Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 #include <linux/fanotify.h>
0003 #include <linux/fdtable.h>
0004 #include <linux/fsnotify_backend.h>
0005 #include <linux/init.h>
0006 #include <linux/jiffies.h>
0007 #include <linux/kernel.h> /* UINT_MAX */
0008 #include <linux/mount.h>
0009 #include <linux/sched.h>
0010 #include <linux/sched/user.h>
0011 #include <linux/sched/signal.h>
0012 #include <linux/types.h>
0013 #include <linux/wait.h>
0014 #include <linux/audit.h>
0015 #include <linux/sched/mm.h>
0016 #include <linux/statfs.h>
0017 #include <linux/stringhash.h>
0018 
0019 #include "fanotify.h"
0020 
0021 static bool fanotify_path_equal(struct path *p1, struct path *p2)
0022 {
0023     return p1->mnt == p2->mnt && p1->dentry == p2->dentry;
0024 }
0025 
0026 static unsigned int fanotify_hash_path(const struct path *path)
0027 {
0028     return hash_ptr(path->dentry, FANOTIFY_EVENT_HASH_BITS) ^
0029         hash_ptr(path->mnt, FANOTIFY_EVENT_HASH_BITS);
0030 }
0031 
0032 static inline bool fanotify_fsid_equal(__kernel_fsid_t *fsid1,
0033                        __kernel_fsid_t *fsid2)
0034 {
0035     return fsid1->val[0] == fsid2->val[0] && fsid1->val[1] == fsid2->val[1];
0036 }
0037 
0038 static unsigned int fanotify_hash_fsid(__kernel_fsid_t *fsid)
0039 {
0040     return hash_32(fsid->val[0], FANOTIFY_EVENT_HASH_BITS) ^
0041         hash_32(fsid->val[1], FANOTIFY_EVENT_HASH_BITS);
0042 }
0043 
0044 static bool fanotify_fh_equal(struct fanotify_fh *fh1,
0045                   struct fanotify_fh *fh2)
0046 {
0047     if (fh1->type != fh2->type || fh1->len != fh2->len)
0048         return false;
0049 
0050     return !fh1->len ||
0051         !memcmp(fanotify_fh_buf(fh1), fanotify_fh_buf(fh2), fh1->len);
0052 }
0053 
0054 static unsigned int fanotify_hash_fh(struct fanotify_fh *fh)
0055 {
0056     long salt = (long)fh->type | (long)fh->len << 8;
0057 
0058     /*
0059      * full_name_hash() works long by long, so it handles fh buf optimally.
0060      */
0061     return full_name_hash((void *)salt, fanotify_fh_buf(fh), fh->len);
0062 }
0063 
0064 static bool fanotify_fid_event_equal(struct fanotify_fid_event *ffe1,
0065                      struct fanotify_fid_event *ffe2)
0066 {
0067     /* Do not merge fid events without object fh */
0068     if (!ffe1->object_fh.len)
0069         return false;
0070 
0071     return fanotify_fsid_equal(&ffe1->fsid, &ffe2->fsid) &&
0072         fanotify_fh_equal(&ffe1->object_fh, &ffe2->object_fh);
0073 }
0074 
0075 static bool fanotify_info_equal(struct fanotify_info *info1,
0076                 struct fanotify_info *info2)
0077 {
0078     if (info1->dir_fh_totlen != info2->dir_fh_totlen ||
0079         info1->dir2_fh_totlen != info2->dir2_fh_totlen ||
0080         info1->file_fh_totlen != info2->file_fh_totlen ||
0081         info1->name_len != info2->name_len ||
0082         info1->name2_len != info2->name2_len)
0083         return false;
0084 
0085     if (info1->dir_fh_totlen &&
0086         !fanotify_fh_equal(fanotify_info_dir_fh(info1),
0087                    fanotify_info_dir_fh(info2)))
0088         return false;
0089 
0090     if (info1->dir2_fh_totlen &&
0091         !fanotify_fh_equal(fanotify_info_dir2_fh(info1),
0092                    fanotify_info_dir2_fh(info2)))
0093         return false;
0094 
0095     if (info1->file_fh_totlen &&
0096         !fanotify_fh_equal(fanotify_info_file_fh(info1),
0097                    fanotify_info_file_fh(info2)))
0098         return false;
0099 
0100     if (info1->name_len &&
0101         memcmp(fanotify_info_name(info1), fanotify_info_name(info2),
0102            info1->name_len))
0103         return false;
0104 
0105     return !info1->name2_len ||
0106         !memcmp(fanotify_info_name2(info1), fanotify_info_name2(info2),
0107             info1->name2_len);
0108 }
0109 
0110 static bool fanotify_name_event_equal(struct fanotify_name_event *fne1,
0111                       struct fanotify_name_event *fne2)
0112 {
0113     struct fanotify_info *info1 = &fne1->info;
0114     struct fanotify_info *info2 = &fne2->info;
0115 
0116     /* Do not merge name events without dir fh */
0117     if (!info1->dir_fh_totlen)
0118         return false;
0119 
0120     if (!fanotify_fsid_equal(&fne1->fsid, &fne2->fsid))
0121         return false;
0122 
0123     return fanotify_info_equal(info1, info2);
0124 }
0125 
0126 static bool fanotify_error_event_equal(struct fanotify_error_event *fee1,
0127                        struct fanotify_error_event *fee2)
0128 {
0129     /* Error events against the same file system are always merged. */
0130     if (!fanotify_fsid_equal(&fee1->fsid, &fee2->fsid))
0131         return false;
0132 
0133     return true;
0134 }
0135 
0136 static bool fanotify_should_merge(struct fanotify_event *old,
0137                   struct fanotify_event *new)
0138 {
0139     pr_debug("%s: old=%p new=%p\n", __func__, old, new);
0140 
0141     if (old->hash != new->hash ||
0142         old->type != new->type || old->pid != new->pid)
0143         return false;
0144 
0145     /*
0146      * We want to merge many dirent events in the same dir (i.e.
