Whamcloud - gitweb
LU-5476 llite: Fix integer overflow in ll_fid2path
[fs/lustre-release.git] / lustre / llite / file.c
index 418464e..adf68ef 100644 (file)
 #include <lustre_lite.h>
 #include <linux/pagemap.h>
 #include <linux/file.h>
+#include <linux/sched.h>
 #include "llite_internal.h"
 #include <lustre/ll_fiemap.h>
+#include <lustre_ioctl.h>
 
 #include "cl_object.h"
 
@@ -206,7 +208,7 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp,
        if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
                struct mdt_body *body;
                body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-               if (!(body->valid & OBD_MD_FLRELEASED))
+               if (!(body->mbo_valid & OBD_MD_FLRELEASED))
                        rc = -EBUSY;
        }
 
@@ -332,10 +334,11 @@ static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                         rc = ll_md_real_close(file->f_dentry->d_inode,
                                               fd->fd_omode);
                 }
-        } else {
-                CERROR("Releasing a file %p with negative dentry %p. Name %s",
-                       file, file->f_dentry, file->f_dentry->d_name.name);
-        }
+       } else {
+               CERROR("released file has negative dentry: file = %p, "
+                      "dentry = %p, name = %s\n",
+                      file, file->f_dentry, file->f_dentry->d_name.name);
+       }
 
 out:
        LUSTRE_FPRIVATE(file) = NULL;
@@ -407,77 +410,69 @@ int ll_file_release(struct inode *inode, struct file *file)
         RETURN(rc);
 }
 
-static int ll_intent_file_open(struct file *file, void *lmm,
-                               int lmmsize, struct lookup_intent *itp)
+static int ll_intent_file_open(struct file *file, void *lmm, int lmmsize,
+                               struct lookup_intent *itp)
 {
-        struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
-        struct dentry *parent = file->f_dentry->d_parent;
-        const char *name = file->f_dentry->d_name.name;
-        const int len = file->f_dentry->d_name.len;
-        struct md_op_data *op_data;
-        struct ptlrpc_request *req;
-        __u32 opc = LUSTRE_OPC_ANY;
-        int rc;
-        ENTRY;
+       struct dentry *de = file->f_dentry;
+       struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
+       struct dentry *parent = de->d_parent;
+       const char *name = NULL;
+       int len = 0;
+       struct md_op_data *op_data;
+       struct ptlrpc_request *req = NULL;
+       int rc;
+       ENTRY;
 
-        if (!parent)
-                RETURN(-ENOENT);
-
-        /* Usually we come here only for NFSD, and we want open lock.
-           But we can also get here with pre 2.6.15 patchless kernels, and in
-           that case that lock is also ok */
-        /* We can also get here if there was cached open handle in revalidate_it
-         * but it disappeared while we were getting from there to ll_file_open.
-         * But this means this file was closed and immediatelly opened which
-         * makes a good candidate for using OPEN lock */
-        /* If lmmsize & lmm are not 0, we are just setting stripe info
-         * parameters. No need for the open lock */
-        if (lmm == NULL && lmmsize == 0) {
-                itp->it_flags |= MDS_OPEN_LOCK;
-                if (itp->it_flags & FMODE_WRITE)
-                        opc = LUSTRE_OPC_CREATE;
-        }
+       LASSERT(parent != NULL);
+       LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
 
-        op_data  = ll_prep_md_op_data(NULL, parent->d_inode,
-                                      file->f_dentry->d_inode, name, len,
-                                      O_RDWR, opc, NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
+       /* if server supports open-by-fid, or file name is invalid, don't pack
+        * name in open request */
+       if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
+           lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
+               name = de->d_name.name;
+               len = de->d_name.len;
+       }
 
