The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/contrib/openzfs/lib/libzfs/libzfs_mount.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * CDDL HEADER START
    3  *
    4  * The contents of this file are subject to the terms of the
    5  * Common Development and Distribution License (the "License").
    6  * You may not use this file except in compliance with the License.
    7  *
    8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
    9  * or https://opensource.org/licenses/CDDL-1.0.
   10  * See the License for the specific language governing permissions
   11  * and limitations under the License.
   12  *
   13  * When distributing Covered Code, include this CDDL HEADER in each
   14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
   15  * If applicable, add the following below this CDDL HEADER, with the
   16  * fields enclosed by brackets "[]" replaced with your own identifying
   17  * information: Portions Copyright [yyyy] [name of copyright owner]
   18  *
   19  * CDDL HEADER END
   20  */
   21 
   22 /*
   23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
   24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
   25  * Copyright (c) 2014, 2022 by Delphix. All rights reserved.
   26  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
   27  * Copyright 2017 RackTop Systems.
   28  * Copyright (c) 2018 Datto Inc.
   29  * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
   30  */
   31 
   32 /*
   33  * Routines to manage ZFS mounts.  We separate all the nasty routines that have
   34  * to deal with the OS.  The following functions are the main entry points --
   35  * they are used by mount and unmount and when changing a filesystem's
   36  * mountpoint.
   37  *
   38  *      zfs_is_mounted()
   39  *      zfs_mount()
   40  *      zfs_mount_at()
   41  *      zfs_unmount()
   42  *      zfs_unmountall()
   43  *
   44  * This file also contains the functions used to manage sharing filesystems:
   45  *
   46  *      zfs_is_shared()
   47  *      zfs_share()
   48  *      zfs_unshare()
   49  *      zfs_unshareall()
   50  *      zfs_commit_shares()
   51  *
   52  * The following functions are available for pool consumers, and will
   53  * mount/unmount and share/unshare all datasets within pool:
   54  *
   55  *      zpool_enable_datasets()
   56  *      zpool_disable_datasets()
   57  */
   58 
   59 #include <dirent.h>
   60 #include <dlfcn.h>
   61 #include <errno.h>
   62 #include <fcntl.h>
   63 #include <libgen.h>
   64 #include <libintl.h>
   65 #include <stdio.h>
   66 #include <stdlib.h>
   67 #include <string.h>
   68 #include <unistd.h>
   69 #include <zone.h>
   70 #include <sys/mntent.h>
   71 #include <sys/mount.h>
   72 #include <sys/stat.h>
   73 #include <sys/vfs.h>
   74 #include <sys/dsl_crypt.h>
   75 
   76 #include <libzfs.h>
   77 
   78 #include "libzfs_impl.h"
   79 #include <thread_pool.h>
   80 
   81 #include <libshare.h>
   82 #include <sys/systeminfo.h>
   83 #define MAXISALEN       257     /* based on sysinfo(2) man page */
   84 
   85 static int mount_tp_nthr = 512; /* tpool threads for multi-threaded mounting */
   86 
   87 static void zfs_mount_task(void *);
   88 
   89 static const proto_table_t proto_table[SA_PROTOCOL_COUNT] = {
   90         [SA_PROTOCOL_NFS] =
   91             {ZFS_PROP_SHARENFS, EZFS_SHARENFSFAILED, EZFS_UNSHARENFSFAILED},
   92         [SA_PROTOCOL_SMB] =
   93             {ZFS_PROP_SHARESMB, EZFS_SHARESMBFAILED, EZFS_UNSHARESMBFAILED},
   94 };
   95 
   96 static const enum sa_protocol share_all_proto[SA_PROTOCOL_COUNT + 1] = {
   97         SA_PROTOCOL_NFS,
   98         SA_PROTOCOL_SMB,
   99         SA_NO_PROTOCOL
  100 };
  101 
  102 
  103 
  104 static boolean_t
  105 dir_is_empty_stat(const char *dirname)
  106 {
  107         struct stat st;
  108 
  109         /*
  110          * We only want to return false if the given path is a non empty
  111          * directory, all other errors are handled elsewhere.
  112          */
  113         if (stat(dirname, &st) < 0 || !S_ISDIR(st.st_mode)) {
  114                 return (B_TRUE);
  115         }
  116 
  117         /*
  118          * An empty directory will still have two entries in it, one
  119          * entry for each of "." and "..".
  120          */
  121         if (st.st_size > 2) {
  122                 return (B_FALSE);
  123         }
  124 
  125         return (B_TRUE);
  126 }
  127 
  128 static boolean_t
  129 dir_is_empty_readdir(const char *dirname)
  130 {
  131         DIR *dirp;
  132         struct dirent64 *dp;
  133         int dirfd;
  134 
  135         if ((dirfd = openat(AT_FDCWD, dirname,
  136             O_RDONLY | O_NDELAY | O_LARGEFILE | O_CLOEXEC, 0)) < 0) {
  137                 return (B_TRUE);
  138         }
  139 
  140         if ((dirp = fdopendir(dirfd)) == NULL) {
  141                 (void) close(dirfd);
  142                 return (B_TRUE);
  143         }
  144 
  145         while ((dp = readdir64(dirp)) != NULL) {
  146 
  147                 if (strcmp(dp->d_name, ".") == 0 ||
  148                     strcmp(dp->d_name, "..") == 0)
  149                         continue;
  150 
  151                 (void) closedir(dirp);
  152                 return (B_FALSE);
  153         }
  154 
  155         (void) closedir(dirp);
  156         return (B_TRUE);
  157 }
  158 
  159 /*
  160  * Returns true if the specified directory is empty.  If we can't open the
  161  * directory at all, return true so that the mount can fail with a more
  162  * informative error message.
  163  */
  164 static boolean_t
  165 dir_is_empty(const char *dirname)
  166 {
  167         struct statfs64 st;
  168 
  169         /*
  170          * If the statvfs call fails or the filesystem is not a ZFS
  171          * filesystem, fall back to the slow path which uses readdir.
  172          */
  173         if ((statfs64(dirname, &st) != 0) ||
  174             (st.f_type != ZFS_SUPER_MAGIC)) {
  175                 return (dir_is_empty_readdir(dirname));
  176         }
  177 
  178         /*
  179          * At this point, we know the provided path is on a ZFS
  180          * filesystem, so we can use stat instead of readdir to
  181          * determine if the directory is empty or not. We try to avoid
  182          * using readdir because that requires opening "dirname"; this
  183          * open file descriptor can potentially end up in a child
  184          * process if there's a concurrent fork, thus preventing the
  185          * zfs_mount() from otherwise succeeding (the open file
  186          * descriptor inherited by the child process will cause the
  187          * parent's mount to fail with EBUSY). The performance
  188          * implications of replacing the open, read, and close with a
  189          * single stat is nice; but is not the main motivation for the
  190          * added complexity.
  191          */
  192         return (dir_is_empty_stat(dirname));
  193 }
  194 
  195 /*
  196  * Checks to see if the mount is active.  If the filesystem is mounted, we fill
  197  * in 'where' with the current mountpoint, and return 1.  Otherwise, we return
  198  * 0.
