Whamcloud - gitweb
LU-13617 llite: don't hold inode_lock for security notify
[fs/lustre-release.git] / lustre / llite / dir.c
index 7138a04..a452779 100644 (file)
@@ -50,6 +50,7 @@
 #include <obd_support.h>
 #include <obd_class.h>
 #include <uapi/linux/lustre/lustre_ioctl.h>
+#include <uapi/linux/llcrypt.h>
 #include <lustre_lib.h>
 #include <lustre_dlm.h>
 #include <lustre_fid.h>
@@ -175,29 +176,6 @@ void ll_release_page(struct inode *inode, struct page *page,
        put_page(page);
 }
 
-/**
- * return IF_* type for given lu_dirent entry.
- * IF_* flag shld be converted to particular OS file type in
- * platform llite module.
- */
-static u16 ll_dirent_type_get(struct lu_dirent *ent)
-{
-       u16 type = 0;
-       struct luda_type *lt;
-       int len = 0;
-
-       if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
-               const unsigned align = sizeof(struct luda_type) - 1;
-
-               len = le16_to_cpu(ent->lde_namelen);
-               len = (len + align) & ~align;
-               lt = (void *)ent->lde_name + len;
-               type = IFTODT(le16_to_cpu(lt->lt_type));
-       }
-
-       return type;
-}
-
 #ifdef HAVE_DIR_CONTEXT
 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
                struct dir_context *ctx)
@@ -256,7 +234,7 @@ int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
                                lhash = hash;
                        fid_le_to_cpu(&fid, &ent->lde_fid);
                        ino = cl_fid_build_ino(&fid, is_api32);
-                       type = ll_dirent_type_get(ent);
+                       type = IFTODT(lu_dirent_type_get(ent));
                        /* For ll_nfs_get_name_filldir(), it will try to access
                         * 'ent' through 'lde_name', so the parameter 'name'
                         * for 'filldir()' must be part of the 'ent'. */
@@ -412,27 +390,6 @@ out:
        RETURN(rc);
 }
 
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
-static int ll_send_mgc_param(struct obd_export *mgc, char *string)
-{
-        struct mgs_send_param *msp;
-        int rc = 0;
-
-        OBD_ALLOC_PTR(msp);
-        if (!msp)
-                return -ENOMEM;
-
-       strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
-        rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
-                                sizeof(struct mgs_send_param), msp, NULL);
-        if (rc)
-                CERROR("Failed to set parameter: %d\n", rc);
-        OBD_FREE_PTR(msp);
-
-        return rc;
-}
-#endif
-
 /**
  * Create striped directory with specified stripe(@lump)
  *
@@ -461,10 +418,11 @@ static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
                                                  strlen(dirname)),
                },
        };
+       bool encrypt = false;
        int err;
        ENTRY;
 
-       if (unlikely(!lmv_magic_supported(lump->lum_magic)))
+       if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
                RETURN(-EINVAL);
 
        if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
@@ -489,7 +447,26 @@ static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
            !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
                RETURN(-ENOENT);
 
-       if (unlikely(!lmv_magic_supported(cpu_to_le32(lump->lum_magic))))
+       if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
+               if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
+                    LMV_HASH_TYPE_CRUSH) {
+                       /* if server doesn't support 'crush' hash type,
+                        * switch to fnv_1a_64.
+                        */
+                       lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
+                       lump->lum_hash_type |= LMV_HASH_TYPE_FNV_1A_64;
+               } else if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
+                    LMV_HASH_TYPE_UNKNOWN) {
+                       /* from 2.14 MDT will choose default hash type if client
+                        * doesn't set a valid one, while old server doesn't
+                        * handle it.
+                        */
+                       lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
+                       lump->lum_hash_type |= LMV_HASH_TYPE_DEFAULT;
+               }
+       }
+
+       if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
                lustre_swab_lmv_user_md(lump);
 
        if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
@@ -501,6 +478,16 @@ static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
        if (IS_ERR(op_data))
                RETURN(PTR_ERR(op_data));
 
