*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
case CIT_WRITE:
/* don't need lock here to check lli_layout_gen as we have held
* extent lock and GROUP lock has to hold to swap layout */
- if (ll_layout_version_get(lli) != vio->vui_layout_gen) {
+ if (ll_layout_version_get(lli) != vio->vui_layout_gen ||
+ OBD_FAIL_CHECK_RESET(OBD_FAIL_LLITE_LOST_LAYOUT, 0)) {
io->ci_need_restart = 1;
- /* this will return application a short read/write */
+ /* this will cause a short read/write */
io->ci_continue = 0;
rc = false;
}
* --bug 17336 */
loff_t size = i_size_read(inode);
unsigned long cur_index = start >>
- PAGE_CACHE_SHIFT;
+ PAGE_SHIFT;
if ((size == 0 && cur_index != 0) ||
- (((size - 1) >> PAGE_CACHE_SHIFT) <
+ (((size - 1) >> PAGE_SHIFT) <
cur_index))
*exceed = 1;
}
if (i_size_read(inode) < kms) {
i_size_write(inode, kms);
CDEBUG(D_VFSTRACE,
- DFID" updating i_size "LPU64"\n",
+ DFID" updating i_size %llu\n",
PFID(lu_object_fid(&obj->co_lu)),
(__u64)i_size_read(inode));
}
if (vio->vui_fd && (vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
descr->cld_mode = CLM_GROUP;
descr->cld_gid = vio->vui_fd->fd_grouplock.lg_gid;
+ enqflags |= CEF_LOCK_MATCH;
} else {
descr->cld_mode = mode;
}
struct vvp_io *vio = cl2vvp_io(env, ios);
struct inode *inode = vvp_object_inode(ios->cis_obj);
- LASSERT(inode == vio->vui_fd->fd_file->f_dentry->d_inode);
- vio->u.fault.ft_mtime = LTIME_S(inode->i_mtime);
+ LASSERT(inode == file_inode(vio->vui_fd->fd_file));
+ vio->u.fault.ft_mtime = inode->i_mtime.tv_sec;
return 0;
}
struct cl_object *obj = io->ci_obj;
struct vvp_io *vio = cl2vvp_io(env, ios);
struct inode *inode = vvp_object_inode(obj);
+ int rc;
CLOBINVRNT(env, obj, vvp_object_invariant(obj));
CDEBUG(D_VFSTRACE, DFID" ignore/verify layout %d/%d, layout version %d "
- "restore needed %d\n",
+ "need write layout %d, restore needed %d\n",
PFID(lu_object_fid(&obj->co_lu)),
io->ci_ignore_layout, io->ci_verify_layout,
- vio->vui_layout_gen, io->ci_restore_needed);
-
- if (io->ci_restore_needed == 1) {
- int rc;
+ vio->vui_layout_gen, io->ci_need_write_intent,
+ io->ci_restore_needed);
+ if (io->ci_restore_needed) {
/* file was detected release, we need to restore it
* before finishing the io
*/
/* if restore registration failed, no restart,
* we will return -ENODATA */
/* The layout will change after restore, so we need to
- * block on layout lock hold by the MDT
+ * block on layout lock held by the MDT
* as MDT will not send new layout in lvb (see LU-3124)
* we have to explicitly fetch it, all this will be done
* by ll_layout_refresh()
}
}
+ /**
+ * dynamic layout change needed, send layout intent
+ * RPC.
