Whamcloud - gitweb
LU-1030 clio: reimplement ll_fsync in clio way
[fs/lustre-release.git] / lustre / llite / file.c
index 231f2f3..5498217 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -202,15 +200,15 @@ int ll_md_real_close(struct inode *inode, int flags)
                 och_usecount = &lli->lli_open_fd_read_count;
         }
 
-        cfs_down(&lli->lli_och_sem);
+        cfs_mutex_lock(&lli->lli_och_mutex);
         if (*och_usecount) { /* There are still users of this handle, so
                                 skip freeing it. */
-                cfs_up(&lli->lli_och_sem);
+                cfs_mutex_unlock(&lli->lli_och_mutex);
                 RETURN(0);
         }
         och=*och_p;
         *och_p = NULL;
-        cfs_up(&lli->lli_och_sem);
+        cfs_mutex_unlock(&lli->lli_och_mutex);
 
         if (och) { /* There might be a race and somebody have freed this och
                       already */
@@ -242,7 +240,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                 struct inode *inode = file->f_dentry->d_inode;
                 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
 
-                cfs_down(&lli->lli_och_sem);
+                cfs_mutex_lock(&lli->lli_och_mutex);
                 if (fd->fd_omode & FMODE_WRITE) {
                         lockmode = LCK_CW;
                         LASSERT(lli->lli_open_fd_write_count);
@@ -256,7 +254,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                         LASSERT(lli->lli_open_fd_read_count);
                         lli->lli_open_fd_read_count--;
                 }
-                cfs_up(&lli->lli_och_sem);
+                cfs_mutex_unlock(&lli->lli_och_mutex);
 
                 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
                                    LDLM_IBITS, &policy, lockmode,
@@ -483,9 +481,7 @@ int ll_local_open(struct file *file, struct lookup_intent *it,
 
 /* Open a file, and (for the very first open) create objects on the OSTs at
  * this time.  If opened with O_LOV_DELAY_CREATE, then we don't do the object
- * creation or open until ll_lov_setstripe() ioctl is called.  We grab
- * lli_open_sem to ensure no other process will create objects, send the
- * stripe MD to the MDS, or try to destroy the objects if that fails.
+ * creation or open until ll_lov_setstripe() ioctl is called.
  *
  * If we already have the stripe MD locally then we don't request it in
  * md_open(), by passing a lmm_size = 0.
@@ -577,27 +573,25 @@ restart:
                 och_usecount = &lli->lli_open_fd_read_count;
         }
 
-        cfs_down(&lli->lli_och_sem);
+        cfs_mutex_lock(&lli->lli_och_mutex);
         if (*och_p) { /* Open handle is present */
                 if (it_disposition(it, DISP_OPEN_OPEN)) {
                         /* Well, there's extra open request that we do not need,
                            let's close it somehow. This will decref request. */
                         rc = it_open_error(DISP_OPEN_OPEN, it);
                         if (rc) {
-                                cfs_up(&lli->lli_och_sem);
+                                cfs_mutex_unlock(&lli->lli_och_mutex);
                                 GOTO(out_openerr, rc);
                         }
 
                         ll_release_openhandle(file->f_dentry, it);
-                        lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats,
-                                             LPROC_LL_OPEN);
                 }
                 (*och_usecount)++;
 
                 rc = ll_local_open(file, it, fd, NULL);
                 if (rc) {
                         (*och_usecount)--;
-                        cfs_up(&lli->lli_och_sem);
+                        cfs_mutex_unlock(&lli->lli_och_mutex);
                         GOTO(out_openerr, rc);
                 }
         } else {
@@ -606,9 +600,9 @@ restart:
                         /* We cannot just request lock handle now, new ELC code
                            means that one of other OPEN locks for this file
                            could be cancelled, and since blocking ast handler
-                           would attempt to grab och_sem as well, that would
+                           would attempt to grab och_mutex as well, that would
                            result in a deadlock */
-                        cfs_up(&lli->lli_och_sem);
+                        cfs_mutex_unlock(&lli->lli_och_mutex);
                         it->it_create_mode |= M_CHECK_STALE;
                         rc = ll_intent_file_open(file, NULL, 0, it);
                         it->it_create_mode &= ~M_CHECK_STALE;
@@ -634,15 +628,14 @@ restart:
 