+       if (IS_ENCRYPTED(parent) ||
+           unlikely(llcrypt_dummy_context_enabled(parent))) {
+               err = llcrypt_get_encryption_info(parent);
+               if (err)
+                       GOTO(out_op_data, err);
+               if (!llcrypt_has_encryption_key(parent))
+                       GOTO(out_op_data, err = -ENOKEY);
+               encrypt = true;
+       }
+
        if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
                /* selinux_dentry_init_security() uses dentry->d_parent and name
                 * to determine the security context for the file. So our fake
@@ -530,17 +517,25 @@ static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
        dentry.d_inode = inode;
 
        if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
-               inode_lock(inode);
+               /* no need to protect selinux_inode_setsecurity() by
+                * inode_lock. Taking it would lead to a client deadlock
+                * LU-13617
+                */
                err = security_inode_notifysecctx(inode,
                                                  op_data->op_file_secctx,
                                                  op_data->op_file_secctx_size);
-               inode_unlock(inode);
        } else {
                err = ll_inode_init_security(&dentry, inode, parent);
        }
        if (err)
                GOTO(out_inode, err);
 
+       if (encrypt) {
+               err = llcrypt_inherit_context(parent, inode, NULL, false);
+               if (err)
+                       GOTO(out_inode, err);
+       }
+
 out_inode:
        if (inode != NULL)
                iput(inode);
@@ -559,10 +554,6 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
        struct md_op_data *op_data;
        struct ptlrpc_request *req = NULL;
        int rc = 0;
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
-       struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
-       struct obd_device *mgc = lsi->lsi_mgc;
-#endif
        int lum_size;
        ENTRY;
 
@@ -624,57 +615,6 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
        if (rc)
                RETURN(rc);
 
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
-       /*
-        * 2.9 server has stored filesystem default stripe in ROOT xattr,
-        * and it's stored into system config for backward compatibility.
-        *
-        * In the following we use the fact that LOV_USER_MAGIC_V1 and
-        * LOV_USER_MAGIC_V3 have the same initial fields so we do not
-        * need the make the distiction between the 2 versions
-        */
-       if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
-           (lump == NULL ||
-            le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
-            le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
-               char *param = NULL;
-               char *buf;
-
-               OBD_ALLOC(param, MGS_PARAM_MAXLEN);
-               if (param == NULL)
-                       GOTO(end, rc = -ENOMEM);
-
-               buf = param;
-               /* Get fsname and assume devname to be -MDT0000. */
-               snprintf(buf, MGS_PARAM_MAXLEN, "%s-MDT0000.lov",
-                        sbi->ll_fsname);
-               buf += strlen(buf);
-
-               /* Set root stripesize */
-               snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
-                        lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
-               rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
-               if (rc)
-                       GOTO(end, rc);
-
-               /* Set root stripecount */
-               snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
-                        lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
-               rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
-               if (rc)
-                       GOTO(end, rc);
-
-               /* Set root stripeoffset */
-               snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
-                        lump ? le16_to_cpu(lump->lmm_stripe_offset) :
-                               (typeof(lump->lmm_stripe_offset))(-1));
-               rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
-
-end:
-               if (param != NULL)
-                       OBD_FREE(param, MGS_PARAM_MAXLEN);
-       }
-#endif
        RETURN(rc);
 }
 
@@ -1100,7 +1040,7 @@ static int copy_and_ct_start(int cmd, struct obd_export *exp,
 
                count = 0;
                for (i = 0; i < sizeof(archive_mask) * 8; i++) {
-                       if ((1 << i) & archive_mask) {
+                       if (BIT(i) & archive_mask) {
                                lk->lk_data[count] = i + 1;
                                count++;
                        }
@@ -1188,29 +1128,34 @@ static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
        case Q_SETQUOTA:
        case Q_SETINFO:
        case LUSTRE_Q_SETDEFAULT:
+       case LUSTRE_Q_SETQUOTAPOOL:
+       case LUSTRE_Q_SETINFOPOOL:
                if (!cfs_capable(CFS_CAP_SYS_ADMIN))
                        RETURN(-EPERM);
                break;
        case Q_GETQUOTA:
        case LUSTRE_Q_GETDEFAULT:
+       case LUSTRE_Q_GETQUOTAPOOL:
                if (check_owner(type, id) &&
                    (!cfs_capable(CFS_CAP_SYS_ADMIN)))
                        RETURN(-EPERM);
                break;
        case Q_GETINFO:
+       case LUSTRE_Q_GETINFOPOOL:
                break;
        default:
                CERROR("unsupported quotactl op: %#x\n", cmd);
                RETURN(-ENOTSUPP);
        }
 