  199  */
  200 boolean_t
  201 is_mounted(libzfs_handle_t *zfs_hdl, const char *special, char **where)
  202 {
  203         struct mnttab entry;
  204 
  205         if (libzfs_mnttab_find(zfs_hdl, special, &entry) != 0)
  206                 return (B_FALSE);
  207 
  208         if (where != NULL)
  209                 *where = zfs_strdup(zfs_hdl, entry.mnt_mountp);
  210 
  211         return (B_TRUE);
  212 }
  213 
  214 boolean_t
  215 zfs_is_mounted(zfs_handle_t *zhp, char **where)
  216 {
  217         return (is_mounted(zhp->zfs_hdl, zfs_get_name(zhp), where));
  218 }
  219 
  220 /*
  221  * Checks any higher order concerns about whether the given dataset is
  222  * mountable, false otherwise.  zfs_is_mountable_internal specifically assumes
  223  * that the caller has verified the sanity of mounting the dataset at
  224  * its mountpoint to the extent the caller wants.
  225  */
  226 static boolean_t
  227 zfs_is_mountable_internal(zfs_handle_t *zhp)
  228 {
  229         if (zfs_prop_get_int(zhp, ZFS_PROP_ZONED) &&
  230             getzoneid() == GLOBAL_ZONEID)
  231                 return (B_FALSE);
  232 
  233         return (B_TRUE);
  234 }
  235 
  236 /*
  237  * Returns true if the given dataset is mountable, false otherwise.  Returns the
  238  * mountpoint in 'buf'.
  239  */
  240 static boolean_t
  241 zfs_is_mountable(zfs_handle_t *zhp, char *buf, size_t buflen,
  242     zprop_source_t *source, int flags)
  243 {
  244         char sourceloc[MAXNAMELEN];
  245         zprop_source_t sourcetype;
  246 
  247         if (!zfs_prop_valid_for_type(ZFS_PROP_MOUNTPOINT, zhp->zfs_type,
  248             B_FALSE))
  249                 return (B_FALSE);
  250 
  251         verify(zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, buf, buflen,
  252             &sourcetype, sourceloc, sizeof (sourceloc), B_FALSE) == 0);
  253 
  254         if (strcmp(buf, ZFS_MOUNTPOINT_NONE) == 0 ||
  255             strcmp(buf, ZFS_MOUNTPOINT_LEGACY) == 0)
  256                 return (B_FALSE);
  257 
  258         if (zfs_prop_get_int(zhp, ZFS_PROP_CANMOUNT) == ZFS_CANMOUNT_OFF)
  259                 return (B_FALSE);
  260 
  261         if (!zfs_is_mountable_internal(zhp))
  262                 return (B_FALSE);
  263 
  264         if (zfs_prop_get_int(zhp, ZFS_PROP_REDACTED) && !(flags & MS_FORCE))
  265                 return (B_FALSE);
  266 
  267         if (source)
  268                 *source = sourcetype;
  269 
  270         return (B_TRUE);
  271 }
  272 
  273 /*
  274  * The filesystem is mounted by invoking the system mount utility rather
  275  * than by the system call mount(2).  This ensures that the /etc/mtab
  276  * file is correctly locked for the update.  Performing our own locking
  277  * and /etc/mtab update requires making an unsafe assumption about how
  278  * the mount utility performs its locking.  Unfortunately, this also means
  279  * in the case of a mount failure we do not have the exact errno.  We must
  280  * make due with return value from the mount process.
  281  *
  282  * In the long term a shared library called libmount is under development
  283  * which provides a common API to address the locking and errno issues.
  284  * Once the standard mount utility has been updated to use this library
  285  * we can add an autoconf check to conditionally use it.
  286  *
  287  * http://www.kernel.org/pub/linux/utils/util-linux/libmount-docs/index.html
  288  */
  289 
  290 static int
  291 zfs_add_option(zfs_handle_t *zhp, char *options, int len,
  292     zfs_prop_t prop, const char *on, const char *off)
  293 {
  294         char *source;
  295         uint64_t value;
  296 
  297         /* Skip adding duplicate default options */
  298         if ((strstr(options, on) != NULL) || (strstr(options, off) != NULL))
  299                 return (0);
  300 
  301         /*
  302          * zfs_prop_get_int() is not used to ensure our mount options
  303          * are not influenced by the current /proc/self/mounts contents.
  304          */
  305         value = getprop_uint64(zhp, prop, &source);
  306 
  307         (void) strlcat(options, ",", len);
  308         (void) strlcat(options, value ? on : off, len);
  309 
  310         return (0);
  311 }
  312 
  313 static int
  314 zfs_add_options(zfs_handle_t *zhp, char *options, int len)
  315 {
  316         int error = 0;
  317 
  318         error = zfs_add_option(zhp, options, len,
  319             ZFS_PROP_ATIME, MNTOPT_ATIME, MNTOPT_NOATIME);
  320         /*
  321          * don't add relatime/strictatime when atime=off, otherwise strictatime
  322          * will force atime=on
  323          */
  324         if (strstr(options, MNTOPT_NOATIME) == NULL) {
  325                 error = zfs_add_option(zhp, options, len,
  326                     ZFS_PROP_RELATIME, MNTOPT_RELATIME, MNTOPT_STRICTATIME);
  327         }
  328         error = error ? error : zfs_add_option(zhp, options, len,
  329             ZFS_PROP_DEVICES, MNTOPT_DEVICES, MNTOPT_NODEVICES);
  330         error = error ? error : zfs_add_option(zhp, options, len,
  331             ZFS_PROP_EXEC, MNTOPT_EXEC, MNTOPT_NOEXEC);
  332         error = error ? error : zfs_add_option(zhp, options, len,
  333             ZFS_PROP_READONLY, MNTOPT_RO, MNTOPT_RW);
  334         error = error ? error : zfs_add_option(zhp, options, len,
  335             ZFS_PROP_SETUID, MNTOPT_SETUID, MNTOPT_NOSETUID);
  336         error = error ? error : zfs_add_option(zhp, options, len,
  337             ZFS_PROP_NBMAND, MNTOPT_NBMAND, MNTOPT_NONBMAND);
  338 
  339         return (error);
  340 }
  341 
  342 int
  343 zfs_mount(zfs_handle_t *zhp, const char *options, int flags)
  344 {
  345         char mountpoint[ZFS_MAXPROPLEN];
  346 
  347         if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL,
  348             flags))
  349                 return (0);
  350 
  351         return (zfs_mount_at(zhp, options, flags, mountpoint));
  352 }
  353 
  354 /*
  355  * Mount the given filesystem.