-       itp->it_flags |= MDS_OPEN_BY_FID;
-        rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
-                            0 /*unused */, &req, ll_md_blocking_ast, 0);
-        ll_finish_md_op_data(op_data);
-        if (rc == -ESTALE) {
-                /* reason for keep own exit path - don`t flood log
-                * with messages with -ESTALE errors.
-                */
-                if (!it_disposition(itp, DISP_OPEN_OPEN) ||
-                     it_open_error(DISP_OPEN_OPEN, itp))
-                        GOTO(out, rc);
-                ll_release_openhandle(file->f_dentry, itp);
-                GOTO(out, rc);
-        }
+       op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
+                                    name, len, 0, LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
+       op_data->op_data = lmm;
+       op_data->op_data_size = lmmsize;
 
-        if (it_disposition(itp, DISP_LOOKUP_NEG))
-                GOTO(out, rc = -ENOENT);
+       rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
+                           &ll_md_blocking_ast, 0);
+       ll_finish_md_op_data(op_data);
+       if (rc == -ESTALE) {
+               /* reason for keep own exit path - don`t flood log
+                * with messages with -ESTALE errors.
+                */
+               if (!it_disposition(itp, DISP_OPEN_OPEN) ||
+                    it_open_error(DISP_OPEN_OPEN, itp))
+                       GOTO(out, rc);
+               ll_release_openhandle(de, itp);
+               GOTO(out, rc);
+       }
 
-        if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
-                rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
-                CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
-                GOTO(out, rc);
-        }
+       if (it_disposition(itp, DISP_LOOKUP_NEG))
+               GOTO(out, rc = -ENOENT);
 
-        rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp);
-        if (!rc && itp->d.lustre.it_lock_mode)
-                ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode,
-                                 itp, NULL);
+       if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
+               rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
+               CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
+               GOTO(out, rc);
+       }
+
+       rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
+       if (!rc && itp->d.lustre.it_lock_mode)
+               ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
 
 out:
        ptlrpc_req_finished(req);
-        ll_intent_drop_lock(itp);
+       ll_intent_drop_lock(itp);
 
-        RETURN(rc);
+       RETURN(rc);
 }
 
 /**
@@ -501,8 +496,8 @@ static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
        struct mdt_body *body;
 
        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-       och->och_fh = body->handle;
-       och->och_fid = body->fid1;
+       och->och_fh = body->mbo_handle;
+       och->och_fid = body->mbo_fid1;
        och->och_lease_handle.cookie = it->d.lustre.it_lock_handle;
        och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
        och->och_flags = it->it_flags;
@@ -531,13 +526,17 @@ static int ll_local_open(struct file *file, struct lookup_intent *it,
                        RETURN(rc);
 
                body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-               ll_ioepoch_open(lli, body->ioepoch);
+               ll_ioepoch_open(lli, body->mbo_ioepoch);
        }
 
        LUSTRE_FPRIVATE(file) = fd;
        ll_readahead_init(inode, &fd->fd_ras);
        fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
 
+       /* ll_cl_context initialize */
+       rwlock_init(&fd->fd_lock);
+       INIT_LIST_HEAD(&fd->fd_lccs);
+
        RETURN(0);
 }
 
@@ -664,9 +663,20 @@ restart:
                            would attempt to grab och_mutex as well, that would
                            result in a deadlock */
                        mutex_unlock(&lli->lli_och_mutex);
-                        it->it_create_mode |= M_CHECK_STALE;
+                       /*
+                        * Normally called under two situations:
+                        * 1. NFS export.
+                        * 2. A race/condition on MDS resulting in no open
+                        *    handle to be returned from LOOKUP|OPEN request,
+                        *    for example if the target entry was a symlink.
+                        *
+                        * Always fetch MDS_OPEN_LOCK if this is not setstripe.
+                        *
+                        * Always specify MDS_OPEN_BY_FID because we don't want
+                        * to get file with different fid.
+                        */
+                       it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
                         rc = ll_intent_file_open(file, NULL, 0, it);
-                        it->it_create_mode &= ~M_CHECK_STALE;
                         if (rc)
                                 GOTO(out_openerr, rc);
 