-        if (valid != QC_GENERAL) {
-                if (cmd == Q_GETINFO)
-                        qctl->qc_cmd = Q_GETOINFO;
-                else if (cmd == Q_GETQUOTA)
-                        qctl->qc_cmd = Q_GETOQUOTA;
-                else
-                        RETURN(-EINVAL);
+       if (valid != QC_GENERAL) {
+               if (cmd == Q_GETINFO)
+                       qctl->qc_cmd = Q_GETOINFO;
+               else if (cmd == Q_GETQUOTA ||
+                        cmd == LUSTRE_Q_GETQUOTAPOOL)
+                       qctl->qc_cmd = Q_GETOQUOTA;
+               else
+                       RETURN(-EINVAL);
 
                 switch (valid) {
                 case QC_MDTIDX:
@@ -1240,22 +1185,26 @@ static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
                 qctl->qc_cmd = cmd;
         } else {
                 struct obd_quotactl *oqctl;
+               int oqctl_len = sizeof(*oqctl);
 
-                OBD_ALLOC_PTR(oqctl);
-                if (oqctl == NULL)
-                        RETURN(-ENOMEM);
+               if (LUSTRE_Q_CMD_IS_POOL(cmd))
+                       oqctl_len += LOV_MAXPOOLNAME + 1;
 
-                QCTL_COPY(oqctl, qctl);
-                rc = obd_quotactl(sbi->ll_md_exp, oqctl);
-                if (rc) {
-                        OBD_FREE_PTR(oqctl);
-                        RETURN(rc);
-                }
+               OBD_ALLOC(oqctl, oqctl_len);
+               if (oqctl == NULL)
+                       RETURN(-ENOMEM);
+
+               QCTL_COPY(oqctl, qctl);
+               rc = obd_quotactl(sbi->ll_md_exp, oqctl);
+               if (rc) {
+                       OBD_FREE(oqctl, oqctl_len);
+                       RETURN(rc);
+               }
                 /* If QIF_SPACE is not set, client should collect the
                  * space usage from OSSs by itself */
-                if (cmd == Q_GETQUOTA &&
-                    !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
-                    !oqctl->qc_dqblk.dqb_curspace) {
+               if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
+                   !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
+                   !oqctl->qc_dqblk.dqb_curspace) {
                         struct obd_quotactl *oqctl_tmp;
 
                         OBD_ALLOC_PTR(oqctl_tmp);
@@ -1292,11 +1241,11 @@ static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
                         OBD_FREE_PTR(oqctl_tmp);
                 }
 out:
-                QCTL_COPY(qctl, oqctl);
-                OBD_FREE_PTR(oqctl);
-        }
+               QCTL_COPY(qctl, oqctl);
+               OBD_FREE(oqctl, oqctl_len);
+       }
 
-        RETURN(rc);
+       RETURN(rc);
 }
 
 int ll_rmfid(struct file *file, void __user *arg)
@@ -1322,7 +1271,7 @@ int ll_rmfid(struct file *file, void __user *arg)
        OBD_ALLOC(lfa, size);
        if (!lfa)
                RETURN(-ENOMEM);
-       OBD_ALLOC(rcs, sizeof(int) * nr);
+       OBD_ALLOC_PTR_ARRAY(rcs, nr);
        if (!rcs)
                GOTO(free_lfa, rc = -ENOMEM);
 
@@ -1340,7 +1289,7 @@ int ll_rmfid(struct file *file, void __user *arg)
        }
 
 free_rcs:
-       OBD_FREE(rcs, sizeof(int) * nr);
+       OBD_FREE_PTR_ARRAY(rcs, nr);
 free_lfa:
        OBD_FREE(lfa, size);
 
@@ -1473,7 +1422,7 @@ out_free:
                lum = (struct lmv_user_md *)data->ioc_inlbuf2;
                lumlen = data->ioc_inllen2;
 