0147      * creates/unlinks/renames), but we do not want to merge dirent
0148      * events referring to subdirs with dirent events referring to
0149      * non subdirs, otherwise, user won't be able to tell from a
0150      * mask FAN_CREATE|FAN_DELETE|FAN_ONDIR if it describes mkdir+
0151      * unlink pair or rmdir+create pair of events.
0152      */
0153     if ((old->mask & FS_ISDIR) != (new->mask & FS_ISDIR))
0154         return false;
0155 
0156     /*
0157      * FAN_RENAME event is reported with special info record types,
0158      * so we cannot merge it with other events.
0159      */
0160     if ((old->mask & FAN_RENAME) != (new->mask & FAN_RENAME))
0161         return false;
0162 
0163     switch (old->type) {
0164     case FANOTIFY_EVENT_TYPE_PATH:
0165         return fanotify_path_equal(fanotify_event_path(old),
0166                        fanotify_event_path(new));
0167     case FANOTIFY_EVENT_TYPE_FID:
0168         return fanotify_fid_event_equal(FANOTIFY_FE(old),
0169                         FANOTIFY_FE(new));
0170     case FANOTIFY_EVENT_TYPE_FID_NAME:
0171         return fanotify_name_event_equal(FANOTIFY_NE(old),
0172                          FANOTIFY_NE(new));
0173     case FANOTIFY_EVENT_TYPE_FS_ERROR:
0174         return fanotify_error_event_equal(FANOTIFY_EE(old),
0175                           FANOTIFY_EE(new));
0176     default:
0177         WARN_ON_ONCE(1);
0178     }
0179 
0180     return false;
0181 }
0182 
0183 /* Limit event merges to limit CPU overhead per event */
0184 #define FANOTIFY_MAX_MERGE_EVENTS 128
0185 
0186 /* and the list better be locked by something too! */
0187 static int fanotify_merge(struct fsnotify_group *group,
0188               struct fsnotify_event *event)
0189 {
0190     struct fanotify_event *old, *new = FANOTIFY_E(event);
0191     unsigned int bucket = fanotify_event_hash_bucket(group, new);
0192     struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
0193     int i = 0;
0194 
0195     pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
0196          group, event, bucket);
0197 
0198     /*
0199      * Don't merge a permission event with any other event so that we know
0200      * the event structure we have created in fanotify_handle_event() is the
0201      * one we should check for permission response.
0202      */
0203     if (fanotify_is_perm_event(new->mask))
0204         return 0;
0205 
0206     hlist_for_each_entry(old, hlist, merge_list) {
0207         if (++i > FANOTIFY_MAX_MERGE_EVENTS)
0208             break;
0209         if (fanotify_should_merge(old, new)) {
0210             old->mask |= new->mask;
0211 
0212             if (fanotify_is_error_event(old->mask))
0213                 FANOTIFY_EE(old)->err_count++;
0214 
0215             return 1;
0216         }
0217     }
0218 
0219     return 0;
0220 }
0221 
0222 /*
0223  * Wait for response to permission event. The function also takes care of
0224  * freeing the permission event (or offloads that in case the wait is canceled
0225  * by a signal). The function returns 0 in case access got allowed by userspace,
0226  * -EPERM in case userspace disallowed the access, and -ERESTARTSYS in case
0227  * the wait got interrupted by a signal.
0228  */
0229 static int fanotify_get_response(struct fsnotify_group *group,
0230                  struct fanotify_perm_event *event,
0231                  struct fsnotify_iter_info *iter_info)
0232 {
0233     int ret;
0234 
0235     pr_debug("%s: group=%p event=%p\n", __func__, group, event);
0236 
0237     ret = wait_event_killable(group->fanotify_data.access_waitq,
0238                   event->state == FAN_EVENT_ANSWERED);
0239     /* Signal pending? */
0240     if (ret < 0) {
0241         spin_lock(&group->notification_lock);
0242         /* Event reported to userspace and no answer yet? */
0243         if (event->state == FAN_EVENT_REPORTED) {
0244             /* Event will get freed once userspace answers to it */
0245             event->state = FAN_EVENT_CANCELED;
0246             spin_unlock(&group->notification_lock);
0247             return ret;
0248         }
0249         /* Event not yet reported? Just remove it. */
0250         if (event->state == FAN_EVENT_INIT) {
0251             fsnotify_remove_queued_event(group, &event->fae.fse);
0252             /* Permission events are not supposed to be hashed */
0253             WARN_ON_ONCE(!hlist_unhashed(&event->fae.merge_list));
0254         }
0255         /*
0256          * Event may be also answered in case signal delivery raced
0257          * with wakeup. In that case we have nothing to do besides
0258          * freeing the event and reporting error.
