* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Intel Corporation.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
{
ENTRY;
- op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
- ATTR_MTIME_SET | ATTR_CTIME_SET;
+ op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
+ ATTR_MTIME | ATTR_MTIME_SET |
+ ATTR_CTIME | ATTR_CTIME_SET;
if (!(och->och_flags & FMODE_WRITE))
goto out;
oinfo.oi_md = lsm;
oinfo.oi_oa = obdo;
- oinfo.oi_oa->o_id = lsm->lsm_object_id;
- oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
+ oinfo.oi_oa->o_oi = lsm->lsm_oi;
oinfo.oi_oa->o_mode = S_IFREG;
oinfo.oi_oa->o_ioepoch = ioepoch;
oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
capa, obdo, ioepoch, sync);
capa_put(capa);
if (rc == 0) {
+ struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi;
+
obdo_refresh_inode(inode, obdo, obdo->o_valid);
- CDEBUG(D_INODE,
- "objid "LPX64" size %llu, blocks %llu, blksize %lu\n",
- lsm ? lsm->lsm_object_id : 0, i_size_read(inode),
+ CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu,"
+ " blksize %lu\n", POSTID(oi), i_size_read(inode),
(unsigned long long)inode->i_blocks,
(unsigned long)ll_inode_blksize(inode));
}
io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
if (write) {
io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
- io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || IS_SYNC(inode);
+ io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
+ file->f_flags & O_DIRECT ||
+ IS_SYNC(inode);
}
io->ci_obj = ll_i2info(inode)->lli_clob;
io->ci_lockreq = CILR_MAYBE;
ssize_t result;
ENTRY;
+restart:
io = ccc_env_thread_io(env);
ll_io_init(io, file, iot == CIT_WRITE);
GOTO(out, result);
out:
cl_io_fini(env, io);
+ /* If any bit been read/written (result != 0), we just return
+ * short read/write instead of restart io. */
+ if (result == 0 && io->ci_need_restart) {
+ 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);
+ goto restart;
+ }
if (iot == CIT_READ) {
if (result >= 0)
ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
LPROC_LL_WRITE_BYTES, result);
fd->fd_write_failed = false;
- } else {
+ } else if (result != -ERESTARTSYS) {
fd->fd_write_failed = true;
}
}
}
#endif
-static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq,
+static int ll_lov_recreate(struct inode *inode, struct ost_id *oi,
obd_count ost_idx)
{
struct obd_export *exp = ll_i2dtexp(inode);
if (lsm2 == NULL)
GOTO(out, rc = -ENOMEM);
- oa->o_id = id;
- oa->o_seq = seq;
+ oa->o_oi = *oi;
oa->o_nlink = ost_idx;
oa->o_flags |= OBD_FL_RECREATE_OBJS;
oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
{
struct ll_recreate_obj ucreat;
+ struct ost_id oi;
ENTRY;
if (!cfs_capable(CFS_CAP_SYS_ADMIN))
sizeof(ucreat)))
RETURN(-EFAULT);
- RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0,
- ucreat.lrc_ost_idx));
+ ostid_set_seq_mdt0(&oi);
+ ostid_set_id(&oi, ucreat.lrc_id);
+ RETURN(ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx));
}
static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
{
struct lu_fid fid;
- obd_id id;
+ struct ost_id oi;
obd_count ost_idx;
ENTRY;
if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid)))
RETURN(-EFAULT);
- id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32);
+ fid_to_ostid(&fid, &oi);
ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
- RETURN(ll_lov_recreate(inode, id, 0, ost_idx));
+ RETURN(ll_lov_recreate(inode, &oi, ost_idx));
}
int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER))
GOTO(out, rc = -EOPNOTSUPP);
- fm_key.oa.o_id = lsm->lsm_object_id;
- fm_key.oa.o_seq = lsm->lsm_object_seq;
+ fm_key.oa.o_oi = lsm->lsm_oi;
fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
if (rc)