@@ -774,7 +784,7 @@ ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
        struct lookup_intent it = { .it_op = IT_OPEN };
        struct ll_sb_info *sbi = ll_i2sbi(inode);
        struct md_op_data *op_data;
-       struct ptlrpc_request *req;
+       struct ptlrpc_request *req = NULL;
        struct lustre_handle old_handle = { 0 };
        struct obd_client_handle *och = NULL;
        int rc;
@@ -840,15 +850,15 @@ ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
 
        it.it_flags = fmode | open_flags;
        it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
-       rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req,
-                               ll_md_blocking_lease_ast,
+       rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
+                           &ll_md_blocking_lease_ast,
        /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
         * it can be cancelled which may mislead applications that the lease is
         * broken;
         * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
         * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
         * doesn't deal with openhandle, so normal openhandle will be leaked. */
-                               LDLM_FL_NO_LRU | LDLM_FL_EXCL);
+                           LDLM_FL_NO_LRU | LDLM_FL_EXCL);
        ll_finish_md_op_data(op_data);
        ptlrpc_req_finished(req);
        if (rc < 0)
@@ -872,7 +882,7 @@ ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
        if (it.d.lustre.it_lock_mode == 0 ||
            it.d.lustre.it_lock_bits != MDS_INODELOCK_OPEN) {
                /* open lock must return for lease */
-               CERROR(DFID "lease granted but no open lock, %d/%Lu.\n",
+               CERROR(DFID "lease granted but no open lock, %d/"LPU64".\n",
                        PFID(ll_inode2fid(inode)), it.d.lustre.it_lock_mode,
                        it.d.lustre.it_lock_bits);
                GOTO(out_close, rc = -EPROTO);
@@ -968,11 +978,11 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
                        oinfo.oi_oa->o_flags |= OBD_FL_FLUSH;
        }
 
-        set = ptlrpc_prep_set();
-        if (set == NULL) {
-                CERROR("can't allocate ptlrpc set\n");
-                rc = -ENOMEM;
-        } else {
+       set = ptlrpc_prep_set();
+       if (set == NULL) {
+               CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM);
+               rc = -ENOMEM;
+       } else {
                 rc = obd_getattr_async(exp, &oinfo, set);
                 if (rc == 0)
                         rc = ptlrpc_set_wait(set);
@@ -1155,8 +1165,8 @@ restart:
                 struct ccc_io *cio = ccc_env_io(env);
                 int write_mutex_locked = 0;
 
-                cio->cui_fd  = LUSTRE_FPRIVATE(file);
-                vio->cui_io_subtype = args->via_io_subtype;
+               cio->cui_fd  = LUSTRE_FPRIVATE(file);
+               vio->cui_io_subtype = args->via_io_subtype;
 
                 switch (vio->cui_io_subtype) {
                 case IO_NORMAL:
@@ -1181,7 +1191,11 @@ restart:
                         CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
                         LBUG();
                 }
+
+               ll_cl_add(file, env, io);
                 result = cl_io_loop(env, io);
+               ll_cl_remove(file, env);
+
                if (args->via_io_subtype == IO_NORMAL)
                        up_read(&lli->lli_trunc_sem);
                if (write_mutex_locked)
@@ -1204,7 +1218,7 @@ out:
                CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n",
                       iot == CIT_READ ? "read" : "write",
                       file->f_dentry->d_name.name, *ppos, count);
-               LASSERTF(io->ci_nob == 0, "%zd", io->ci_nob);
+               LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob);
                goto restart;
        }
 
@@ -1512,6 +1526,7 @@ int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
        }
 
        ll_inode_size_lock(inode);
+       oit.it_flags |= MDS_OPEN_BY_FID;
        rc = ll_intent_file_open(file, lum, lum_size, &oit);
        if (rc)
                GOTO(out_unlock, rc);
@@ -1566,9 +1581,9 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
         LASSERT(body != NULL); /* checked by mdc_getattr_name */
 
-        lmmsize = body->eadatasize;
+       lmmsize = body->mbo_eadatasize;
 