  356  */
  357 int
  358 zfs_mount_at(zfs_handle_t *zhp, const char *options, int flags,
  359     const char *mountpoint)
  360 {
  361         struct stat buf;
  362         char mntopts[MNT_LINE_MAX];
  363         char overlay[ZFS_MAXPROPLEN];
  364         char prop_encroot[MAXNAMELEN];
  365         boolean_t is_encroot;
  366         zfs_handle_t *encroot_hp = zhp;
  367         libzfs_handle_t *hdl = zhp->zfs_hdl;
  368         uint64_t keystatus;
  369         int remount = 0, rc;
  370 
  371         if (options == NULL) {
  372                 (void) strlcpy(mntopts, MNTOPT_DEFAULTS, sizeof (mntopts));
  373         } else {
  374                 (void) strlcpy(mntopts, options, sizeof (mntopts));
  375         }
  376 
  377         if (strstr(mntopts, MNTOPT_REMOUNT) != NULL)
  378                 remount = 1;
  379 
  380         /* Potentially duplicates some checks if invoked by zfs_mount(). */
  381         if (!zfs_is_mountable_internal(zhp))
  382                 return (0);
  383 
  384         /*
  385          * If the pool is imported read-only then all mounts must be read-only
  386          */
  387         if (zpool_get_prop_int(zhp->zpool_hdl, ZPOOL_PROP_READONLY, NULL))
  388                 (void) strlcat(mntopts, "," MNTOPT_RO, sizeof (mntopts));
  389 
  390         /*
  391          * Append default mount options which apply to the mount point.
  392          * This is done because under Linux (unlike Solaris) multiple mount
  393          * points may reference a single super block.  This means that just
  394          * given a super block there is no back reference to update the per
  395          * mount point options.
  396          */
  397         rc = zfs_add_options(zhp, mntopts, sizeof (mntopts));
  398         if (rc) {
  399                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  400                     "default options unavailable"));
  401                 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
  402                     dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
  403                     mountpoint));
  404         }
  405 
  406         /*
  407          * If the filesystem is encrypted the key must be loaded  in order to
  408          * mount. If the key isn't loaded, the MS_CRYPT flag decides whether
  409          * or not we attempt to load the keys. Note: we must call
  410          * zfs_refresh_properties() here since some callers of this function
  411          * (most notably zpool_enable_datasets()) may implicitly load our key
  412          * by loading the parent's key first.
  413          */
  414         if (zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION) != ZIO_CRYPT_OFF) {
  415                 zfs_refresh_properties(zhp);
  416                 keystatus = zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS);
  417 
  418                 /*
  419                  * If the key is unavailable and MS_CRYPT is set give the
  420                  * user a chance to enter the key. Otherwise just fail
  421                  * immediately.
  422                  */
  423                 if (keystatus == ZFS_KEYSTATUS_UNAVAILABLE) {
  424                         if (flags & MS_CRYPT) {
  425                                 rc = zfs_crypto_get_encryption_root(zhp,
  426                                     &is_encroot, prop_encroot);
  427                                 if (rc) {
  428                                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  429                                             "Failed to get encryption root for "
  430                                             "'%s'."), zfs_get_name(zhp));
  431                                         return (rc);
  432                                 }
  433 
  434                                 if (!is_encroot) {
  435                                         encroot_hp = zfs_open(hdl, prop_encroot,
  436                                             ZFS_TYPE_DATASET);
  437                                         if (encroot_hp == NULL)
  438                                                 return (hdl->libzfs_error);
  439                                 }
  440 
  441                                 rc = zfs_crypto_load_key(encroot_hp,
  442                                     B_FALSE, NULL);
  443 
  444                                 if (!is_encroot)
  445                                         zfs_close(encroot_hp);
  446                                 if (rc)
  447                                         return (rc);
  448                         } else {
  449                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  450                                     "encryption key not loaded"));
  451                                 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
  452                                     dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
  453                                     mountpoint));
  454                         }
  455                 }
  456 
  457         }
  458 
  459         /*
  460          * Append zfsutil option so the mount helper allow the mount
  461          */
  462         strlcat(mntopts, "," MNTOPT_ZFSUTIL, sizeof (mntopts));
  463 
  464         /* Create the directory if it doesn't already exist */
  465         if (lstat(mountpoint, &buf) != 0) {
  466                 if (mkdirp(mountpoint, 0755) != 0) {
  467                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  468                             "failed to create mountpoint: %s"),
  469                             strerror(errno));
  470                         return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
  471                             dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
  472                             mountpoint));
  473                 }
  474         }
  475 
  476         /*
  477          * Overlay mounts are enabled by default but may be disabled
  478          * via the 'overlay' property. The -O flag remains for compatibility.
  479          */
  480         if (!(flags & MS_OVERLAY)) {
  481                 if (zfs_prop_get(zhp, ZFS_PROP_OVERLAY, overlay,
  482                     sizeof (overlay), NULL, NULL, 0, B_FALSE) == 0) {
  483                         if (strcmp(overlay, "on") == 0) {
  484                                 flags |= MS_OVERLAY;
  485                         }
  486                 }
  487         }
  488 
  489         /*
  490          * Determine if the mountpoint is empty.  If so, refuse to perform the
  491          * mount.  We don't perform this check if 'remount' is
  492          * specified or if overlay option (-O) is given
  493          */
  494         if ((flags & MS_OVERLAY) == 0 && !remount &&
  495             !dir_is_empty(mountpoint)) {
  496                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  497                     "directory is not empty"));
  498                 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
  499                     dgettext(TEXT_DOMAIN, "cannot mount '%s'"), mountpoint));
  500         }
  501 
  502         /* perform the mount */
  503         rc = do_mount(zhp, mountpoint, mntopts, flags);
  504         if (rc) {
  505                 /*
  506                  * Generic errors are nasty, but there are just way too many
  507                  * from mount(), and they're well-understood.  We pick a few
  508                  * common ones to improve upon.
  509                  */
  510                 if (rc == EBUSY) {
  511                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  512                             "mountpoint or dataset is busy"));
  513                 } else if (rc == EPERM) {
  514                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  515                             "Insufficient privileges"));
  516                 } else if (rc == ENOTSUP) {
  517                         int spa_version;
  518 
  519                         VERIFY(zfs_spa_version(zhp, &spa_version) == 0);
  520                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
  521                             "Can't mount a version %llu "
  522                             "file system on a version %d pool. Pool must be"
  523                             " upgraded to mount this file system."),
  524                             (u_longlong_t)zfs_prop_get_int(zhp,
  525                             ZFS_PROP_VERSION), spa_version);
  526                 } else {
  527                         zfs_error_aux(hdl, "%s", strerror(rc));
  528                 }
  529                 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
  530                     dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
  531                     zhp->zfs_name));
  532         }
  533 
  534         /* remove the mounted entry before re-adding on remount */
  535         if (remount)
  536                 libzfs_mnttab_remove(hdl, zhp->zfs_name);
  537 
  538         /* add the mounted entry into our cache */
  539         libzfs_mnttab_add(hdl, zfs_get_name(zhp), mountpoint, mntopts);
  540         return (0);
  541 }
  542 
  543 /*
  544  * Unmount a single filesystem.