+ */
+ if (io->ci_need_write_intent) {
+ loff_t start = 0;
+ loff_t end = OBD_OBJECT_EOF;
+
+ io->ci_need_write_intent = 0;
+
+ LASSERT(io->ci_type == CIT_WRITE ||
+ cl_io_is_trunc(io) || cl_io_is_mkwrite(io));
+
+ if (io->ci_type == CIT_WRITE) {
+ if (!cl_io_is_append(io)) {
+ start = io->u.ci_rw.rw_range.cir_pos;
+ end = start + io->u.ci_rw.rw_range.cir_count;
+ }
+ } else if (cl_io_is_trunc(io)) {
+ end = io->u.ci_setattr.sa_attr.lvb_size;
+ } else { /* mkwrite */
+ pgoff_t index = io->u.ci_fault.ft_index;
+
+ start = cl_offset(io->ci_obj, index);
+ end = cl_offset(io->ci_obj, index + 1);
+ }
+
+ CDEBUG(D_VFSTRACE, DFID" write layout, type %u [%llu, %llu)\n",
+ PFID(lu_object_fid(&obj->co_lu)), io->ci_type,
+ start, end);
+ rc = ll_layout_write_intent(inode, start, end);
+ io->ci_result = rc;
+ if (!rc)
+ io->ci_need_restart = 1;
+ }
+
if (!io->ci_ignore_layout && io->ci_verify_layout) {
__u32 gen = 0;
/* today successful restore is the only possible
* case */
/* restore was done, clear restoring state */
- ll_i2info(vvp_object_inode(obj))->lli_flags &=
- ~LLIF_FILE_RESTORING;
+ ll_file_clear_flag(ll_i2info(vvp_object_inode(obj)),
+ LLIF_FILE_RESTORING);
}
}
}
return CLM_READ;
}
-static int vvp_mmap_locks(const struct lu_env *env,
- struct vvp_io *vio, struct cl_io *io)
+static int vvp_mmap_locks(const struct lu_env *env, struct cl_io *io)
{
struct vvp_thread_info *vti = vvp_env_info(env);
- struct mm_struct *mm = current->mm;
- struct vm_area_struct *vma;
- struct cl_lock_descr *descr = &vti->vti_descr;
- ldlm_policy_data_t policy;
- unsigned long addr;
- unsigned long seg;
- ssize_t count;
- int result = 0;
- ENTRY;
-
- LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
-
- if (!cl_is_normalio(env, io))
- RETURN(0);
-
- if (vio->vui_iov == NULL) /* nfs or loop back device write */
+ struct mm_struct *mm = current->mm;
+ struct vm_area_struct *vma;
+ struct cl_lock_descr *descr = &vti->vti_descr;
+ union ldlm_policy_data policy;
+ struct iovec iov;
+ struct iov_iter i;
+ int result = 0;
+ ENTRY;
+
+ LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
+
+ if (!cl_is_normalio(env, io))
RETURN(0);
/* No MM (e.g. NFS)? No vmas too. */
if (mm == NULL)
RETURN(0);
- for (seg = 0; seg < vio->vui_nrsegs; seg++) {
- const struct iovec *iv = &vio->vui_iov[seg];
+ iov_for_each(iov, i, io->u.ci_rw.rw_iter) {
+ unsigned long addr = (unsigned long)iov.iov_base;
+ size_t count = iov.iov_len;
- addr = (unsigned long)iv->iov_base;
- count = iv->iov_len;
if (count == 0)
continue;
down_read(&mm->mmap_sem);
while((vma = our_vma(mm, addr, count)) != NULL) {
- struct inode *inode = vma->vm_file->f_dentry->d_inode;
+ struct dentry *de = file_dentry(vma->vm_file);
+ struct inode *inode = de->d_inode;
int flags = CEF_MUST;
if (ll_file_nolock(vma->vm_file)) {
if (!cl_is_normalio(env, io))
return;
- LASSERT(vio->vui_tot_nrsegs >= vio->vui_nrsegs);
- LASSERT(vio->vui_tot_count >= nob);
-
- vio->vui_iov += vio->vui_nrsegs;
- vio->vui_tot_nrsegs -= vio->vui_nrsegs;
- vio->vui_tot_count -= nob;
-
- /* update the iov */
- if (vio->vui_iov_olen > 0) {
- struct iovec *iv;
-
- vio->vui_iov--;
- vio->vui_tot_nrsegs++;
- iv = &vio->vui_iov[0];
- if (io->ci_continue) {
- iv->iov_base += iv->iov_len;
- LASSERT(vio->vui_iov_olen > iv->iov_len);