-        if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
+       if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
                         lmmsize == 0) {
                 GOTO(out, rc = -ENODATA);
         }
@@ -1597,13 +1612,14 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
                  * avoid swab not existent lsm objects */
                 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
-                        if (S_ISREG(body->mode))
-                                lustre_swab_lov_user_md_objects(
-                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
-                                 stripe_count);
-                } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
-                        lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
-                        if (S_ISREG(body->mode))
+                       if (S_ISREG(body->mbo_mode))
+                               lustre_swab_lov_user_md_objects(
+                                   ((struct lov_user_md_v1 *)lmm)->lmm_objects,
+                                   stripe_count);
+               } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
+                       lustre_swab_lov_user_md_v3(
+                               (struct lov_user_md_v3 *)lmm);
+                       if (S_ISREG(body->mbo_mode))
                                 lustre_swab_lov_user_md_objects(
                                  ((struct lov_user_md_v3 *)lmm)->lmm_objects,
                                  stripe_count);
@@ -1878,38 +1894,39 @@ out:
        RETURN(rc);
 }
 
-int ll_fid2path(struct inode *inode, void *arg)
+int ll_fid2path(struct inode *inode, void __user *arg)
 {
        struct obd_export       *exp = ll_i2mdexp(inode);
-       struct getinfo_fid2path *gfout, *gfin;
-       int                      outsize, rc;
+       const struct getinfo_fid2path __user *gfin = arg;
+       __u32                    pathlen;
+       struct getinfo_fid2path *gfout;
+       size_t                   outsize;
+       int                      rc;
+
        ENTRY;
 
        if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
            !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
                RETURN(-EPERM);
 
-       /* Need to get the buflen */
-       OBD_ALLOC_PTR(gfin);
-       if (gfin == NULL)
-               RETURN(-ENOMEM);
-       if (copy_from_user(gfin, arg, sizeof(*gfin))) {
-               OBD_FREE_PTR(gfin);
+       /* Only need to get the buflen */
+       if (get_user(pathlen, &gfin->gf_pathlen))
                RETURN(-EFAULT);
-       }
 
-       outsize = sizeof(*gfout) + gfin->gf_pathlen;
+       if (pathlen > PATH_MAX)
+               RETURN(-EINVAL);
+
+       outsize = sizeof(*gfout) + pathlen;
        OBD_ALLOC(gfout, outsize);
-       if (gfout == NULL) {
-               OBD_FREE_PTR(gfin);
+       if (gfout == NULL)
                RETURN(-ENOMEM);
-       }
-       memcpy(gfout, gfin, sizeof(*gfout));
-       OBD_FREE_PTR(gfin);
+
+       if (copy_from_user(gfout, arg, sizeof(*gfout)))
+               GOTO(gf_free, rc = -EFAULT);
 
        /* Call mdc_iocontrol */
        rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
-       if (rc)
+       if (rc != 0)
                GOTO(gf_free, rc);
 
        if (copy_to_user(arg, gfout, outsize))
@@ -2292,10 +2309,14 @@ static int ll_hsm_import(struct inode *inode, struct file *file,
                         ATTR_MTIME | ATTR_MTIME_SET |
                         ATTR_ATIME | ATTR_ATIME_SET;
 
+       mutex_lock(&inode->i_mutex);
+
        rc = ll_setattr_raw(file->f_dentry, attr, true);
        if (rc == -ENODATA)
                rc = 0;
 
+       mutex_unlock(&inode->i_mutex);
+
 out:
        if (hss != NULL)
                OBD_FREE_PTR(hss);
@@ -2306,6 +2327,12 @@ out:
        RETURN(rc);
 }
 