  545  */
  546 static int
  547 unmount_one(zfs_handle_t *zhp, const char *mountpoint, int flags)
  548 {
  549         int error;
  550 
  551         error = do_unmount(zhp, mountpoint, flags);
  552         if (error != 0) {
  553                 int libzfs_err;
  554 
  555                 switch (error) {
  556                 case EBUSY:
  557                         libzfs_err = EZFS_BUSY;
  558                         break;
  559                 case EIO:
  560                         libzfs_err = EZFS_IO;
  561                         break;
  562                 case ENOENT:
  563                         libzfs_err = EZFS_NOENT;
  564                         break;
  565                 case ENOMEM:
  566                         libzfs_err = EZFS_NOMEM;
  567                         break;
  568                 case EPERM:
  569                         libzfs_err = EZFS_PERM;
  570                         break;
  571                 default:
  572                         libzfs_err = EZFS_UMOUNTFAILED;
  573                 }
  574                 if (zhp) {
  575                         return (zfs_error_fmt(zhp->zfs_hdl, libzfs_err,
  576                             dgettext(TEXT_DOMAIN, "cannot unmount '%s'"),
  577                             mountpoint));
  578                 } else {
  579                         return (-1);
  580                 }
  581         }
  582 
  583         return (0);
  584 }
  585 
  586 /*
  587  * Unmount the given filesystem.
  588  */
  589 int
  590 zfs_unmount(zfs_handle_t *zhp, const char *mountpoint, int flags)
  591 {
  592         libzfs_handle_t *hdl = zhp->zfs_hdl;
  593         struct mnttab entry;
  594         char *mntpt = NULL;
  595         boolean_t encroot, unmounted = B_FALSE;
  596 
  597         /* check to see if we need to unmount the filesystem */
  598         if (mountpoint != NULL || ((zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) &&
  599             libzfs_mnttab_find(hdl, zhp->zfs_name, &entry) == 0)) {
  600                 /*
  601                  * mountpoint may have come from a call to
  602                  * getmnt/getmntany if it isn't NULL. If it is NULL,
  603                  * we know it comes from libzfs_mnttab_find which can
  604                  * then get freed later. We strdup it to play it safe.
  605                  */
  606                 if (mountpoint == NULL)
  607                         mntpt = zfs_strdup(hdl, entry.mnt_mountp);
  608                 else
  609                         mntpt = zfs_strdup(hdl, mountpoint);
  610 
  611                 /*
  612                  * Unshare and unmount the filesystem
  613                  */
  614                 if (zfs_unshare(zhp, mntpt, share_all_proto) != 0) {
  615                         free(mntpt);
  616                         return (-1);
  617                 }
  618                 zfs_commit_shares(NULL);
  619 
  620                 if (unmount_one(zhp, mntpt, flags) != 0) {
  621                         free(mntpt);
  622                         (void) zfs_share(zhp, NULL);
  623                         zfs_commit_shares(NULL);
  624                         return (-1);
  625                 }
  626 
  627                 libzfs_mnttab_remove(hdl, zhp->zfs_name);
  628                 free(mntpt);
  629                 unmounted = B_TRUE;
  630         }
  631 
  632         /*
  633          * If the MS_CRYPT flag is provided we must ensure we attempt to
  634          * unload the dataset's key regardless of whether we did any work
  635          * to unmount it. We only do this for encryption roots.
  636          */
  637         if ((flags & MS_CRYPT) != 0 &&
  638             zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION) != ZIO_CRYPT_OFF) {
  639                 zfs_refresh_properties(zhp);
  640 
  641                 if (zfs_crypto_get_encryption_root(zhp, &encroot, NULL) != 0 &&
  642                     unmounted) {
  643                         (void) zfs_mount(zhp, NULL, 0);
  644                         return (-1);
  645                 }
  646 
  647                 if (encroot && zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) ==
  648                     ZFS_KEYSTATUS_AVAILABLE &&
  649                     zfs_crypto_unload_key(zhp) != 0) {
  650                         (void) zfs_mount(zhp, NULL, 0);
  651                         return (-1);
  652                 }
  653         }
  654 
  655         zpool_disable_volume_os(zhp->zfs_name);
  656 
  657         return (0);
  658 }
  659 
  660 /*
  661  * Unmount this filesystem and any children inheriting the mountpoint property.
  662  * To do this, just act like we're changing the mountpoint property, but don't
  663  * remount the filesystems afterwards.
  664  */
  665 int
  666 zfs_unmountall(zfs_handle_t *zhp, int flags)
  667 {
  668         prop_changelist_t *clp;
  669         int ret;
  670 
  671         clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT,
  672             CL_GATHER_ITER_MOUNTED, flags);
  673         if (clp == NULL)
  674                 return (-1);
  675 
  676         ret = changelist_prefix(clp);
  677         changelist_free(clp);
  678 
  679         return (ret);
  680 }
  681 
  682 /*
  683  * Unshare a filesystem by mountpoint.
  684  */
  685 static int
  686 unshare_one(libzfs_handle_t *hdl, const char *name, const char *mountpoint,
  687     enum sa_protocol proto)
  688 {
  689         int err = sa_disable_share(mountpoint, proto);
  690         if (err != SA_OK)
  691                 return (zfs_error_fmt(hdl, proto_table[proto].p_unshare_err,
  692                     dgettext(TEXT_DOMAIN, "cannot unshare '%s': %s"),
  693                     name, sa_errorstr(err)));
  694 
  695         return (0);
  696 }
  697 
  698 /*
  699  * Share the given filesystem according to the options in the specified
  700  * protocol specific properties (sharenfs, sharesmb).  We rely
  701  * on "libshare" to do the dirty work for us.
  702  */
  703 int
  704 zfs_share(zfs_handle_t *zhp, const enum sa_protocol *proto)
  705 {
  706         char mountpoint[ZFS_MAXPROPLEN];
  707         char shareopts[ZFS_MAXPROPLEN];
  708         char sourcestr[ZFS_MAXPROPLEN];
  709         const enum sa_protocol *curr_proto;
  710         zprop_source_t sourcetype;
  711         int err = 0;
  712 
  713         if (proto == NULL)
  714                 proto = share_all_proto;
  715 
  716         if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL, 0))
  717                 return (0);
  718 
  719         for (curr_proto = proto; *curr_proto != SA_NO_PROTOCOL; curr_proto++) {
  720                 /*
  721                  * Return success if there are no share options.
  722                  */
  723                 if (zfs_prop_get(zhp, proto_table[*curr_proto].p_prop,
  724                     shareopts, sizeof (shareopts), &sourcetype, sourcestr,
  725                     ZFS_MAXPROPLEN, B_FALSE) != 0 ||
  726                     strcmp(shareopts, "off") == 0)
  727                         continue;
  728 
  729                 /*
  730                  * If the 'zoned' property is set, then zfs_is_mountable()
  731                  * will have already bailed out if we are in the global zone.
  732                  * But local zones cannot be NFS servers, so we ignore it for
  733                  * local zones as well.
  734                  */
  735                 if (zfs_prop_get_int(zhp, ZFS_PROP_ZONED))
  736                         continue;
  737 
  738                 err = sa_enable_share(zfs_get_name(zhp), mountpoint, shareopts,
  739                     *curr_proto);
  740                 if (err != SA_OK) {
  741                         return (zfs_error_fmt(zhp->zfs_hdl,
  742                             proto_table[*curr_proto].p_share_err,
  743                             dgettext(TEXT_DOMAIN, "cannot share '%s: %s'"),
  744                             zfs_get_name(zhp), sa_errorstr(err)));
  745                 }
  746 
  747         }
  748         return (0);
  749 }
  750 
  751 /*
  752  * Check to see if the filesystem is currently shared.