0259          */
0260         spin_unlock(&group->notification_lock);
0261         goto out;
0262     }
0263 
0264     /* userspace responded, convert to something usable */
0265     switch (event->response & ~FAN_AUDIT) {
0266     case FAN_ALLOW:
0267         ret = 0;
0268         break;
0269     case FAN_DENY:
0270     default:
0271         ret = -EPERM;
0272     }
0273 
0274     /* Check if the response should be audited */
0275     if (event->response & FAN_AUDIT)
0276         audit_fanotify(event->response & ~FAN_AUDIT);
0277 
0278     pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
0279          group, event, ret);
0280 out:
0281     fsnotify_destroy_event(group, &event->fae.fse);
0282 
0283     return ret;
0284 }
0285 
0286 /*
0287  * This function returns a mask for an event that only contains the flags
0288  * that have been specifically requested by the user. Flags that may have
0289  * been included within the event mask, but have not been explicitly
0290  * requested by the user, will not be present in the returned mask.
0291  */
0292 static u32 fanotify_group_event_mask(struct fsnotify_group *group,
0293                      struct fsnotify_iter_info *iter_info,
0294                      u32 *match_mask, u32 event_mask,
0295                      const void *data, int data_type,
0296                      struct inode *dir)
0297 {
0298     __u32 marks_mask = 0, marks_ignore_mask = 0;
0299     __u32 test_mask, user_mask = FANOTIFY_OUTGOING_EVENTS |
0300                      FANOTIFY_EVENT_FLAGS;
0301     const struct path *path = fsnotify_data_path(data, data_type);
0302     unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
0303     struct fsnotify_mark *mark;
0304     bool ondir = event_mask & FAN_ONDIR;
0305     int type;
0306 
0307     pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
0308          __func__, iter_info->report_mask, event_mask, data, data_type);
0309 
0310     if (!fid_mode) {
0311         /* Do we have path to open a file descriptor? */
0312         if (!path)
0313             return 0;
0314         /* Path type events are only relevant for files and dirs */
0315         if (!d_is_reg(path->dentry) && !d_can_lookup(path->dentry))
0316             return 0;
0317     } else if (!(fid_mode & FAN_REPORT_FID)) {
0318         /* Do we have a directory inode to report? */
0319         if (!dir && !ondir)
0320             return 0;
0321     }
0322 
0323     fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
0324         /*
0325          * Apply ignore mask depending on event flags in ignore mask.
0326          */
0327         marks_ignore_mask |=
0328             fsnotify_effective_ignore_mask(mark, ondir, type);
0329 
0330         /*
0331          * Send the event depending on event flags in mark mask.
0332          */
0333         if (!fsnotify_mask_applicable(mark->mask, ondir, type))
0334             continue;
0335 
0336         marks_mask |= mark->mask;
0337 
0338         /* Record the mark types of this group that matched the event */
0339         *match_mask |= 1U << type;
0340     }
0341 
0342     test_mask = event_mask & marks_mask & ~marks_ignore_mask;
0343 
0344     /*
0345      * For dirent modification events (create/delete/move) that do not carry
0346      * the child entry name information, we report FAN_ONDIR for mkdir/rmdir
0347      * so user can differentiate them from creat/unlink.
0348      *
0349      * For backward compatibility and consistency, do not report FAN_ONDIR
0350      * to user in legacy fanotify mode (reporting fd) and report FAN_ONDIR
0351      * to user in fid mode for all event types.
0352      *
0353      * We never report FAN_EVENT_ON_CHILD to user, but we do pass it in to
0354      * fanotify_alloc_event() when group is reporting fid as indication
0355      * that event happened on child.
0356      */
0357     if (fid_mode) {
0358         /* Do not report event flags without any event */
0359         if (!(test_mask & ~FANOTIFY_EVENT_FLAGS))
0360             return 0;
0361     } else {
0362         user_mask &= ~FANOTIFY_EVENT_FLAGS;
0363     }
0364 
0365     return test_mask & user_mask;
0366 }
0367 
0368 /*
0369  * Check size needed to encode fanotify_fh.
0370  *
0371  * Return size of encoded fh without fanotify_fh header.
0372  * Return 0 on failure to encode.
0373  */
0374 static int fanotify_encode_fh_len(struct inode *inode)
0375 {
0376     int dwords = 0;
0377     int fh_len;
0378 
0379     if (!inode)
0380         return 0;
0381 
0382     exportfs_encode_inode_fh(inode, NULL, &dwords, NULL);
0383     fh_len = dwords << 2;
0384 
0385     /*
0386      * struct fanotify_error_event might be preallocated and is
0387      * limited to MAX_HANDLE_SZ.  This should never happen, but
0388      * safeguard by forcing an invalid file handle.
0389      */
0390     if (WARN_ON_ONCE(fh_len > MAX_HANDLE_SZ))
0391         return 0;
0392 
0393     return fh_len;
0394 }
0395 
0396 /*
0397  * Encode fanotify_fh.
0398  *
0399  * Return total size of encoded fh including fanotify_fh header.
0400  * Return 0 on failure to encode.