GOTO(gf_free, rc);
+
if (copy_to_user(arg, gfout, outsize))
rc = -EFAULT;
RETURN(rc);
}
-long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+struct ll_swap_stack {
+ struct iattr ia1, ia2;
+ __u64 dv1, dv2;
+ struct inode *inode1, *inode2;
+ bool check_dv1, check_dv2;
+};
+
+static int ll_swap_layouts(struct file *file1, struct file *file2,
+ struct lustre_swap_layouts *lsl)
{
- struct inode *inode = file->f_dentry->d_inode;
- struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
- int flags;
+ struct mdc_swap_layouts msl;
+ struct md_op_data *op_data;
+ __u32 gid;
+ __u64 dv;
+ struct ll_swap_stack *llss = NULL;
+ int rc;
- ENTRY;
+ OBD_ALLOC_PTR(llss);
+ if (llss == NULL)
+ RETURN(-ENOMEM);
+
+ llss->inode1 = file1->f_dentry->d_inode;
+ llss->inode2 = file2->f_dentry->d_inode;
+
+ if (!S_ISREG(llss->inode2->i_mode))
+ GOTO(free, rc = -EINVAL);
+
+ if (ll_permission(llss->inode1, MAY_WRITE, NULL) ||
+ ll_permission(llss->inode2, MAY_WRITE, NULL))
+ GOTO(free, rc = -EPERM);
+
+ if (llss->inode2->i_sb != llss->inode1->i_sb)
+ GOTO(free, rc = -EXDEV);
+
+ /* we use 2 bool because it is easier to swap than 2 bits */
+ if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
+ llss->check_dv1 = true;
+
+ if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
+ llss->check_dv2 = true;
+
+ /* we cannot use lsl->sl_dvX directly because we may swap them */
+ llss->dv1 = lsl->sl_dv1;
+ llss->dv2 = lsl->sl_dv2;
+
+ rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
+ if (rc == 0) /* same file, done! */
+ GOTO(free, rc = 0);
+
+ if (rc < 0) { /* sequentialize it */
+ swap(llss->inode1, llss->inode2);
+ swap(file1, file2);
+ swap(llss->dv1, llss->dv2);
+ swap(llss->check_dv1, llss->check_dv2);
+ }
+
+ gid = lsl->sl_gid;
+ if (gid != 0) { /* application asks to flush dirty cache */
+ rc = ll_get_grouplock(llss->inode1, file1, gid);
+ if (rc < 0)
+ GOTO(free, rc);
+
+ rc = ll_get_grouplock(llss->inode2, file2, gid);
+ if (rc < 0) {
+ ll_put_grouplock(llss->inode1, file1, gid);
+ GOTO(free, rc);
+ }
+ }
+
+ /* to be able to restore mtime and atime after swap
+ * we need to first save them */
+ if (lsl->sl_flags &
+ (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
+ llss->ia1.ia_mtime = llss->inode1->i_mtime;
+ llss->ia1.ia_atime = llss->inode1->i_atime;
+ llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
+ llss->ia2.ia_mtime = llss->inode2->i_mtime;
+ llss->ia2.ia_atime = llss->inode2->i_atime;
+ llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
+ }
+
+ /* ultimate check, before swaping the layouts we check if
+ * dataversion has changed (if requested) */
+ if (llss->check_dv1) {
+ rc = ll_data_version(llss->inode1, &dv, 0);
+ if (rc)
+ GOTO(putgl, rc);
+ if (dv != llss->dv1)
+ GOTO(putgl, rc = -EAGAIN);
+ }
+
+ if (llss->check_dv2) {
+ rc = ll_data_version(llss->inode2, &dv, 0);
+ if (rc)
+ GOTO(putgl, rc);
+ if (dv != llss->dv2)
+ GOTO(putgl, rc = -EAGAIN);
+ }
+
+ /* struct md_op_data is used to send the swap args to the mdt
+ * only flags is missing, so we use struct mdc_swap_layouts
+ * through the md_op_data->op_data */
+ /* flags from user space have to be converted before they are send to
+ * server, no flag is sent today, they are only used on the client */
+ msl.msl_flags = 0;
+ rc = -ENOMEM;
+ op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
+ 0, LUSTRE_OPC_ANY, &msl);
+ if (op_data != NULL) {
+ rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS,
+ ll_i2mdexp(llss->inode1),
+ sizeof(*op_data), op_data, NULL);
+ ll_finish_md_op_data(op_data);
+ }
+
+putgl:
+ if (gid != 0) {
+ ll_put_grouplock(llss->inode2, file2, gid);
+ ll_put_grouplock(llss->inode1, file1, gid);