  753  */
  754 boolean_t
  755 zfs_is_shared(zfs_handle_t *zhp, char **where,
  756     const enum sa_protocol *proto)
  757 {
  758         char *mountpoint;
  759         if (proto == NULL)
  760                 proto = share_all_proto;
  761 
  762         if (ZFS_IS_VOLUME(zhp))
  763                 return (B_FALSE);
  764 
  765         if (!zfs_is_mounted(zhp, &mountpoint))
  766                 return (B_FALSE);
  767 
  768         for (const enum sa_protocol *p = proto; *p != SA_NO_PROTOCOL; ++p)
  769                 if (sa_is_shared(mountpoint, *p)) {
  770                         if (where != NULL)
  771                                 *where = mountpoint;
  772                         else
  773                                 free(mountpoint);
  774                         return (B_TRUE);
  775                 }
  776 
  777         free(mountpoint);
  778         return (B_FALSE);
  779 }
  780 
  781 void
  782 zfs_commit_shares(const enum sa_protocol *proto)
  783 {
  784         if (proto == NULL)
  785                 proto = share_all_proto;
  786 
  787         for (const enum sa_protocol *p = proto; *p != SA_NO_PROTOCOL; ++p)
  788                 sa_commit_shares(*p);
  789 }
  790 
  791 void
  792 zfs_truncate_shares(const enum sa_protocol *proto)
  793 {
  794         if (proto == NULL)
  795                 proto = share_all_proto;
  796 
  797         for (const enum sa_protocol *p = proto; *p != SA_NO_PROTOCOL; ++p)
  798                 sa_truncate_shares(*p);
  799 }
  800 
  801 /*
  802  * Unshare the given filesystem.
  803  */
  804 int
  805 zfs_unshare(zfs_handle_t *zhp, const char *mountpoint,
  806     const enum sa_protocol *proto)
  807 {
  808         libzfs_handle_t *hdl = zhp->zfs_hdl;
  809         struct mnttab entry;
  810 
  811         if (proto == NULL)
  812                 proto = share_all_proto;
  813 
  814         if (mountpoint != NULL || ((zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) &&
  815             libzfs_mnttab_find(hdl, zfs_get_name(zhp), &entry) == 0)) {
  816 
  817                 /* check to see if need to unmount the filesystem */
  818                 const char *mntpt = mountpoint ?: entry.mnt_mountp;
  819 
  820                 for (const enum sa_protocol *curr_proto = proto;
  821                     *curr_proto != SA_NO_PROTOCOL; curr_proto++)
  822                         if (sa_is_shared(mntpt, *curr_proto) &&
  823                             unshare_one(hdl, zhp->zfs_name,
  824                             mntpt, *curr_proto) != 0)
  825                                         return (-1);
  826         }
  827 
  828         return (0);
  829 }
  830 
  831 /*
  832  * Same as zfs_unmountall(), but for NFS and SMB unshares.
  833  */
  834 int
  835 zfs_unshareall(zfs_handle_t *zhp, const enum sa_protocol *proto)
  836 {
  837         prop_changelist_t *clp;
  838         int ret;
  839 
  840         if (proto == NULL)
  841                 proto = share_all_proto;
  842 
  843         clp = changelist_gather(zhp, ZFS_PROP_SHARENFS, 0, 0);
  844         if (clp == NULL)
  845                 return (-1);
  846 
  847         ret = changelist_unshare(clp, proto);
  848         changelist_free(clp);
  849 
  850         return (ret);
  851 }
  852 
  853 /*
  854  * Remove the mountpoint associated with the current dataset, if necessary.
  855  * We only remove the underlying directory if:
  856  *
  857  *      - The mountpoint is not 'none' or 'legacy'
  858  *      - The mountpoint is non-empty
  859  *      - The mountpoint is the default or inherited
  860  *      - The 'zoned' property is set, or we're in a local zone
  861  *
  862  * Any other directories we leave alone.
  863  */
  864 void
  865 remove_mountpoint(zfs_handle_t *zhp)
  866 {
  867         char mountpoint[ZFS_MAXPROPLEN];
  868         zprop_source_t source;
  869 
  870         if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint),
  871             &source, 0))
  872                 return;
  873 
  874         if (source == ZPROP_SRC_DEFAULT ||
  875             source == ZPROP_SRC_INHERITED) {
  876                 /*
  877                  * Try to remove the directory, silently ignoring any errors.
  878                  * The filesystem may have since been removed or moved around,
  879                  * and this error isn't really useful to the administrator in
  880                  * any way.
  881                  */
  882                 (void) rmdir(mountpoint);
  883         }
  884 }
  885 
  886 /*
  887  * Add the given zfs handle to the cb_handles array, dynamically reallocating
  888  * the array if it is out of space.
  889  */
  890 void
  891 libzfs_add_handle(get_all_cb_t *cbp, zfs_handle_t *zhp)
  892 {
  893         if (cbp->cb_alloc == cbp->cb_used) {
  894                 size_t newsz;
  895                 zfs_handle_t **newhandles;
  896 
  897                 newsz = cbp->cb_alloc != 0 ? cbp->cb_alloc * 2 : 64;
  898                 newhandles = zfs_realloc(zhp->zfs_hdl,
  899                     cbp->cb_handles, cbp->cb_alloc * sizeof (zfs_handle_t *),
  900                     newsz * sizeof (zfs_handle_t *));
  901                 cbp->cb_handles = newhandles;
  902                 cbp->cb_alloc = newsz;
  903         }
  904         cbp->cb_handles[cbp->cb_used++] = zhp;
  905 }
  906 
  907 /*
  908  * Recursive helper function used during file system enumeration
  909  */
  910 static int
  911 zfs_iter_cb(zfs_handle_t *zhp, void *data)
  912 {
  913         get_all_cb_t *cbp = data;
  914 
  915         if (!(zfs_get_type(zhp) & ZFS_TYPE_FILESYSTEM)) {
  916                 zfs_close(zhp);
  917                 return (0);
  918         }
  919 
  920         if (zfs_prop_get_int(zhp, ZFS_PROP_CANMOUNT) == ZFS_CANMOUNT_NOAUTO) {
  921                 zfs_close(zhp);
  922                 return (0);
  923         }
  924 
  925         if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) ==
  926             ZFS_KEYSTATUS_UNAVAILABLE) {
  927                 zfs_close(zhp);
  928                 return (0);
  929         }
  930 
  931         /*
  932          * If this filesystem is inconsistent and has a receive resume
  933          * token, we can not mount it.
  934          */
  935         if (zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT) &&
  936             zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
  937             NULL, 0, NULL, NULL, 0, B_TRUE) == 0) {
  938                 zfs_close(zhp);
  939                 return (0);
  940         }
  941 
  942         libzfs_add_handle(cbp, zhp);
  943         if (zfs_iter_filesystems(zhp, 0, zfs_iter_cb, cbp) != 0) {
  944                 zfs_close(zhp);
  945                 return (-1);
  946         }
  947         return (0);
  948 }
  949 
  950 /*
  951  * Sort comparator that compares two mountpoint paths. We sort these paths so
  952  * that subdirectories immediately follow their parents. This means that we
  953  * effectively treat the '/' character as the lowest value non-nul char.