0401  */
0402 static int fanotify_encode_fh(struct fanotify_fh *fh, struct inode *inode,
0403                   unsigned int fh_len, unsigned int *hash,
0404                   gfp_t gfp)
0405 {
0406     int dwords, type = 0;
0407     char *ext_buf = NULL;
0408     void *buf = fh->buf;
0409     int err;
0410 
0411     fh->type = FILEID_ROOT;
0412     fh->len = 0;
0413     fh->flags = 0;
0414 
0415     /*
0416      * Invalid FHs are used by FAN_FS_ERROR for errors not
0417      * linked to any inode. The f_handle won't be reported
0418      * back to userspace.
0419      */
0420     if (!inode)
0421         goto out;
0422 
0423     /*
0424      * !gpf means preallocated variable size fh, but fh_len could
0425      * be zero in that case if encoding fh len failed.
0426      */
0427     err = -ENOENT;
0428     if (fh_len < 4 || WARN_ON_ONCE(fh_len % 4) || fh_len > MAX_HANDLE_SZ)
0429         goto out_err;
0430 
0431     /* No external buffer in a variable size allocated fh */
0432     if (gfp && fh_len > FANOTIFY_INLINE_FH_LEN) {
0433         /* Treat failure to allocate fh as failure to encode fh */
0434         err = -ENOMEM;
0435         ext_buf = kmalloc(fh_len, gfp);
0436         if (!ext_buf)
0437             goto out_err;
0438 
0439         *fanotify_fh_ext_buf_ptr(fh) = ext_buf;
0440         buf = ext_buf;
0441         fh->flags |= FANOTIFY_FH_FLAG_EXT_BUF;
0442     }
0443 
0444     dwords = fh_len >> 2;
0445     type = exportfs_encode_inode_fh(inode, buf, &dwords, NULL);
0446     err = -EINVAL;
0447     if (!type || type == FILEID_INVALID || fh_len != dwords << 2)
0448         goto out_err;
0449 
0450     fh->type = type;
0451     fh->len = fh_len;
0452 
0453 out:
0454     /*
0455      * Mix fh into event merge key.  Hash might be NULL in case of
0456      * unhashed FID events (i.e. FAN_FS_ERROR).
0457      */
0458     if (hash)
0459         *hash ^= fanotify_hash_fh(fh);
0460 
0461     return FANOTIFY_FH_HDR_LEN + fh_len;
0462 
0463 out_err:
0464     pr_warn_ratelimited("fanotify: failed to encode fid (type=%d, len=%d, err=%i)\n",
0465                 type, fh_len, err);
0466     kfree(ext_buf);
0467     *fanotify_fh_ext_buf_ptr(fh) = NULL;
0468     /* Report the event without a file identifier on encode error */
0469     fh->type = FILEID_INVALID;
0470     fh->len = 0;
0471     return 0;
0472 }
0473 
0474 /*
0475  * FAN_REPORT_FID is ambiguous in that it reports the fid of the child for
0476  * some events and the fid of the parent for create/delete/move events.
0477  *
0478  * With the FAN_REPORT_TARGET_FID flag, the fid of the child is reported
0479  * also in create/delete/move events in addition to the fid of the parent
0480  * and the name of the child.
0481  */
0482 static inline bool fanotify_report_child_fid(unsigned int fid_mode, u32 mask)
0483 {
0484     if (mask & ALL_FSNOTIFY_DIRENT_EVENTS)
0485         return (fid_mode & FAN_REPORT_TARGET_FID);
0486 
0487     return (fid_mode & FAN_REPORT_FID) && !(mask & FAN_ONDIR);
0488 }
0489 
0490 /*
0491  * The inode to use as identifier when reporting fid depends on the event
0492  * and the group flags.
0493  *
0494  * With the group flag FAN_REPORT_TARGET_FID, always report the child fid.
0495  *
0496  * Without the group flag FAN_REPORT_TARGET_FID, report the modified directory
0497  * fid on dirent events and the child fid otherwise.
0498  *
0499  * For example:
0500  * FS_ATTRIB reports the child fid even if reported on a watched parent.
0501  * FS_CREATE reports the modified dir fid without FAN_REPORT_TARGET_FID.
0502  *       and reports the created child fid with FAN_REPORT_TARGET_FID.
0503  */
0504 static struct inode *fanotify_fid_inode(u32 event_mask, const void *data,
0505                     int data_type, struct inode *dir,
0506                     unsigned int fid_mode)
0507 {
0508     if ((event_mask & ALL_FSNOTIFY_DIRENT_EVENTS) &&
0509         !(fid_mode & FAN_REPORT_TARGET_FID))
0510         return dir;
0511 
0512     return fsnotify_data_inode(data, data_type);
0513 }
0514 
0515 /*
0516  * The inode to use as identifier when reporting dir fid depends on the event.
0517  * Report the modified directory inode on dirent modification events.
0518  * Report the "victim" inode if "victim" is a directory.
0519  * Report the parent inode if "victim" is not a directory and event is
0520  * reported to parent.
0521  * Otherwise, do not report dir fid.