                 LASSERT(it_disposition(it, DISP_ENQ_OPEN_REF));
 
-                ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
                 rc = ll_local_open(file, it, fd, *och_p);
                 if (rc)
                         GOTO(out_och_free, rc);
         }
-        cfs_up(&lli->lli_och_sem);
+        cfs_mutex_unlock(&lli->lli_och_mutex);
         fd = NULL;
 
-        /* Must do this outside lli_och_sem lock to prevent deadlock where
+        /* Must do this outside lli_och_mutex lock to prevent deadlock where
            different kind of OPEN lock for this same inode gets cancelled
            by ldlm_cancel_lru */
         if (!S_ISREG(inode->i_mode))
@@ -673,13 +666,15 @@ out_och_free:
                         *och_p = NULL; /* OBD_FREE writes some magic there */
                         (*och_usecount)--;
                 }
-                cfs_up(&lli->lli_och_sem);
+                cfs_mutex_unlock(&lli->lli_och_mutex);
 
 out_openerr:
                 if (opendir_set != 0)
                         ll_stop_statahead(inode, lli->lli_opendir_key);
                 if (fd != NULL)
                         ll_file_data_put(fd);
+        } else {
+                ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
         }
 
         return rc;
@@ -708,7 +703,8 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
                                OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
                                OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
-                               OBD_MD_FLGROUP | OBD_MD_FLEPOCH;
+                               OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
+                               OBD_MD_FLDATAVERSION;
         oinfo.oi_capa = capa;
         if (sync) {
                 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
@@ -728,7 +724,8 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
         if (rc == 0)
                 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
                                          OBD_MD_FLATIME | OBD_MD_FLMTIME |
-                                         OBD_MD_FLCTIME | OBD_MD_FLSIZE);
+                                         OBD_MD_FLCTIME | OBD_MD_FLSIZE |
+                                         OBD_MD_FLDATAVERSION);
         RETURN(rc);
 }
 
@@ -839,7 +836,7 @@ static ssize_t ll_file_io_generic(const struct lu_env *env,
         if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
                 struct vvp_io *vio = vvp_env_io(env);
                 struct ccc_io *cio = ccc_env_io(env);
-                int write_sem_locked = 0;
+                int write_mutex_locked = 0;
 
                 cio->cui_fd  = LUSTRE_FPRIVATE(file);
                 vio->cui_io_subtype = args->via_io_subtype;
@@ -854,9 +851,10 @@ static ssize_t ll_file_io_generic(const struct lu_env *env,
 #endif
                         if ((iot == CIT_WRITE) &&
                             !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
-                                if(cfs_down_interruptible(&lli->lli_write_sem))
+                                if (cfs_mutex_lock_interruptible(&lli->
+                                                               lli_write_mutex))
                                         GOTO(out, result = -ERESTARTSYS);
-                                write_sem_locked = 1;
+                                write_mutex_locked = 1;
                         } else if (iot == CIT_READ) {
                                 cfs_down_read(&lli->lli_trunc_sem);
                         }
@@ -874,8 +872,8 @@ static ssize_t ll_file_io_generic(const struct lu_env *env,
                         LBUG();
                 }
                 result = cl_io_loop(env, io);
-                if (write_sem_locked)
-                        cfs_up(&lli->lli_write_sem);
+                if (write_mutex_locked)
+                        cfs_mutex_unlock(&lli->lli_write_mutex);
                 else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ)
                         cfs_up_read(&lli->lli_trunc_sem);
         } else {
@@ -1249,10 +1247,11 @@ static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq,
         oa->o_nlink = ost_idx;
         oa->o_flags |= OBD_FL_RECREATE_OBJS;
         oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
-        obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid, OBD_MD_FLTYPE |
-                        OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME);
+        obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
+                                   OBD_MD_FLMTIME | OBD_MD_FLCTIME);
+        obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
         memcpy(lsm2, lsm, lsm_size);
-        rc = obd_create(exp, oa, &lsm2, &oti);
+        rc = obd_create(NULL, exp, oa, &lsm2, &oti);
 