+static inline long ll_lease_type_from_fmode(fmode_t fmode)
+{
+       return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
+              ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
+}
+
 static long
 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
@@ -2516,20 +2543,20 @@ ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                struct ll_inode_info *lli = ll_i2info(inode);
                struct obd_client_handle *och = NULL;
                bool lease_broken;
-               fmode_t mode = 0;
+               fmode_t fmode;
 
                switch (arg) {
-               case F_WRLCK:
+               case LL_LEASE_WRLCK:
                        if (!(file->f_mode & FMODE_WRITE))
                                RETURN(-EPERM);
-                       mode = FMODE_WRITE;
+                       fmode = FMODE_WRITE;
                        break;
-               case F_RDLCK:
+               case LL_LEASE_RDLCK:
                        if (!(file->f_mode & FMODE_READ))
                                RETURN(-EPERM);
-                       mode = FMODE_READ;
+                       fmode = FMODE_READ;
                        break;
-               case F_UNLCK:
+               case LL_LEASE_UNLCK:
                        mutex_lock(&lli->lli_och_mutex);
                        if (fd->fd_lease_och != NULL) {
                                och = fd->fd_lease_och;
@@ -2537,25 +2564,26 @@ ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                        }
                        mutex_unlock(&lli->lli_och_mutex);
 
-                       if (och != NULL) {
-                               mode = och->och_flags &(FMODE_READ|FMODE_WRITE);
-                               rc = ll_lease_close(och, inode, &lease_broken);
-                               if (rc == 0 && lease_broken)
-                                       mode = 0;
-                       } else {
-                               rc = -ENOLCK;
-                       }
+                       if (och == NULL)
+                               RETURN(-ENOLCK);
+
+                       fmode = och->och_flags;
+                       rc = ll_lease_close(och, inode, &lease_broken);
+                       if (rc < 0)
+                               RETURN(rc);
 
-                       /* return the type of lease or error */
-                       RETURN(rc < 0 ? rc : (int)mode);
+                       if (lease_broken)
+                               fmode = 0;
+
+                       RETURN(ll_lease_type_from_fmode(fmode));
                default:
                        RETURN(-EINVAL);
                }
 
-               CDEBUG(D_INODE, "Set lease with mode %d\n", mode);
+               CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
 
                /* apply for lease */
-               och = ll_lease_open(inode, file, mode, 0);
+               och = ll_lease_open(inode, file, fmode, 0);
                if (IS_ERR(och))
                        RETURN(PTR_ERR(och));
 
@@ -2576,8 +2604,8 @@ ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
        case LL_IOC_GET_LEASE: {
                struct ll_inode_info *lli = ll_i2info(inode);
                struct ldlm_lock *lock = NULL;
+               fmode_t fmode = 0;
 
-               rc = 0;
                mutex_lock(&lli->lli_och_mutex);
                if (fd->fd_lease_och != NULL) {
                        struct obd_client_handle *och = fd->fd_lease_och;
@@ -2586,14 +2614,15 @@ ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                        if (lock != NULL) {
                                lock_res_and_lock(lock);
                                if (!ldlm_is_cancel(lock))
-                                       rc = och->och_flags &
-                                               (FMODE_READ | FMODE_WRITE);
+                                       fmode = och->och_flags;
+
                                unlock_res_and_lock(lock);
                                LDLM_LOCK_PUT(lock);
                        }
                }
                mutex_unlock(&lli->lli_och_mutex);
-               RETURN(rc);
+
+               RETURN(ll_lease_type_from_fmode(fmode));
        }
        case LL_IOC_HSM_IMPORT: {
                struct hsm_user_import *hui;
@@ -2892,6 +2921,7 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
        struct md_op_data *op_data;
        struct lustre_handle lockh = {0};
        ldlm_policy_data_t flock = {{0}};
+       int fl_type = file_lock->fl_type;
        __u64 flags = 0;
        int rc;
        int rc2 = 0;
@@ -2927,7 +2957,7 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
        if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
                flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
 
-        switch (file_lock->fl_type) {
+       switch (fl_type) {
         case F_RDLCK:
                 einfo.ei_mode = LCK_PR;
                 break;
@@ -2946,8 +2976,7 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
                 einfo.ei_mode = LCK_PW;
                 break;
         default:
-                CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
-                        file_lock->fl_type);
+               CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
                 RETURN (-ENOTSUPP);
         }
 