0522  */
0523 static struct inode *fanotify_dfid_inode(u32 event_mask, const void *data,
0524                      int data_type, struct inode *dir)
0525 {
0526     struct inode *inode = fsnotify_data_inode(data, data_type);
0527 
0528     if (event_mask & ALL_FSNOTIFY_DIRENT_EVENTS)
0529         return dir;
0530 
0531     if (inode && S_ISDIR(inode->i_mode))
0532         return inode;
0533 
0534     return dir;
0535 }
0536 
0537 static struct fanotify_event *fanotify_alloc_path_event(const struct path *path,
0538                             unsigned int *hash,
0539                             gfp_t gfp)
0540 {
0541     struct fanotify_path_event *pevent;
0542 
0543     pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp);
0544     if (!pevent)
0545         return NULL;
0546 
0547     pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH;
0548     pevent->path = *path;
0549     *hash ^= fanotify_hash_path(path);
0550     path_get(path);
0551 
0552     return &pevent->fae;
0553 }
0554 
0555 static struct fanotify_event *fanotify_alloc_perm_event(const struct path *path,
0556                             gfp_t gfp)
0557 {
0558     struct fanotify_perm_event *pevent;
0559 
0560     pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
0561     if (!pevent)
0562         return NULL;
0563 
0564     pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
0565     pevent->response = 0;
0566     pevent->state = FAN_EVENT_INIT;
0567     pevent->path = *path;
0568     path_get(path);
0569 
0570     return &pevent->fae;
0571 }
0572 
0573 static struct fanotify_event *fanotify_alloc_fid_event(struct inode *id,
0574                                __kernel_fsid_t *fsid,
0575                                unsigned int *hash,
0576                                gfp_t gfp)
0577 {
0578     struct fanotify_fid_event *ffe;
0579 
0580     ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp);
0581     if (!ffe)
0582         return NULL;
0583 
0584     ffe->fae.type = FANOTIFY_EVENT_TYPE_FID;
0585     ffe->fsid = *fsid;
0586     *hash ^= fanotify_hash_fsid(fsid);
0587     fanotify_encode_fh(&ffe->object_fh, id, fanotify_encode_fh_len(id),
0588                hash, gfp);
0589 
0590     return &ffe->fae;
0591 }
0592 
0593 static struct fanotify_event *fanotify_alloc_name_event(struct inode *dir,
0594                             __kernel_fsid_t *fsid,
0595                             const struct qstr *name,
0596                             struct inode *child,
0597                             struct dentry *moved,
0598                             unsigned int *hash,
0599                             gfp_t gfp)
0600 {
0601     struct fanotify_name_event *fne;
0602     struct fanotify_info *info;
0603     struct fanotify_fh *dfh, *ffh;
0604     struct inode *dir2 = moved ? d_inode(moved->d_parent) : NULL;
0605     const struct qstr *name2 = moved ? &moved->d_name : NULL;
0606     unsigned int dir_fh_len = fanotify_encode_fh_len(dir);
0607     unsigned int dir2_fh_len = fanotify_encode_fh_len(dir2);
0608     unsigned int child_fh_len = fanotify_encode_fh_len(child);
0609     unsigned long name_len = name ? name->len : 0;
0610     unsigned long name2_len = name2 ? name2->len : 0;
0611     unsigned int len, size;
0612 
0613     /* Reserve terminating null byte even for empty name */
0614     size = sizeof(*fne) + name_len + name2_len + 2;
0615     if (dir_fh_len)
0616         size += FANOTIFY_FH_HDR_LEN + dir_fh_len;
0617     if (dir2_fh_len)
0618         size += FANOTIFY_FH_HDR_LEN + dir2_fh_len;
0619     if (child_fh_len)
0620         size += FANOTIFY_FH_HDR_LEN + child_fh_len;
0621     fne = kmalloc(size, gfp);
0622     if (!fne)
0623         return NULL;
0624 
0625     fne->fae.type = FANOTIFY_EVENT_TYPE_FID_NAME;
0626     fne->fsid = *fsid;
0627     *hash ^= fanotify_hash_fsid(fsid);
0628     info = &fne->info;
0629     fanotify_info_init(info);
0630     if (dir_fh_len) {
0631         dfh = fanotify_info_dir_fh(info);
0632         len = fanotify_encode_fh(dfh, dir, dir_fh_len, hash, 0);
0633         fanotify_info_set_dir_fh(info, len);
0634     }
0635     if (dir2_fh_len) {
0636         dfh = fanotify_info_dir2_fh(info);
0637         len = fanotify_encode_fh(dfh, dir2, dir2_fh_len, hash, 0);
0638         fanotify_info_set_dir2_fh(info, len);
0639     }
0640     if (child_fh_len) {
0641         ffh = fanotify_info_file_fh(info);
0642         len = fanotify_encode_fh(ffh, child, child_fh_len, hash, 0);
0643         fanotify_info_set_file_fh(info, len);
0644     }
0645     if (name_len) {
0646         fanotify_info_copy_name(info, name);
0647         *hash ^= full_name_hash((void *)name_len, name->name, name_len);
0648     }
0649     if (name2_len) {
0650         fanotify_info_copy_name2(info, name2);
0651         *hash ^= full_name_hash((void *)name2_len, name2->name,
0652                     name2_len);
0653     }
0654 
0655     pr_debug("%s: size=%u dir_fh_len=%u child_fh_len=%u name_len=%u name='%.*s'\n",
0656          __func__, size, dir_fh_len, child_fh_len,
0657          info->name_len, info->name_len, fanotify_info_name(info));
0658 
0659     if (dir2_fh_len) {
0660         pr_debug("%s: dir2_fh_len=%u name2_len=%u name2='%.*s'\n",
0661              __func__, dir2_fh_len, info->name2_len,