  954  * Since filesystems from non-global zones can have the same mountpoint
  955  * as other filesystems, the comparator sorts global zone filesystems to
  956  * the top of the list. This means that the global zone will traverse the
  957  * filesystem list in the correct order and can stop when it sees the
  958  * first zoned filesystem. In a non-global zone, only the delegated
  959  * filesystems are seen.
  960  *
  961  * An example sorted list using this comparator would look like:
  962  *
  963  * /foo
  964  * /foo/bar
  965  * /foo/bar/baz
  966  * /foo/baz
  967  * /foo.bar
  968  * /foo (NGZ1)
  969  * /foo (NGZ2)
  970  *
  971  * The mounting code depends on this ordering to deterministically iterate
  972  * over filesystems in order to spawn parallel mount tasks.
  973  */
  974 static int
  975 mountpoint_cmp(const void *arga, const void *argb)
  976 {
  977         zfs_handle_t *const *zap = arga;
  978         zfs_handle_t *za = *zap;
  979         zfs_handle_t *const *zbp = argb;
  980         zfs_handle_t *zb = *zbp;
  981         char mounta[MAXPATHLEN];
  982         char mountb[MAXPATHLEN];
  983         const char *a = mounta;
  984         const char *b = mountb;
  985         boolean_t gota, gotb;
  986         uint64_t zoneda, zonedb;
  987 
  988         zoneda = zfs_prop_get_int(za, ZFS_PROP_ZONED);
  989         zonedb = zfs_prop_get_int(zb, ZFS_PROP_ZONED);
  990         if (zoneda && !zonedb)
  991                 return (1);
  992         if (!zoneda && zonedb)
  993                 return (-1);
  994 
  995         gota = (zfs_get_type(za) == ZFS_TYPE_FILESYSTEM);
  996         if (gota) {
  997                 verify(zfs_prop_get(za, ZFS_PROP_MOUNTPOINT, mounta,
  998                     sizeof (mounta), NULL, NULL, 0, B_FALSE) == 0);
  999         }
 1000         gotb = (zfs_get_type(zb) == ZFS_TYPE_FILESYSTEM);
 1001         if (gotb) {
 1002                 verify(zfs_prop_get(zb, ZFS_PROP_MOUNTPOINT, mountb,
 1003                     sizeof (mountb), NULL, NULL, 0, B_FALSE) == 0);
 1004         }
 1005 
 1006         if (gota && gotb) {
 1007                 while (*a != '\0' && (*a == *b)) {
 1008                         a++;
 1009                         b++;
 1010                 }
 1011                 if (*a == *b)
 1012                         return (0);
 1013                 if (*a == '\0')
 1014                         return (-1);
 1015                 if (*b == '\0')
 1016                         return (1);
 1017                 if (*a == '/')
 1018                         return (-1);
 1019                 if (*b == '/')
 1020                         return (1);
 1021                 return (*a < *b ? -1 : *a > *b);
 1022         }
 1023 
 1024         if (gota)
 1025                 return (-1);
 1026         if (gotb)
 1027                 return (1);
 1028 
 1029         /*
 1030          * If neither filesystem has a mountpoint, revert to sorting by
 1031          * dataset name.
 1032          */
 1033         return (strcmp(zfs_get_name(za), zfs_get_name(zb)));
 1034 }
 1035 
 1036 /*
 1037  * Return true if path2 is a child of path1 or path2 equals path1 or
 1038  * path1 is "/" (path2 is always a child of "/").
 1039  */
 1040 static boolean_t
 1041 libzfs_path_contains(const char *path1, const char *path2)
 1042 {
 1043         return (strcmp(path1, path2) == 0 || strcmp(path1, "/") == 0 ||
 1044             (strstr(path2, path1) == path2 && path2[strlen(path1)] == '/'));
 1045 }
 1046 
 1047 /*
 1048  * Given a mountpoint specified by idx in the handles array, find the first
 1049  * non-descendent of that mountpoint and return its index. Descendant paths
 1050  * start with the parent's path. This function relies on the ordering
 1051  * enforced by mountpoint_cmp().
 1052  */
 1053 static int
 1054 non_descendant_idx(zfs_handle_t **handles, size_t num_handles, int idx)
 1055 {
 1056         char parent[ZFS_MAXPROPLEN];
 1057         char child[ZFS_MAXPROPLEN];
 1058         int i;
 1059 
 1060         verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, parent,
 1061             sizeof (parent), NULL, NULL, 0, B_FALSE) == 0);
 1062 
 1063         for (i = idx + 1; i < num_handles; i++) {
 1064                 verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT, child,
 1065                     sizeof (child), NULL, NULL, 0, B_FALSE) == 0);
 1066                 if (!libzfs_path_contains(parent, child))
 1067                         break;
 1068         }
 1069         return (i);
 1070 }
 1071 
 1072 typedef struct mnt_param {
 1073         libzfs_handle_t *mnt_hdl;
 1074         tpool_t         *mnt_tp;
 1075         zfs_handle_t    **mnt_zhps; /* filesystems to mount */
 1076         size_t          mnt_num_handles;
 1077         int             mnt_idx;        /* Index of selected entry to mount */
 1078         zfs_iter_f      mnt_func;
 1079         void            *mnt_data;
 1080 } mnt_param_t;
 1081 
 1082 /*
 1083  * Allocate and populate the parameter struct for mount function, and
 1084  * schedule mounting of the entry selected by idx.
 1085  */
 1086 static void
 1087 zfs_dispatch_mount(libzfs_handle_t *hdl, zfs_handle_t **handles,
 1088     size_t num_handles, int idx, zfs_iter_f func, void *data, tpool_t *tp)
 1089 {
 1090         mnt_param_t *mnt_param = zfs_alloc(hdl, sizeof (mnt_param_t));
 1091 
 1092         mnt_param->mnt_hdl = hdl;
 1093         mnt_param->mnt_tp = tp;
 1094         mnt_param->mnt_zhps = handles;
 1095         mnt_param->mnt_num_handles = num_handles;
 1096         mnt_param->mnt_idx = idx;
 1097         mnt_param->mnt_func = func;
 1098         mnt_param->mnt_data = data;
 1099 
 1100         (void) tpool_dispatch(tp, zfs_mount_task, (void*)mnt_param);
 1101 }
 1102 
 1103 /*
 1104  * This is the structure used to keep state of mounting or sharing operations
 1105  * during a call to zpool_enable_datasets().
 1106  */
 1107 typedef struct mount_state {
 1108         /*
 1109          * ms_mntstatus is set to -1 if any mount fails. While multiple threads
 1110          * could update this variable concurrently, no synchronization is
 1111          * needed as it's only ever set to -1.