- iv->iov_len = vio->vui_iov_olen - iv->iov_len;
- } else {
- /* restore the iov_len, in case of restart io. */
- iv->iov_len = vio->vui_iov_olen;
- }
- vio->vui_iov_olen = 0;
- }
-}
-
-static void vvp_io_update_iov(const struct lu_env *env,
- struct vvp_io *vio, struct cl_io *io)
-{
- int i;
- size_t size = io->u.ci_rw.crw_count;
-
- vio->vui_iov_olen = 0;
- if (!cl_is_normalio(env, io) || vio->vui_tot_nrsegs == 0)
- return;
-
- for (i = 0; i < vio->vui_tot_nrsegs; i++) {
- struct iovec *iv = &vio->vui_iov[i];
-
- if (iv->iov_len < size) {
- size -= iv->iov_len;
- } else {
- if (iv->iov_len > size) {
- vio->vui_iov_olen = iv->iov_len;
- iv->iov_len = size;
- }
- break;
- }
+ vio->vui_tot_count -= nob;
+ if (io->ci_pio) {
+ iov_iter_advance(&io->u.ci_rw.rw_iter, nob);
+ io->u.ci_rw.rw_iocb.ki_pos = io->u.ci_rw.rw_range.cir_pos;
+#ifdef HAVE_KIOCB_KI_LEFT
+ io->u.ci_rw.rw_iocb.ki_left = vio->vui_tot_count;
+#elif defined(HAVE_KI_NBYTES)
+ io->u.ci_rw.rw_iocb.ki_nbytes = vio->vui_tot_count;
+#endif
+ } else {
+ /* It was truncated to stripe size in vvp_io_rw_lock() */
+ iov_iter_reexpand(&io->u.ci_rw.rw_iter, vio->vui_tot_count);
}
-
- vio->vui_nrsegs = i + 1;
- LASSERTF(vio->vui_tot_nrsegs >= vio->vui_nrsegs,
- "tot_nrsegs: %lu, nrsegs: %lu\n",
- vio->vui_tot_nrsegs, vio->vui_nrsegs);
}
static int vvp_io_rw_lock(const struct lu_env *env, struct cl_io *io,
enum cl_lock_mode mode, loff_t start, loff_t end)
{
- struct vvp_io *vio = vvp_env_io(env);
int result;
int ast_flags = 0;
LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
ENTRY;
- vvp_io_update_iov(env, vio, io);
+ if (cl_is_normalio(env, io))
+ iov_iter_truncate(&io->u.ci_rw.rw_iter,
+ io->u.ci_rw.rw_range.cir_count);
- if (io->u.ci_rw.crw_nonblock)
+ if (io->u.ci_rw.rw_nonblock)
ast_flags |= CEF_NONBLOCK;
- result = vvp_mmap_locks(env, vio, io);
+ result = vvp_mmap_locks(env, io);
if (result == 0)
result = vvp_io_one_lock(env, io, ast_flags, mode, start, end);
- RETURN(result);
+ RETURN(result);
}
static int vvp_io_read_lock(const struct lu_env *env,
const struct cl_io_slice *ios)
{
- struct cl_io *io = ios->cis_io;
- struct cl_io_rw_common *rd = &io->u.ci_rd.rd;
- int result;
+ struct cl_io *io = ios->cis_io;
+ struct cl_io_range *range = &io->u.ci_rw.rw_range;
+ int rc;
ENTRY;
- result = vvp_io_rw_lock(env, io, CLM_READ, rd->crw_pos,
- rd->crw_pos + rd->crw_count - 1);
- RETURN(result);
+ rc = vvp_io_rw_lock(env, io, CLM_READ, range->cir_pos,
+ range->cir_pos + range->cir_count - 1);
+ RETURN(rc);
}
static int vvp_io_fault_lock(const struct lu_env *env,
static int vvp_io_write_lock(const struct lu_env *env,
const struct cl_io_slice *ios)
{
- struct cl_io *io = ios->cis_io;
- loff_t start;
- loff_t end;
+ struct cl_io *io = ios->cis_io;
+ loff_t start;
+ loff_t end;
+ int rc;
- if (io->u.ci_wr.wr_append) {
- start = 0;
- end = OBD_OBJECT_EOF;
- } else {
- start = io->u.ci_wr.wr.crw_pos;
- end = start + io->u.ci_wr.wr.crw_count - 1;
- }
- return vvp_io_rw_lock(env, io, CLM_WRITE, start, end);
+ ENTRY;
+ if (io->u.ci_rw.rw_append) {
+ start = 0;
+ end = OBD_OBJECT_EOF;
+ } else {
+ start = io->u.ci_rw.rw_range.cir_pos;
+ end = start + io->u.ci_rw.rw_range.cir_count - 1;
+ }
+ rc = vvp_io_rw_lock(env, io, CLM_WRITE, start, end);
+ RETURN(rc);
}
static int vvp_io_setattr_iter_init(const struct lu_env *env,