0662              info->name2_len, fanotify_info_name2(info));
0663     }
0664 
0665     return &fne->fae;
0666 }
0667 
0668 static struct fanotify_event *fanotify_alloc_error_event(
0669                         struct fsnotify_group *group,
0670                         __kernel_fsid_t *fsid,
0671                         const void *data, int data_type,
0672                         unsigned int *hash)
0673 {
0674     struct fs_error_report *report =
0675             fsnotify_data_error_report(data, data_type);
0676     struct inode *inode;
0677     struct fanotify_error_event *fee;
0678     int fh_len;
0679 
0680     if (WARN_ON_ONCE(!report))
0681         return NULL;
0682 
0683     fee = mempool_alloc(&group->fanotify_data.error_events_pool, GFP_NOFS);
0684     if (!fee)
0685         return NULL;
0686 
0687     fee->fae.type = FANOTIFY_EVENT_TYPE_FS_ERROR;
0688     fee->error = report->error;
0689     fee->err_count = 1;
0690     fee->fsid = *fsid;
0691 
0692     inode = report->inode;
0693     fh_len = fanotify_encode_fh_len(inode);
0694 
0695     /* Bad fh_len. Fallback to using an invalid fh. Should never happen. */
0696     if (!fh_len && inode)
0697         inode = NULL;
0698 
0699     fanotify_encode_fh(&fee->object_fh, inode, fh_len, NULL, 0);
0700 
0701     *hash ^= fanotify_hash_fsid(fsid);
0702 
0703     return &fee->fae;
0704 }
0705 
0706 static struct fanotify_event *fanotify_alloc_event(
0707                 struct fsnotify_group *group,
0708                 u32 mask, const void *data, int data_type,
0709                 struct inode *dir, const struct qstr *file_name,
0710                 __kernel_fsid_t *fsid, u32 match_mask)
0711 {
0712     struct fanotify_event *event = NULL;
0713     gfp_t gfp = GFP_KERNEL_ACCOUNT;
0714     unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS);
0715     struct inode *id = fanotify_fid_inode(mask, data, data_type, dir,
0716                           fid_mode);
0717     struct inode *dirid = fanotify_dfid_inode(mask, data, data_type, dir);
0718     const struct path *path = fsnotify_data_path(data, data_type);
0719     struct mem_cgroup *old_memcg;
0720     struct dentry *moved = NULL;
0721     struct inode *child = NULL;
0722     bool name_event = false;
0723     unsigned int hash = 0;
0724     bool ondir = mask & FAN_ONDIR;
0725     struct pid *pid;
0726 
0727     if ((fid_mode & FAN_REPORT_DIR_FID) && dirid) {
0728         /*
0729          * For certain events and group flags, report the child fid
0730          * in addition to reporting the parent fid and maybe child name.
0731          */
0732         if (fanotify_report_child_fid(fid_mode, mask) && id != dirid)
0733             child = id;
0734 
0735         id = dirid;
0736 
0737         /*
0738          * We record file name only in a group with FAN_REPORT_NAME
0739          * and when we have a directory inode to report.
0740          *
0741          * For directory entry modification event, we record the fid of
0742          * the directory and the name of the modified entry.
0743          *
0744          * For event on non-directory that is reported to parent, we
0745          * record the fid of the parent and the name of the child.
0746          *
0747          * Even if not reporting name, we need a variable length
0748          * fanotify_name_event if reporting both parent and child fids.
0749          */
0750         if (!(fid_mode & FAN_REPORT_NAME)) {
0751             name_event = !!child;
0752             file_name = NULL;
0753         } else if ((mask & ALL_FSNOTIFY_DIRENT_EVENTS) || !ondir) {
0754             name_event = true;
0755         }
0756 
0757         /*
0758          * In the special case of FAN_RENAME event, use the match_mask
0759          * to determine if we need to report only the old parent+name,
0760          * only the new parent+name or both.
0761          * 'dirid' and 'file_name' are the old parent+name and
0762          * 'moved' has the new parent+name.
0763          */
0764         if (mask & FAN_RENAME) {
0765             bool report_old, report_new;
0766 
0767             if (WARN_ON_ONCE(!match_mask))
0768                 return NULL;
0769 
0770             /* Report both old and new parent+name if sb watching */
0771             report_old = report_new =
0772                 match_mask & (1U << FSNOTIFY_ITER_TYPE_SB);
0773             report_old |=
0774                 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE);
0775             report_new |=
0776                 match_mask & (1U << FSNOTIFY_ITER_TYPE_INODE2);
0777 
0778             if (!report_old) {
0779                 /* Do not report old parent+name */
0780                 dirid = NULL;
0781                 file_name = NULL;
0782             }
0783             if (report_new) {
0784                 /* Report new parent+name */
0785                 moved = fsnotify_data_dentry(data, data_type);
0786             }
0787         }
0788     }
0789 
0790     /*
0791      * For queues with unlimited length lost events are not expected and
0792      * can possibly have security implications. Avoid losing events when
0793      * memory is short. For the limited size queues, avoid OOM killer in the
0794      * target monitoring memcg as it may have security repercussion.