@@ -2969,15 +2998,16 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
         case F_GETLK64:
 #endif
                 flags = LDLM_FL_TEST_LOCK;
-                /* Save the old mode so that if the mode in the lock changes we
-                 * can decrement the appropriate reader or writer refcount. */
-                file_lock->fl_type = einfo.ei_mode;
                 break;
         default:
                 CERROR("unknown fcntl lock command: %d\n", cmd);
                 RETURN (-EINVAL);
         }
 
+       /* Save the old mode so that if the mode in the lock changes we
+        * can decrement the appropriate reader or writer refcount. */
+       file_lock->fl_type = einfo.ei_mode;
+
         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
                                      LUSTRE_OPC_ANY, NULL);
         if (IS_ERR(op_data))
@@ -2988,8 +3018,12 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
               flock.l_flock.pid, flags, einfo.ei_mode,
               flock.l_flock.start, flock.l_flock.end);
 
-        rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
-                        op_data, &lockh, &flock, 0, NULL /* req */, flags);
+       rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
+                       flags);
+
+       /* Restore the file lock type if not TEST lock. */
+       if (!(flags & LDLM_FL_TEST_LOCK))
+               file_lock->fl_type = fl_type;
 
         if ((file_lock->fl_flags & FL_FLOCK) &&
             (rc == 0 || file_lock->fl_type == F_UNLCK))
@@ -3001,8 +3035,8 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 
        if (rc2 && file_lock->fl_type != F_UNLCK) {
                einfo.ei_mode = LCK_NL;
-               md_enqueue(sbi->ll_md_exp, &einfo, NULL,
-                       op_data, &lockh, &flock, 0, NULL /* req */, flags);
+               md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
+                          &lockh, flags);
                rc = rc2;
        }
 
@@ -3035,7 +3069,7 @@ int ll_get_fid_by_name(struct inode *parent, const char *name,
        if (body == NULL)
                GOTO(out_req, rc = -EFAULT);
        if (fid != NULL)
-               *fid = body->fid1;
+               *fid = body->mbo_fid1;
 out_req:
        ptlrpc_req_finished(req);
        RETURN(rc);
@@ -3246,14 +3280,9 @@ static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
                 if (IS_ERR(op_data))
                         RETURN(PTR_ERR(op_data));
 
-                oit.it_create_mode |= M_CHECK_STALE;
-                rc = md_intent_lock(exp, op_data, NULL, 0,
-                                    /* we are not interested in name
-                                       based lookup */
-                                    &oit, 0, &req,
-                                    ll_md_blocking_ast, 0);
+               rc = md_intent_lock(exp, op_data, &oit, &req,
+                                   &ll_md_blocking_ast, 0);
                 ll_finish_md_op_data(op_data);
-                oit.it_create_mode &= ~M_CHECK_STALE;
                 if (rc < 0) {
                         rc = ll_inode_revalidate_fini(inode, rc);
                         GOTO (out, rc);
@@ -3501,8 +3530,14 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
 # endif
 #endif
 {
-        int rc = 0;
-        ENTRY;
+       int rc = 0;
+       struct ll_sb_info *sbi;
+       struct root_squash_info *squash;
+       struct cred *cred = NULL;
+       const struct cred *old_cred = NULL;
+       cfs_cap_t cap;
+       bool squash_id = false;
+       ENTRY;
 
 #ifdef MAY_NOT_BLOCK
        if (mask & MAY_NOT_BLOCK)
@@ -3525,11 +3560,46 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
               PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
 
-       if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
-               return lustre_check_remote_perm(inode, mask);
+       /* squash fsuid/fsgid if needed */
+       sbi = ll_i2sbi(inode);
+       squash = &sbi->ll_squash;
+       if (unlikely(squash->rsi_uid != 0 &&
+                    uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
+                    !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
+                       squash_id = true;
+       }
+       if (squash_id) {
+               CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
+                      __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
+                      squash->rsi_uid, squash->rsi_gid);
+
+               /* update current process's credentials
+                * and FS capability */
+               cred = prepare_creds();
+               if (cred == NULL)
+                       RETURN(-ENOMEM);
+
+               cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
+               cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
+               for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
+                       if ((1 << cap) & CFS_CAP_FS_MASK)
+                               cap_lower(cred->cap_effective, cap);
+               }
+               old_cred = override_creds(cred);
+       }
+
+       ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
 