return result;
}
-static int vvp_io_setattr_trunc(const struct lu_env *env,
- const struct cl_io_slice *ios,
- struct inode *inode, loff_t size)
-{
- inode_dio_wait(inode);
- return 0;
-}
-
static int vvp_io_setattr_time(const struct lu_env *env,
const struct cl_io_slice *ios)
{
static int vvp_io_setattr_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
- struct cl_io *io = ios->cis_io;
- struct inode *inode = vvp_object_inode(io->ci_obj);
- int result = 0;
+ struct cl_io *io = ios->cis_io;
+ struct inode *inode = vvp_object_inode(io->ci_obj);
+ struct ll_inode_info *lli = ll_i2info(inode);
- mutex_lock(&inode->i_mutex);
- if (cl_io_is_trunc(io))
- result = vvp_io_setattr_trunc(env, ios, inode,
- io->u.ci_setattr.sa_attr.lvb_size);
- if (result == 0 && io->u.ci_setattr.sa_valid & TIMES_SET_FLAGS)
- result = vvp_io_setattr_time(env, ios);
- return result;
+ if (cl_io_is_trunc(io)) {
+ down_write(&lli->lli_trunc_sem);
+ inode_lock(inode);
+ inode_dio_wait(inode);
+ } else {
+ inode_lock(inode);
+ }
+
+ if (io->u.ci_setattr.sa_valid & TIMES_SET_FLAGS)
+ return vvp_io_setattr_time(env, ios);
+
+ return 0;
}
static void vvp_io_setattr_end(const struct lu_env *env,
const struct cl_io_slice *ios)
{
- struct cl_io *io = ios->cis_io;
- struct inode *inode = vvp_object_inode(io->ci_obj);
+ struct cl_io *io = ios->cis_io;
+ struct inode *inode = vvp_object_inode(io->ci_obj);
+ struct ll_inode_info *lli = ll_i2info(inode);
if (cl_io_is_trunc(io)) {
/* Truncate in memory pages - they must be clean pages
* because osc has already notified to destroy osc_extents. */
vvp_do_vmtruncate(inode, io->u.ci_setattr.sa_attr.lvb_size);
inode_dio_write_done(inode);
+ inode_unlock(inode);
+ up_write(&lli->lli_trunc_sem);
+ } else {
+ inode_unlock(inode);
}
- mutex_unlock(&inode->i_mutex);
}
static void vvp_io_setattr_fini(const struct lu_env *env,
const struct cl_io_slice *ios)
{
+ bool restore_needed = ios->cis_io->ci_restore_needed;
+ struct inode *inode = vvp_object_inode(ios->cis_obj);
+
vvp_io_fini(env, ios);
+
+ if (restore_needed && !ios->cis_io->ci_restore_needed) {
+ /* restore finished, set data modified flag for HSM */
+ ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED);
+ }
}
static int vvp_io_read_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
- struct vvp_io *vio = cl2vvp_io(env, ios);
- struct cl_io *io = ios->cis_io;
- struct cl_object *obj = io->ci_obj;
- struct inode *inode = vvp_object_inode(obj);
- struct file *file = vio->vui_fd->fd_file;
-
- int result;
- loff_t pos = io->u.ci_rd.rd.crw_pos;
- long cnt = io->u.ci_rd.rd.crw_count;
- long tot = vio->vui_tot_count;
- int exceed = 0;
+ struct vvp_io *vio = cl2vvp_io(env, ios);
+ struct cl_io *io = ios->cis_io;
+ struct cl_object *obj = io->ci_obj;
+ struct inode *inode = vvp_object_inode(obj);
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct file *file = vio->vui_fd->fd_file;
+ struct cl_io_range *range = &io->u.ci_rw.rw_range;
+ loff_t pos = range->cir_pos; /* for generic_file_splice_read() only */
+ size_t tot = vio->vui_tot_count;
+ int exceed = 0;
+ int result;
CLOBINVRNT(env, obj, vvp_object_invariant(obj));
- CDEBUG(D_VFSTRACE, "read: -> [%lli, %lli)\n", pos, pos + cnt);
+ CDEBUG(D_VFSTRACE, "%s: read [%llu, %llu)\n",