 1112          */
 1113         int             ms_mntstatus;
 1114         int             ms_mntflags;
 1115         const char      *ms_mntopts;
 1116 } mount_state_t;
 1117 
 1118 static int
 1119 zfs_mount_one(zfs_handle_t *zhp, void *arg)
 1120 {
 1121         mount_state_t *ms = arg;
 1122         int ret = 0;
 1123 
 1124         /*
 1125          * don't attempt to mount encrypted datasets with
 1126          * unloaded keys
 1127          */
 1128         if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) ==
 1129             ZFS_KEYSTATUS_UNAVAILABLE)
 1130                 return (0);
 1131 
 1132         if (zfs_mount(zhp, ms->ms_mntopts, ms->ms_mntflags) != 0)
 1133                 ret = ms->ms_mntstatus = -1;
 1134         return (ret);
 1135 }
 1136 
 1137 static int
 1138 zfs_share_one(zfs_handle_t *zhp, void *arg)
 1139 {
 1140         mount_state_t *ms = arg;
 1141         int ret = 0;
 1142 
 1143         if (zfs_share(zhp, NULL) != 0)
 1144                 ret = ms->ms_mntstatus = -1;
 1145         return (ret);
 1146 }
 1147 
 1148 /*
 1149  * Thread pool function to mount one file system. On completion, it finds and
 1150  * schedules its children to be mounted. This depends on the sorting done in
 1151  * zfs_foreach_mountpoint(). Note that the degenerate case (chain of entries
 1152  * each descending from the previous) will have no parallelism since we always
 1153  * have to wait for the parent to finish mounting before we can schedule
 1154  * its children.
 1155  */
 1156 static void
 1157 zfs_mount_task(void *arg)
 1158 {
 1159         mnt_param_t *mp = arg;
 1160         int idx = mp->mnt_idx;
 1161         zfs_handle_t **handles = mp->mnt_zhps;
 1162         size_t num_handles = mp->mnt_num_handles;
 1163         char mountpoint[ZFS_MAXPROPLEN];
 1164 
 1165         verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, mountpoint,
 1166             sizeof (mountpoint), NULL, NULL, 0, B_FALSE) == 0);
 1167 
 1168         if (mp->mnt_func(handles[idx], mp->mnt_data) != 0)
 1169                 goto out;
 1170 
 1171         /*
 1172          * We dispatch tasks to mount filesystems with mountpoints underneath
 1173          * this one. We do this by dispatching the next filesystem with a
 1174          * descendant mountpoint of the one we just mounted, then skip all of
 1175          * its descendants, dispatch the next descendant mountpoint, and so on.
 1176          * The non_descendant_idx() function skips over filesystems that are
 1177          * descendants of the filesystem we just dispatched.
 1178          */
 1179         for (int i = idx + 1; i < num_handles;
 1180             i = non_descendant_idx(handles, num_handles, i)) {
 1181                 char child[ZFS_MAXPROPLEN];
 1182                 verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT,
 1183                     child, sizeof (child), NULL, NULL, 0, B_FALSE) == 0);
 1184 
 1185                 if (!libzfs_path_contains(mountpoint, child))
 1186                         break; /* not a descendant, return */
 1187                 zfs_dispatch_mount(mp->mnt_hdl, handles, num_handles, i,
 1188                     mp->mnt_func, mp->mnt_data, mp->mnt_tp);
 1189         }
 1190 
 1191 out:
 1192         free(mp);
 1193 }
 1194 
 1195 /*
 1196  * Issue the func callback for each ZFS handle contained in the handles
 1197  * array. This function is used to mount all datasets, and so this function
 1198  * guarantees that filesystems for parent mountpoints are called before their
 1199  * children. As such, before issuing any callbacks, we first sort the array
 1200  * of handles by mountpoint.
 1201  *
 1202  * Callbacks are issued in one of two ways:
 1203  *
 1204  * 1. Sequentially: If the parallel argument is B_FALSE or the ZFS_SERIAL_MOUNT
 1205  *    environment variable is set, then we issue callbacks sequentially.
 1206  *
 1207  * 2. In parallel: If the parallel argument is B_TRUE and the ZFS_SERIAL_MOUNT
 1208  *    environment variable is not set, then we use a tpool to dispatch threads
 1209  *    to mount filesystems in parallel. This function dispatches tasks to mount
 1210  *    the filesystems at the top-level mountpoints, and these tasks in turn
 1211  *    are responsible for recursively mounting filesystems in their children
 1212  *    mountpoints.
 1213  */
 1214 void
 1215 zfs_foreach_mountpoint(libzfs_handle_t *hdl, zfs_handle_t **handles,
 1216     size_t num_handles, zfs_iter_f func, void *data, boolean_t parallel)
 1217 {
 1218         zoneid_t zoneid = getzoneid();
 1219 
 1220         /*
 1221          * The ZFS_SERIAL_MOUNT environment variable is an undocumented
 1222          * variable that can be used as a convenience to do a/b comparison
 1223          * of serial vs. parallel mounting.
 1224          */
 1225         boolean_t serial_mount = !parallel ||
 1226             (getenv("ZFS_SERIAL_MOUNT") != NULL);
 1227 
 1228         /*
 1229          * Sort the datasets by mountpoint. See mountpoint_cmp for details
 1230          * of how these are sorted.
 1231          */
 1232         qsort(handles, num_handles, sizeof (zfs_handle_t *), mountpoint_cmp);
 1233 
 1234         if (serial_mount) {
 1235                 for (int i = 0; i < num_handles; i++) {
 1236                         func(handles[i], data);
 1237                 }
 1238                 return;
 1239         }
 1240 
 1241         /*
 1242          * Issue the callback function for each dataset using a parallel
 1243          * algorithm that uses a thread pool to manage threads.
 1244          */
 1245         tpool_t *tp = tpool_create(1, mount_tp_nthr, 0, NULL);
 1246 
 1247         /*
 1248          * There may be multiple "top level" mountpoints outside of the pool's
 1249          * root mountpoint, e.g.: /foo /bar. Dispatch a mount task for each of
 1250          * these.
 1251          */
 1252         for (int i = 0; i < num_handles;
 1253             i = non_descendant_idx(handles, num_handles, i)) {
 1254                 /*
 1255                  * Since the mountpoints have been sorted so that the zoned
 1256                  * filesystems are at the end, a zoned filesystem seen from
 1257                  * the global zone means that we're done.
 1258                  */
 1259                 if (zoneid == GLOBAL_ZONEID &&
 1260                     zfs_prop_get_int(handles[i], ZFS_PROP_ZONED))
 1261                         break;
 1262                 zfs_dispatch_mount(hdl, handles, num_handles, i, func, data,
 1263                     tp);
 1264         }
 1265 
 1266         tpool_wait(tp); /* wait for all scheduled mounts to complete */
 1267         tpool_destroy(tp);
 1268 }
 1269 
 1270 /*
 1271  * Mount and share all datasets within the given pool.  This assumes that no
 1272  * datasets within the pool are currently mounted.