-       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
-       rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
+       if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+               rc = lustre_check_remote_perm(inode, mask);
+       else
+               rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
+
+       /* restore current process's credentials and FS capability */
+       if (squash_id) {
+               revert_creds(old_cred);
+               put_cred(cred);
+       }
 
        RETURN(rc);
 }
@@ -3602,15 +3672,15 @@ struct inode_operations ll_file_inode_operations = {
 /* dynamic ioctl number support routins */
 static struct llioc_ctl_data {
        struct rw_semaphore     ioc_sem;
-        cfs_list_t              ioc_head;
+       struct list_head        ioc_head;
 } llioc = {
-        __RWSEM_INITIALIZER(llioc.ioc_sem),
-        CFS_LIST_HEAD_INIT(llioc.ioc_head)
+       __RWSEM_INITIALIZER(llioc.ioc_sem),
+       LIST_HEAD_INIT(llioc.ioc_head)
 };
 
 
 struct llioc_data {
-        cfs_list_t              iocd_list;
+       struct list_head        iocd_list;
         unsigned int            iocd_size;
         llioc_callback_t        iocd_cb;
         unsigned int            iocd_count;
@@ -3639,7 +3709,7 @@ void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
         memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
 
        down_write(&llioc.ioc_sem);
-        cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
+       list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
        up_write(&llioc.ioc_sem);
 
         RETURN(in_data);
@@ -3653,11 +3723,11 @@ void ll_iocontrol_unregister(void *magic)
                 return;
 
        down_write(&llioc.ioc_sem);
-        cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
+       list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
                 if (tmp == magic) {
                         unsigned int size = tmp->iocd_size;
 
-                        cfs_list_del(&tmp->iocd_list);
+                       list_del(&tmp->iocd_list);
                        up_write(&llioc.ioc_sem);
 
                         OBD_FREE(tmp, size);
@@ -3681,7 +3751,7 @@ ll_iocontrol_call(struct inode *inode, struct file *file,
         int rc = -EINVAL, i;
 
        down_read(&llioc.ioc_sem);
-        cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
+       list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
                 for (i = 0; i < data->iocd_count; i++) {
                         if (cmd != data->iocd_cmd[i])
                                 continue;
@@ -3787,7 +3857,7 @@ static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
        if (body == NULL)
                GOTO(out, rc = -EPROTO);
 
-       lmmsize = body->eadatasize;
+       lmmsize = body->mbo_eadatasize;
        if (lmmsize == 0) /* empty layout */
                GOTO(out, rc = 0);
 
@@ -3838,7 +3908,7 @@ static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
        LASSERT(lock != NULL);
        LASSERT(ldlm_has_layout(lock));
 
-       LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d\n",
+       LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d",
                   PFID(&lli->lli_fid), inode, reconf);
 
        /* in case this is a caching lock and reinstate with new inode */
@@ -3950,8 +4020,8 @@ int ll_layout_refresh(struct inode *inode, __u32 *gen)
        struct ldlm_enqueue_info einfo = {
                .ei_type = LDLM_IBITS,
                .ei_mode = LCK_CR,
-               .ei_cb_bl = ll_md_blocking_ast,
-               .ei_cb_cp = ldlm_completion_ast,
+               .ei_cb_bl = &ll_md_blocking_ast,
+               .ei_cb_cp = &ldlm_completion_ast,
        };
        int rc;
        ENTRY;
@@ -3993,12 +4063,11 @@ again:
        it.it_op = IT_LAYOUT;
        lockh.cookie = 0ULL;
 
-       LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)\n",
+       LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
                          ll_get_fsname(inode->i_sb, NULL, 0),
                          PFID(&lli->lli_fid), inode);
 
-       rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
-                       NULL, 0, NULL, 0);
+       rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
        if (it.d.lustre.it_data != NULL)
                ptlrpc_req_finished(it.d.lustre.it_data);
        it.d.lustre.it_data = NULL;