-               if (!lmv_magic_supported(lum->lum_magic)) {
+               if (!lmv_user_magic_supported(lum->lum_magic)) {
                        CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
                               lum->lum_magic, -EINVAL);
                        GOTO(lmv_out_free, rc = -EINVAL);
@@ -1494,11 +1443,7 @@ out_free:
                        GOTO(lmv_out_free, rc = -EINVAL);
                }
 
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
-               mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
-#else
                mode = data->ioc_type;
-#endif
                rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
 lmv_out_free:
                OBD_FREE_LARGE(buf, len);
@@ -1856,11 +1801,12 @@ out_rmdir:
                        stx.stx_atime.tv_sec = body->mbo_atime;
                        stx.stx_ctime.tv_sec = body->mbo_ctime;
                        stx.stx_mtime.tv_sec = body->mbo_mtime;
+                       stx.stx_btime.tv_sec = body->mbo_btime;
                        stx.stx_rdev_major = MAJOR(body->mbo_rdev);
                        stx.stx_rdev_minor = MINOR(body->mbo_rdev);
                        stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
                        stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
-                       stx.stx_mask |= STATX_BASIC_STATS;
+                       stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
 
                        /*
                         * For a striped directory, the size and blocks returned
@@ -1906,24 +1852,32 @@ out_req:
                return rc;
        }
        case OBD_IOC_QUOTACTL: {
-                struct if_quotactl *qctl;
+               struct if_quotactl *qctl;
+               int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
 
-                OBD_ALLOC_PTR(qctl);
-                if (!qctl)
-                        RETURN(-ENOMEM);
+               OBD_ALLOC(qctl, qctl_len);
+               if (!qctl)
+                       RETURN(-ENOMEM);
 
                if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
-                        GOTO(out_quotactl, rc = -EFAULT);
-
-                rc = quotactl_ioctl(sbi, qctl);
+                       GOTO(out_quotactl, rc = -EFAULT);
+
+               if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
+                       char __user *from = (char __user *)arg +
+                                       offsetof(typeof(*qctl), qc_poolname);
+                       if (copy_from_user(qctl->qc_poolname, from,
+                                          LOV_MAXPOOLNAME + 1))
+                               GOTO(out_quotactl, rc = -EFAULT);
+               }
 
+               rc = quotactl_ioctl(sbi, qctl);
                if (rc == 0 &&
                    copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
                         rc = -EFAULT;
 
-        out_quotactl:
-                OBD_FREE_PTR(qctl);
-                RETURN(rc);
+out_quotactl:
+               OBD_FREE(qctl, qctl_len);
+               RETURN(rc);
         }
         case OBD_IOC_GETDTNAME:
         case OBD_IOC_GETMDNAME:
@@ -2148,9 +2102,9 @@ migrate_free:
 
                RETURN(rc);
        }
-       case LL_IOC_FSGETXATTR:
+       case FS_IOC_FSGETXATTR:
                RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
-       case LL_IOC_FSSETXATTR:
+       case FS_IOC_FSSETXATTR:
                RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
        case LL_IOC_PCC_DETACH_BY_FID: {
                struct lu_pcc_detach_fid *detach;
@@ -2189,6 +2143,33 @@ out_detach:
                OBD_FREE_PTR(detach);
                RETURN(rc);
        }
+#ifdef HAVE_LUSTRE_CRYPTO
+       case LL_IOC_SET_ENCRYPTION_POLICY:
+               if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
+                       return -EOPNOTSUPP;
+               return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
+       case LL_IOC_GET_ENCRYPTION_POLICY_EX:
+               if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
+                       return -EOPNOTSUPP;
+               return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
+       case LL_IOC_ADD_ENCRYPTION_KEY:
+               if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
+                       return -EOPNOTSUPP;
+               return llcrypt_ioctl_add_key(file, (void __user *)arg);
+       case LL_IOC_REMOVE_ENCRYPTION_KEY:
+               if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
+                       return -EOPNOTSUPP;
+               return llcrypt_ioctl_remove_key(file, (void __user *)arg);
+       case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
+               if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
+                       return -EOPNOTSUPP;
+               return llcrypt_ioctl_remove_key_all_users(file,
+                                                         (void __user *)arg);
+       case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
+               if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
+                       return -EOPNOTSUPP;
+               return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
+#endif
        default:
                RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
                                     (void __user *)arg));