+ }
+
+ /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
+ if (rc != 0)
+ GOTO(free, rc);
+
+ /* clear useless flags */
+ if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
+ llss->ia1.ia_valid &= ~ATTR_MTIME;
+ llss->ia2.ia_valid &= ~ATTR_MTIME;
+ }
+
+ if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
+ llss->ia1.ia_valid &= ~ATTR_ATIME;
+ llss->ia2.ia_valid &= ~ATTR_ATIME;
+ }
+
+ /* update time if requested */
+ rc = 0;
+ if (llss->ia2.ia_valid != 0) {
+ mutex_lock(&llss->inode1->i_mutex);
+ rc = ll_setattr(file1->f_dentry, &llss->ia2);
+ mutex_unlock(&llss->inode1->i_mutex);
+ }
+
+ if (llss->ia1.ia_valid != 0) {
+ int rc1;
+
+ mutex_lock(&llss->inode2->i_mutex);
+ rc1 = ll_setattr(file2->f_dentry, &llss->ia1);
+ mutex_unlock(&llss->inode2->i_mutex);
+ if (rc == 0)
+ rc = rc1;
+ }
+
+free:
+ if (llss != NULL)
+ OBD_FREE_PTR(llss);
+
+ RETURN(rc);
+}
+
+long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+ struct inode *inode = file->f_dentry->d_inode;
+ struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
+ int flags, rc;
+ ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino,
inode->i_generation, inode, cmd);
RETURN(ll_lov_setstripe(inode, file, arg));
case LL_IOC_LOV_SETEA:
RETURN(ll_lov_setea(inode, file, arg));
+ case LL_IOC_LOV_SWAP_LAYOUTS: {
+ struct file *file2;
+ struct lustre_swap_layouts lsl;
+
+ if (cfs_copy_from_user(&lsl, (char *)arg,
+ sizeof(struct lustre_swap_layouts)))
+ RETURN(-EFAULT);
+
+ if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
+ RETURN(-EPERM);
+
+ file2 = fget(lsl.sl_fd);
+ if (file2 == NULL)
+ RETURN(-EBADF);
+
+ rc = -EPERM;
+ if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
+ rc = ll_swap_layouts(file, file2, &lsl);
+ fput(file2);
+ RETURN(rc);
+ }
case LL_IOC_LOV_GETSTRIPE:
RETURN(ll_lov_getstripe(inode, arg));
case LL_IOC_RECREATE_OBJ:
eof = i_size_read(inode);
}
- retval = generic_file_llseek_size(file, offset, origin,
+ retval = ll_generic_file_llseek_size(file, offset, origin,
ll_file_maxbytes(inode), eof);
RETURN(retval);
}
* Return how many pages have been written.
*/
int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
- enum cl_fsync_mode mode)
+ enum cl_fsync_mode mode, int ignore_layout)
{
struct cl_env_nest nest;
struct lu_env *env;
io = ccc_env_thread_io(env);
io->ci_obj = cl_i2info(inode)->lli_clob;
- io->ci_ignore_layout = 1;
+ io->ci_ignore_layout = ignore_layout;
/* initialize parameters for sync */
fio = &io->u.ci_fsync;
RETURN(result);
}
+/*
+ * When dentry is provided (the 'else' case), *file->f_dentry may be
+ * null and dentry must be used directly rather than pulled from
+ * *file->f_dentry as is done otherwise.
+ */
+
#ifdef HAVE_FILE_FSYNC_4ARGS
-int ll_fsync(struct file *file, loff_t start, loff_t end, int data)
+int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
+{
+ struct dentry *dentry = file->f_dentry;
#elif defined(HAVE_FILE_FSYNC_2ARGS)
-int ll_fsync(struct file *file, int data)
+int ll_fsync(struct file *file, int datasync)
+{
+ struct dentry *dentry = file->f_dentry;
#else
-int ll_fsync(struct file *file, struct dentry *dentry, int data)
-#endif
+int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
{
- struct inode *inode = file->f_dentry->d_inode;
+#endif
+ struct inode *inode = dentry->d_inode;
struct ll_inode_info *lli = ll_i2info(inode);
struct ptlrpc_request *req;
struct obd_capa *oc;
if (!err)
ptlrpc_req_finished(req);
- if (data) {
+ if (datasync && S_ISREG(inode->i_mode)) {
struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
- CL_FSYNC_ALL);
+ CL_FSYNC_ALL, 0);
if (rc == 0 && err < 0)
rc = err;
if (rc < 0)
here to preserve get_cwd functionality on 2.6.