 1273  */
 1274 int
 1275 zpool_enable_datasets(zpool_handle_t *zhp, const char *mntopts, int flags)
 1276 {
 1277         get_all_cb_t cb = { 0 };
 1278         mount_state_t ms = { 0 };
 1279         zfs_handle_t *zfsp;
 1280         int ret = 0;
 1281 
 1282         if ((zfsp = zfs_open(zhp->zpool_hdl, zhp->zpool_name,
 1283             ZFS_TYPE_DATASET)) == NULL)
 1284                 goto out;
 1285 
 1286         /*
 1287          * Gather all non-snapshot datasets within the pool. Start by adding
 1288          * the root filesystem for this pool to the list, and then iterate
 1289          * over all child filesystems.
 1290          */
 1291         libzfs_add_handle(&cb, zfsp);
 1292         if (zfs_iter_filesystems(zfsp, 0, zfs_iter_cb, &cb) != 0)
 1293                 goto out;
 1294 
 1295         /*
 1296          * Mount all filesystems
 1297          */
 1298         ms.ms_mntopts = mntopts;
 1299         ms.ms_mntflags = flags;
 1300         zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used,
 1301             zfs_mount_one, &ms, B_TRUE);
 1302         if (ms.ms_mntstatus != 0)
 1303                 ret = ms.ms_mntstatus;
 1304 
 1305         /*
 1306          * Share all filesystems that need to be shared. This needs to be
 1307          * a separate pass because libshare is not mt-safe, and so we need
 1308          * to share serially.
 1309          */
 1310         ms.ms_mntstatus = 0;
 1311         zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used,
 1312             zfs_share_one, &ms, B_FALSE);
 1313         if (ms.ms_mntstatus != 0)
 1314                 ret = ms.ms_mntstatus;
 1315         else
 1316                 zfs_commit_shares(NULL);
 1317 
 1318 out:
 1319         for (int i = 0; i < cb.cb_used; i++)
 1320                 zfs_close(cb.cb_handles[i]);
 1321         free(cb.cb_handles);
 1322 
 1323         return (ret);
 1324 }
 1325 
 1326 struct sets_s {
 1327         char *mountpoint;
 1328         zfs_handle_t *dataset;
 1329 };
 1330 
 1331 static int
 1332 mountpoint_compare(const void *a, const void *b)
 1333 {
 1334         const struct sets_s *mounta = (struct sets_s *)a;
 1335         const struct sets_s *mountb = (struct sets_s *)b;
 1336 
 1337         return (strcmp(mountb->mountpoint, mounta->mountpoint));
 1338 }
 1339 
 1340 /*
 1341  * Unshare and unmount all datasets within the given pool.  We don't want to
 1342  * rely on traversing the DSL to discover the filesystems within the pool,
 1343  * because this may be expensive (if not all of them are mounted), and can fail
 1344  * arbitrarily (on I/O error, for example).  Instead, we walk /proc/self/mounts
 1345  * and gather all the filesystems that are currently mounted.
 1346  */
 1347 int
 1348 zpool_disable_datasets(zpool_handle_t *zhp, boolean_t force)
 1349 {
 1350         int used, alloc;
 1351         FILE *mnttab;
 1352         struct mnttab entry;
 1353         size_t namelen;
 1354         struct sets_s *sets = NULL;
 1355         libzfs_handle_t *hdl = zhp->zpool_hdl;
 1356         int i;
 1357         int ret = -1;
 1358         int flags = (force ? MS_FORCE : 0);
 1359 
 1360         namelen = strlen(zhp->zpool_name);
 1361 
 1362         if ((mnttab = fopen(MNTTAB, "re")) == NULL)
 1363                 return (ENOENT);
 1364 
 1365         used = alloc = 0;
 1366         while (getmntent(mnttab, &entry) == 0) {
 1367                 /*
 1368                  * Ignore non-ZFS entries.
 1369                  */
 1370                 if (entry.mnt_fstype == NULL ||
 1371                     strcmp(entry.mnt_fstype, MNTTYPE_ZFS) != 0)
 1372                         continue;
 1373 
 1374                 /*
 1375                  * Ignore filesystems not within this pool.
 1376                  */
 1377                 if (entry.mnt_mountp == NULL ||
 1378                     strncmp(entry.mnt_special, zhp->zpool_name, namelen) != 0 ||
 1379                     (entry.mnt_special[namelen] != '/' &&
 1380                     entry.mnt_special[namelen] != '\0'))
 1381                         continue;
 1382 
 1383                 /*
 1384                  * At this point we've found a filesystem within our pool.  Add
 1385                  * it to our growing list.
 1386                  */
 1387                 if (used == alloc) {
 1388                         if (alloc == 0) {
 1389                                 sets = zfs_alloc(hdl,
 1390                                     8 * sizeof (struct sets_s));
 1391                                 alloc = 8;
 1392                         } else {
 1393                                 sets = zfs_realloc(hdl, sets,
 1394                                     alloc * sizeof (struct sets_s),
 1395                                     alloc * 2 * sizeof (struct sets_s));
 1396 
 1397                                 alloc *= 2;
 1398                         }
 1399                 }
 1400 
 1401                 sets[used].mountpoint = zfs_strdup(hdl, entry.mnt_mountp);
 1402 
 1403                 /*
 1404                  * This is allowed to fail, in case there is some I/O error.  It
 1405                  * is only used to determine if we need to remove the underlying
 1406                  * mountpoint, so failure is not fatal.
 1407                  */
 1408                 sets[used].dataset = make_dataset_handle(hdl,
 1409                     entry.mnt_special);
 1410 
 1411                 used++;
 1412         }
 1413 
 1414         /*
 1415          * At this point, we have the entire list of filesystems, so sort it by
 1416          * mountpoint.
 1417          */
 1418         if (used != 0)
 1419                 qsort(sets, used, sizeof (struct sets_s), mountpoint_compare);
 1420 
 1421         /*
 1422          * Walk through and first unshare everything.
 1423          */
 1424         for (i = 0; i < used; i++) {
 1425                 for (enum sa_protocol i = 0; i < SA_PROTOCOL_COUNT; ++i) {
 1426                         if (sa_is_shared(sets[i].mountpoint, i) &&
 1427                             unshare_one(hdl, sets[i].mountpoint,
 1428                             sets[i].mountpoint, i) != 0)
 1429                                 goto out;
 1430                 }
 1431         }
 1432         zfs_commit_shares(NULL);
 1433 
 1434         /*
 1435          * Now unmount everything, removing the underlying directories as
 1436          * appropriate.
 1437          */
 1438         for (i = 0; i < used; i++) {
 1439                 if (unmount_one(sets[i].dataset, sets[i].mountpoint,
 1440                     flags) != 0)
 1441                         goto out;
 1442         }
 1443 
 1444         for (i = 0; i < used; i++) {
 1445                 if (sets[i].dataset)
 1446                         remove_mountpoint(sets[i].dataset);
 1447         }
 1448 
 1449         zpool_disable_datasets_os(zhp, force);
 1450 
 1451         ret = 0;
 1452 out:
 1453         (void) fclose(mnttab);
 1454         for (i = 0; i < used; i++) {
 1455                 if (sets[i].dataset)
 1456                         zfs_close(sets[i].dataset);
 1457                 free(sets[i].mountpoint);
 1458         }
 1459         free(sets);
 1460 
 1461         return (ret);
 1462 }

Cache object: 2f6098e0854a8d0c0994d7b0302a0d35


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.