0795      */
0796     if (group->max_events == UINT_MAX)
0797         gfp |= __GFP_NOFAIL;
0798     else
0799         gfp |= __GFP_RETRY_MAYFAIL;
0800 
0801     /* Whoever is interested in the event, pays for the allocation. */
0802     old_memcg = set_active_memcg(group->memcg);
0803 
0804     if (fanotify_is_perm_event(mask)) {
0805         event = fanotify_alloc_perm_event(path, gfp);
0806     } else if (fanotify_is_error_event(mask)) {
0807         event = fanotify_alloc_error_event(group, fsid, data,
0808                            data_type, &hash);
0809     } else if (name_event && (file_name || moved || child)) {
0810         event = fanotify_alloc_name_event(dirid, fsid, file_name, child,
0811                           moved, &hash, gfp);
0812     } else if (fid_mode) {
0813         event = fanotify_alloc_fid_event(id, fsid, &hash, gfp);
0814     } else {
0815         event = fanotify_alloc_path_event(path, &hash, gfp);
0816     }
0817 
0818     if (!event)
0819         goto out;
0820 
0821     if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
0822         pid = get_pid(task_pid(current));
0823     else
0824         pid = get_pid(task_tgid(current));
0825 
0826     /* Mix event info, FAN_ONDIR flag and pid into event merge key */
0827     hash ^= hash_long((unsigned long)pid | ondir, FANOTIFY_EVENT_HASH_BITS);
0828     fanotify_init_event(event, hash, mask);
0829     event->pid = pid;
0830 
0831 out:
0832     set_active_memcg(old_memcg);
0833     return event;
0834 }
0835 
0836 /*
0837  * Get cached fsid of the filesystem containing the object from any connector.
0838  * All connectors are supposed to have the same fsid, but we do not verify that
0839  * here.
0840  */
0841 static __kernel_fsid_t fanotify_get_fsid(struct fsnotify_iter_info *iter_info)
0842 {
0843     struct fsnotify_mark *mark;
0844     int type;
0845     __kernel_fsid_t fsid = {};
0846 
0847     fsnotify_foreach_iter_mark_type(iter_info, mark, type) {
0848         struct fsnotify_mark_connector *conn;
0849 
0850         conn = READ_ONCE(mark->connector);
0851         /* Mark is just getting destroyed or created? */
0852         if (!conn)
0853             continue;
0854         if (!(conn->flags & FSNOTIFY_CONN_FLAG_HAS_FSID))
0855             continue;
0856         /* Pairs with smp_wmb() in fsnotify_add_mark_list() */
0857         smp_rmb();
0858         fsid = conn->fsid;
0859         if (WARN_ON_ONCE(!fsid.val[0] && !fsid.val[1]))
0860             continue;
0861         return fsid;
0862     }
0863 
0864     return fsid;
0865 }
0866 
0867 /*
0868  * Add an event to hash table for faster merge.
0869  */
0870 static void fanotify_insert_event(struct fsnotify_group *group,
0871                   struct fsnotify_event *fsn_event)
0872 {
0873     struct fanotify_event *event = FANOTIFY_E(fsn_event);
0874     unsigned int bucket = fanotify_event_hash_bucket(group, event);
0875     struct hlist_head *hlist = &group->fanotify_data.merge_hash[bucket];
0876 
0877     assert_spin_locked(&group->notification_lock);
0878 
0879     if (!fanotify_is_hashed_event(event->mask))
0880         return;
0881 
0882     pr_debug("%s: group=%p event=%p bucket=%u\n", __func__,
0883          group, event, bucket);
0884 
0885     hlist_add_head(&event->merge_list, hlist);
0886 }
0887 
0888 static int fanotify_handle_event(struct fsnotify_group *group, u32 mask,
0889                  const void *data, int data_type,
0890                  struct inode *dir,
0891                  const struct qstr *file_name, u32 cookie,
0892                  struct fsnotify_iter_info *iter_info)
0893 {
0894     int ret = 0;
0895     struct fanotify_event *event;
0896     struct fsnotify_event *fsn_event;
0897     __kernel_fsid_t fsid = {};
0898     u32 match_mask = 0;
0899 
0900     BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
0901     BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
0902     BUILD_BUG_ON(FAN_ATTRIB != FS_ATTRIB);
0903     BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
0904     BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
0905     BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
0906     BUILD_BUG_ON(FAN_MOVED_TO != FS_MOVED_TO);
0907     BUILD_BUG_ON(FAN_MOVED_FROM != FS_MOVED_FROM);
0908     BUILD_BUG_ON(FAN_CREATE != FS_CREATE);
0909     BUILD_BUG_ON(FAN_DELETE != FS_DELETE);
0910     BUILD_BUG_ON(FAN_DELETE_SELF != FS_DELETE_SELF);
0911     BUILD_BUG_ON(FAN_MOVE_SELF != FS_MOVE_SELF);
0912     BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
0913     BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
0914     BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
0915     BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
0916     BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
0917     BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
0918     BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
0919     BUILD_BUG_ON(FAN_FS_ERROR != FS_ERROR);
0920     BUILD_BUG_ON(FAN_RENAME != FS_RENAME);
0921 
0922     BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 21);
0923 
0924     mask = fanotify_group_event_mask(group, iter_info, &match_mask,
0925                      mask, data, data_type, dir);
0926     if (!mask)
0927         return 0;
0928 
0929     pr_debug("%s: group=%p mask=%x report_mask=%x\n", __func__,
0930          group, mask, match_mask);
0931 
0932     if (fanotify_is_perm_event(mask)) {
0933         /*
0934          * fsnotify_prepare_user_wait() fails if we race with mark
0935          * deletion.  Just let the operation pass in that case.