Bug 10503 */
if (!dentry->d_inode->i_nlink)
- d_lustre_invalidate(dentry);
+ d_lustre_invalidate(dentry, 0);
ll_lookup_finish_locks(&oit, dentry);
} else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
RETURN(result);
}
+/* Fetch layout from MDT with getxattr request, if it's not ready yet */
+static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
+
+{
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ struct obd_capa *oc;
+ struct ptlrpc_request *req;
+ struct mdt_body *body;
+ void *lvbdata;
+ void *lmm;
+ int lmmsize;
+ int rc;
+ ENTRY;
+
+ if (lock->l_lvb_data != NULL)
+ RETURN(0);
+
+ /* if layout lock was granted right away, the layout is returned
+ * within DLM_LVB of dlm reply; otherwise if the lock was ever
+ * blocked and then granted via completion ast, we have to fetch
+ * layout here. Please note that we can't use the LVB buffer in
+ * completion AST because it doesn't have a large enough buffer */
+ oc = ll_mdscapa_get(inode);
+ rc = ll_get_max_mdsize(sbi, &lmmsize);
+ if (rc == 0)
+ rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
+ OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
+ lmmsize, 0, &req);
+ capa_put(oc);
+ if (rc < 0)
+ RETURN(rc);
+
+ body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+ if (body == NULL || body->eadatasize > lmmsize)
+ GOTO(out, rc = -EPROTO);
+
+ lmmsize = body->eadatasize;
+ if (lmmsize == 0) /* empty layout */
+ GOTO(out, rc = 0);
+
+ lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
+ if (lmm == NULL)
+ GOTO(out, rc = -EFAULT);
+
+ OBD_ALLOC_LARGE(lvbdata, lmmsize);
+ if (lvbdata == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ memcpy(lvbdata, lmm, lmmsize);
+ lock_res_and_lock(lock);
+ if (lock->l_lvb_data == NULL) {
+ lock->l_lvb_data = lvbdata;
+ lock->l_lvb_len = lmmsize;
+ lvbdata = NULL;
+ }
+ unlock_res_and_lock(lock);
+
+ if (lvbdata != NULL)
+ OBD_FREE_LARGE(lvbdata, lmmsize);
+ EXIT;
+
+out:
+ ptlrpc_req_finished(req);
+ return rc;
+}
+
/**
* Apply the layout to the inode. Layout lock is held and will be released
* in this function.
struct cl_object_conf conf;
int rc = 0;
bool lvb_ready;
+ bool wait_layout = false;
ENTRY;
LASSERT(lustre_handle_is_used(lockh));
LDLM_DEBUG(lock, "File %p/"DFID" being reconfigured: %d.\n",
inode, PFID(&lli->lli_fid), reconf);
+ /* in case this is a caching lock and reinstate with new inode */
+ md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
+
lock_res_and_lock(lock);
lvb_ready = !!(lock->l_flags & LDLM_FL_LVB_READY);
unlock_res_and_lock(lock);
/* checking lvb_ready is racy but this is okay. The worst case is
* that multi processes may configure the file on the same time. */
if (lvb_ready || !reconf) {
- LDLM_LOCK_PUT(lock);
-
rc = -ENODATA;
if (lvb_ready) {
/* layout_gen must be valid if layout lock is not
*gen = lli->lli_layout_gen;
rc = 0;
}
- ldlm_lock_decref(lockh, mode);
- RETURN(rc);
+ GOTO(out, rc);
}
+ rc = ll_layout_fetch(inode, lock);
+ if (rc < 0)
+ GOTO(out, rc);
+
/* for layout lock, lmm is returned in lock's lvb.
* lvb_data is immutable if the lock is held so it's safe to access it
* without res lock. See the description in ldlm_lock_decref_internal()
rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
lock->l_lvb_data, lock->l_lvb_len);
if (rc >= 0) {
+ *gen = LL_LAYOUT_GEN_EMPTY;
if (md.lsm != NULL)
*gen = md.lsm->lsm_layout_gen;
rc = 0;
PFID(&lli->lli_fid), rc);
}
}
- if (rc < 0) {
- LDLM_LOCK_PUT(lock);
- ldlm_lock_decref(lockh, mode);
- RETURN(rc);
- }
+ if (rc < 0)
+ GOTO(out, rc);
/* set layout to file. Unlikely this will fail as old layout was
* surely eliminated */
conf.coc_lock = lock;
conf.u.coc_md = &md;
rc = ll_layout_conf(inode, &conf);
- LDLM_LOCK_PUT(lock);
-
- ldlm_lock_decref(lockh, mode);
if (md.lsm != NULL)
obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
+ /* refresh layout failed, need to wait */
+ wait_layout = rc == -EBUSY;
+ EXIT;
+
+out:
+ LDLM_LOCK_PUT(lock);
+ ldlm_lock_decref(lockh, mode);
+
/* wait for IO to complete if it's still being used. */
- if (rc == -EBUSY) {
+ if (wait_layout) {
CDEBUG(D_INODE, "%s: %p/"DFID" wait for layout reconf.\n",
ll_get_fsname(inode->i_sb, NULL, 0),
inode, PFID(&lli->lli_fid));
CDEBUG(D_INODE, "file: "DFID" waiting layout return: %d.\n",
PFID(&lli->lli_fid), rc);
}
-
RETURN(rc);
}
.ei_mode = LCK_CR,
.ei_cb_bl = ll_md_blocking_ast,
.ei_cb_cp = ldlm_completion_ast,
- .ei_cbdata = inode };
+ .ei_cbdata = NULL };
int rc;
ENTRY;
- *gen = LL_LAYOUT_GEN_ZERO;
+ *gen = lli->lli_layout_gen;
if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
RETURN(0);