         OBD_FREE_LARGE(lsm2, lsm_size);
         GOTO(out, rc);
@@ -1641,8 +1640,8 @@ int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
         fm_key.oa.o_seq = lsm->lsm_object_seq;
         fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 
-        obdo_from_inode(&fm_key.oa, inode, &ll_i2info(inode)->lli_fid,
-                        OBD_MD_FLSIZE);
+        obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
+        obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
         /* If filesize is 0, then there would be no objects for mapping */
         if (fm_key.oa.o_size == 0) {
                 fiemap->fm_mapped_extents = 0;
@@ -1651,7 +1650,8 @@ int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
 
         memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
 
-        rc = obd_get_info(exp, sizeof(fm_key), &fm_key, &vallen, fiemap, lsm);
+        rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
+                          fiemap, lsm);
         if (rc)
                 CERROR("obd_get_info failed: rc = %d\n", rc);
 
@@ -1746,15 +1746,51 @@ error:
         RETURN(rc);
 }
 
-#ifdef HAVE_UNLOCKED_IOCTL
+/*
+ * Read the data_version for inode.
+ *
+ * This value is computed using stripe object version on OST.
+ * Version is computed using server side locking.
+ *
+ * @param extent_lock  Take extent lock. Not needed if a process is already
+ *                     holding the OST object group locks.
+ */
+static int ll_data_version(struct inode *inode, __u64 *data_version,
+                           int extent_lock)
+{
+        struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
+        struct ll_sb_info    *sbi = ll_i2sbi(inode);
+        struct obdo          *obdo = NULL;
+        int                   rc;
+        ENTRY;
+
+        /* If no stripe, we consider version is 0. */
+        if (!lsm) {
+                *data_version = 0;
+                CDEBUG(D_INODE, "No object for inode\n");
+                RETURN(0);
+        }
+
+        OBD_ALLOC_PTR(obdo);
+        if (obdo == NULL)
+                RETURN(-ENOMEM);
+
+        rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock);
+        if (!rc) {
+                if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
+                        rc = -EOPNOTSUPP;
+                else
+                        *data_version = obdo->o_data_version;
+        }
+
+        OBD_FREE_PTR(obdo);
+
+        RETURN(rc);
+}
+
 long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
         struct inode *inode = file->f_dentry->d_inode;
-#else
-int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
-                  unsigned long arg)
-{
-#endif
         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
         int flags;
 
@@ -1836,6 +1872,23 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
         }
         case OBD_IOC_FID2PATH:
                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
+        case LL_IOC_DATA_VERSION: {
+                struct ioc_data_version idv;
+                int rc;
+
+                if (cfs_copy_from_user(&idv, (char *)arg, sizeof(idv)))
+                        RETURN(-EFAULT);
+
+                rc = ll_data_version(inode, &idv.idv_version,
+                                     !(idv.idv_flags & LL_DV_NOFLUSH));
+
+                if (rc == 0 &&
+                    cfs_copy_to_user((char *) arg, &idv, sizeof(idv)))
+                        RETURN(-EFAULT);
+
+                RETURN(rc);
+        }
+
         case LL_IOC_GET_MDTIDX: {
                 int mdtidx;
 
@@ -1929,10 +1982,54 @@ int ll_flush(struct file *file)
         return rc ? -EIO : 0;
 }
 
-#ifndef HAVE_FILE_FSYNC_2ARGS
-int ll_fsync(struct file *file, struct dentry *dentry, int data)
-#else
+/**
+ * Called to make sure a portion of file has been written out.
+ * if @local_only is not true, it will send OST_SYNC RPCs to ost.
+ */
+int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end)
+{
+       struct cl_env_nest nest;
+       struct lu_env *env;
+       struct cl_io *io;
+       struct obd_capa *capa = NULL;
+       struct cl_fsync_io *fio;
+       int result;
+       ENTRY;
+
+       env = cl_env_nested_get(&nest);
+       if (IS_ERR(env))
+               RETURN(PTR_ERR(env));
+
+       capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
+
+       io = ccc_env_thread_io(env);
+       io->ci_obj = cl_i2info(inode)->lli_clob;
+
+       /* initialize parameters for sync */
+       fio = &io->u.ci_fsync;
+       fio->fi_capa = capa;
+       fio->fi_start = start;
+       fio->fi_end = end;
+       fio->fi_fid = ll_inode2fid(inode);
+
+       if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
+               result = cl_io_loop(env, io);
+       else
+               result = io->ci_result;
+       cl_io_fini(env, io);
+       cl_env_nested_put(&nest, env);
+
+       capa_put(capa);
+
+       RETURN(result);
+}
+
+#ifdef HAVE_FILE_FSYNC_4ARGS
+int ll_fsync(struct file *file, loff_t start, loff_t end, int data)
+#elif defined(HAVE_FILE_FSYNC_2ARGS)
 int ll_fsync(struct file *file, int data)
+#else
+int ll_fsync(struct file *file, struct dentry *dentry, int data)
 #endif
 {
         struct inode *inode = file->f_dentry->d_inode;
@@ -1974,32 +2071,9 @@ int ll_fsync(struct file *file, int data)
                 ptlrpc_req_finished(req);
 