0936          */
0937         if (!fsnotify_prepare_user_wait(iter_info))
0938             return 0;
0939     }
0940 
0941     if (FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS)) {
0942         fsid = fanotify_get_fsid(iter_info);
0943         /* Racing with mark destruction or creation? */
0944         if (!fsid.val[0] && !fsid.val[1])
0945             return 0;
0946     }
0947 
0948     event = fanotify_alloc_event(group, mask, data, data_type, dir,
0949                      file_name, &fsid, match_mask);
0950     ret = -ENOMEM;
0951     if (unlikely(!event)) {
0952         /*
0953          * We don't queue overflow events for permission events as
0954          * there the access is denied and so no event is in fact lost.
0955          */
0956         if (!fanotify_is_perm_event(mask))
0957             fsnotify_queue_overflow(group);
0958         goto finish;
0959     }
0960 
0961     fsn_event = &event->fse;
0962     ret = fsnotify_insert_event(group, fsn_event, fanotify_merge,
0963                     fanotify_insert_event);
0964     if (ret) {
0965         /* Permission events shouldn't be merged */
0966         BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
0967         /* Our event wasn't used in the end. Free it. */
0968         fsnotify_destroy_event(group, fsn_event);
0969 
0970         ret = 0;
0971     } else if (fanotify_is_perm_event(mask)) {
0972         ret = fanotify_get_response(group, FANOTIFY_PERM(event),
0973                         iter_info);
0974     }
0975 finish:
0976     if (fanotify_is_perm_event(mask))
0977         fsnotify_finish_user_wait(iter_info);
0978 
0979     return ret;
0980 }
0981 
0982 static void fanotify_free_group_priv(struct fsnotify_group *group)
0983 {
0984     kfree(group->fanotify_data.merge_hash);
0985     if (group->fanotify_data.ucounts)
0986         dec_ucount(group->fanotify_data.ucounts,
0987                UCOUNT_FANOTIFY_GROUPS);
0988 
0989     if (mempool_initialized(&group->fanotify_data.error_events_pool))
0990         mempool_exit(&group->fanotify_data.error_events_pool);
0991 }
0992 
0993 static void fanotify_free_path_event(struct fanotify_event *event)
0994 {
0995     path_put(fanotify_event_path(event));
0996     kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event));
0997 }
0998 
0999 static void fanotify_free_perm_event(struct fanotify_event *event)
1000 {
1001     path_put(fanotify_event_path(event));
1002     kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
1003 }
1004 
1005 static void fanotify_free_fid_event(struct fanotify_event *event)
1006 {
1007     struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
1008 
1009     if (fanotify_fh_has_ext_buf(&ffe->object_fh))
1010         kfree(fanotify_fh_ext_buf(&ffe->object_fh));
1011     kmem_cache_free(fanotify_fid_event_cachep, ffe);
1012 }
1013 
1014 static void fanotify_free_name_event(struct fanotify_event *event)
1015 {
1016     kfree(FANOTIFY_NE(event));
1017 }
1018 
1019 static void fanotify_free_error_event(struct fsnotify_group *group,
1020                       struct fanotify_event *event)
1021 {
1022     struct fanotify_error_event *fee = FANOTIFY_EE(event);
1023 
1024     mempool_free(fee, &group->fanotify_data.error_events_pool);
1025 }
1026 
1027 static void fanotify_free_event(struct fsnotify_group *group,
1028                 struct fsnotify_event *fsn_event)
1029 {
1030     struct fanotify_event *event;
1031 
1032     event = FANOTIFY_E(fsn_event);
1033     put_pid(event->pid);
1034     switch (event->type) {
1035     case FANOTIFY_EVENT_TYPE_PATH:
1036         fanotify_free_path_event(event);
1037         break;
1038     case FANOTIFY_EVENT_TYPE_PATH_PERM:
1039         fanotify_free_perm_event(event);
1040         break;
1041     case FANOTIFY_EVENT_TYPE_FID:
1042         fanotify_free_fid_event(event);
1043         break;
1044     case FANOTIFY_EVENT_TYPE_FID_NAME:
1045         fanotify_free_name_event(event);
1046         break;
1047     case FANOTIFY_EVENT_TYPE_OVERFLOW:
1048         kfree(event);
1049         break;
1050     case FANOTIFY_EVENT_TYPE_FS_ERROR:
1051         fanotify_free_error_event(group, event);
1052         break;
1053     default:
1054         WARN_ON_ONCE(1);
1055     }
1056 }
1057 
1058 static void fanotify_freeing_mark(struct fsnotify_mark *mark,
1059                   struct fsnotify_group *group)
1060 {
1061     if (!FAN_GROUP_FLAG(group, FAN_UNLIMITED_MARKS))
1062         dec_ucount(group->fanotify_data.ucounts, UCOUNT_FANOTIFY_MARKS);
1063 }
1064 
1065 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
1066 {
1067     kmem_cache_free(fanotify_mark_cache, fsn_mark);
1068 }
1069 
1070 const struct fsnotify_ops fanotify_fsnotify_ops = {
1071     .handle_event = fanotify_handle_event,
1072     .free_group_priv = fanotify_free_group_priv,
1073     .free_event = fanotify_free_event,
1074     .freeing_mark = fanotify_freeing_mark,
1075     .free_mark = fanotify_free_mark,
1076 };
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!