc = singlemount(mtcur, 0);
- free(mtcur->mnt_opts);
- }
- goto clean_up;
- }
- /* If we're trying to mount something specific and this isn't it,
- * skip it. Note we must match both the exact text in fstab (ala
- * "proc") or a full path from root */
- if (argc) {
- // Is this what we're looking for?
- if (strcmp(argv[0], mtcur->mnt_fsname) &&
- strcmp(storage_path, mtcur->mnt_fsname) &&
- strcmp(argv[0], mtcur->mnt_dir) &&
- strcmp(storage_path, mtcur->mnt_dir)) continue;
- // Remember this entry. Something later may have overmounted
- // it, and we want the _last_ match.
- mtcur = mtnext;
- // If we're mounting all.
- } else {
- // Do we need to match a filesystem type?
- // TODO: support "-t type1,type2"; "-t notype1,type2"
- if (fstype && strcmp(mtcur->mnt_type, fstype)) continue;
- // Skip noauto and swap anyway.
- if (parse_mount_options(mtcur->mnt_opts, 0)
- & (MOUNT_NOAUTO | MOUNT_SWAP)) continue;
- // No, mount -a won't mount anything,
- // even user mounts, for mere humans.
- if (nonroot)
- bb_error_msg_and_die(must_be_root);
- // Mount this thing.
- // NFS mounts want this to be xrealloc-able
- mtcur->mnt_opts = xstrdup(mtcur->mnt_opts);
- if (singlemount(mtcur, 1)) {
- /* Count number of failed mounts */
- rc++;
- }
- free(mtcur->mnt_opts);
- }
- }
- if (ENABLE_FEATURE_CLEAN_UP) endmntent(fstab);
- clean_up:
- if (ENABLE_FEATURE_CLEAN_UP) {
- free(storage_path);
- free(cmdopts);
- }
- return rc;
- }
|