         if (data && lsm) {
-                struct obd_info *oinfo;
-
-                OBD_ALLOC_PTR(oinfo);
-                if (!oinfo)
-                        RETURN(rc ? rc : -ENOMEM);
-                OBDO_ALLOC(oinfo->oi_oa);
-                if (!oinfo->oi_oa) {
-                        OBD_FREE_PTR(oinfo);
-                        RETURN(rc ? rc : -ENOMEM);
-                }
-                oinfo->oi_oa->o_id = lsm->lsm_object_id;
-                oinfo->oi_oa->o_seq = lsm->lsm_object_seq;
-                oinfo->oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
-                obdo_from_inode(oinfo->oi_oa, inode, &ll_i2info(inode)->lli_fid,
-                                OBD_MD_FLTYPE | OBD_MD_FLATIME |
-                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
-                                OBD_MD_FLGROUP);
-                oinfo->oi_md = lsm;
-                oinfo->oi_capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
-                err = obd_sync_rqset(ll_i2sbi(inode)->ll_dt_exp, oinfo, 0,
-                                     OBD_OBJECT_EOF);
-                capa_put(oinfo->oi_capa);
+               err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF);
                 if (!rc)
                         rc = err;
-                OBDO_FREE(oinfo->oi_oa);
-                OBD_FREE_PTR(oinfo);
                 lli->lli_write_rc = rc < 0 ? rc : 0;
         }
 
@@ -2117,13 +2191,11 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 
         if ((file_lock->fl_flags & FL_FLOCK) &&
             (rc == 0 || file_lock->fl_type == F_UNLCK))
-                ll_flock_lock_file_wait(file, file_lock, (cmd == F_SETLKW));
-#ifdef HAVE_F_OP_FLOCK
+                flock_lock_file_wait(file, file_lock);
         if ((file_lock->fl_flags & FL_POSIX) &&
             (rc == 0 || file_lock->fl_type == F_UNLCK) &&
             !(flags & LDLM_FL_TEST_LOCK))
                 posix_lock_file_wait(file, file_lock);
-#endif
 
         RETURN(rc);
 }
@@ -2526,11 +2598,7 @@ struct file_operations ll_file_operations = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
-#ifdef HAVE_UNLOCKED_IOCTL
         .unlocked_ioctl = ll_file_ioctl,
-#else
-        .ioctl          = ll_file_ioctl,
-#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
@@ -2550,11 +2618,7 @@ struct file_operations ll_file_operations_flock = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
-#ifdef HAVE_UNLOCKED_IOCTL
         .unlocked_ioctl = ll_file_ioctl,
-#else
-        .ioctl          = ll_file_ioctl,
-#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
@@ -2567,9 +2631,7 @@ struct file_operations ll_file_operations_flock = {
 #endif
         .fsync          = ll_fsync,
         .flush          = ll_flush,
-#ifdef HAVE_F_OP_FLOCK
         .flock          = ll_file_flock,
-#endif
         .lock           = ll_file_flock
 };
 
@@ -2579,11 +2641,7 @@ struct file_operations ll_file_operations_noflock = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
-#ifdef HAVE_UNLOCKED_IOCTL
         .unlocked_ioctl = ll_file_ioctl,
-#else
-        .ioctl          = ll_file_ioctl,
-#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
@@ -2596,9 +2654,7 @@ struct file_operations ll_file_operations_noflock = {
 #endif
         .fsync          = ll_fsync,
         .flush          = ll_flush,
-#ifdef HAVE_F_OP_FLOCK
         .flock          = ll_file_noflock,
-#endif
         .lock           = ll_file_noflock
 };