+ file_dentry(file)->d_name.name,
+ range->cir_pos, range->cir_pos + range->cir_count);
+
+ if (vio->vui_io_subtype == IO_NORMAL)
+ down_read(&lli->lli_trunc_sem);
if (!can_populate_pages(env, io, inode))
return 0;
- result = vvp_prep_size(env, obj, io, pos, tot, &exceed);
- if (result != 0)
- return result;
- else if (exceed != 0)
- goto out;
+ result = vvp_prep_size(env, obj, io, range->cir_pos, tot, &exceed);
+ if (result != 0)
+ return result;
+ else if (exceed != 0)
+ goto out;
- LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu,
- "Read ino %lu, %lu bytes, offset %lld, size %llu\n",
- inode->i_ino, cnt, pos, i_size_read(inode));
+ LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu,
+ "Read ino %lu, %lu bytes, offset %lld, size %llu\n",
+ inode->i_ino, range->cir_count, range->cir_pos,
+ i_size_read(inode));
/* turn off the kernel's read-ahead */
vio->vui_fd->fd_file->f_ra.ra_pages = 0;
/* initialize read-ahead window once per syscall */
if (!vio->vui_ra_valid) {
vio->vui_ra_valid = true;
- vio->vui_ra_start = cl_index(obj, pos);
- vio->vui_ra_count = cl_index(obj, tot + PAGE_CACHE_SIZE - 1);
+ vio->vui_ra_start = cl_index(obj, range->cir_pos);
+ vio->vui_ra_count = cl_index(obj, tot + PAGE_SIZE - 1);
ll_ras_enter(file);
}
file_accessed(file);
switch (vio->vui_io_subtype) {
case IO_NORMAL:
- LASSERT(vio->vui_iocb->ki_pos == pos);
- result = generic_file_aio_read(vio->vui_iocb,
- vio->vui_iov, vio->vui_nrsegs,
- vio->vui_iocb->ki_pos);
+ LASSERTF(io->u.ci_rw.rw_iocb.ki_pos == range->cir_pos,
+ "ki_pos %lld [%lld, %lld)\n",
+ io->u.ci_rw.rw_iocb.ki_pos,
+ range->cir_pos, range->cir_pos + range->cir_count);
+ result = generic_file_read_iter(&io->u.ci_rw.rw_iocb,
+ &io->u.ci_rw.rw_iter);
break;
case IO_SPLICE:
result = generic_file_splice_read(file, &pos,
- vio->u.splice.vui_pipe, cnt,
+ vio->u.splice.vui_pipe,
+ range->cir_count,
vio->u.splice.vui_flags);
/* LU-1109: do splice read stripe by stripe otherwise if it
* may make nfsd stuck if this read occupied all internal pipe
out:
if (result >= 0) {
- if (result < cnt)
+ if (result < range->cir_count)
io->ci_continue = 0;
io->ci_nob += result;
ll_rw_stats_tally(ll_i2sbi(inode), current->pid, vio->vui_fd,
- pos, result, READ);
+ range->cir_pos, result, READ);
result = 0;
}
SetPageUptodate(cl_page_vmpage(page));
cl_page_disown(env, io, page);
- /* held in ll_cl_init() */
lu_ref_del(&page->cp_reference, "cl_io", io);
cl_page_put(env, page);
}
cl_page_disown(env, io, page);
- /* held in ll_cl_init() */
lu_ref_del(&page->cp_reference, "cl_io", cl_io_top(io));
cl_page_put(env, page);
}
cl_page_disown(env, io, page);
- /* held in ll_cl_init() */
lu_ref_del(&page->cp_reference, "cl_io", io);
cl_page_put(env, page);
}
static int vvp_io_write_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
- struct vvp_io *vio = cl2vvp_io(env, ios);
- struct cl_io *io = ios->cis_io;
- struct cl_object *obj = io->ci_obj;
- struct inode *inode = vvp_object_inode(obj);
- ssize_t result = 0;
- loff_t pos = io->u.ci_wr.wr.crw_pos;
- size_t cnt = io->u.ci_wr.wr.crw_count;
+ struct vvp_io *vio = cl2vvp_io(env, ios);
+ struct cl_io *io = ios->cis_io;
+ struct cl_object *obj = io->ci_obj;
+ struct inode *inode = vvp_object_inode(obj);
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct file *file = vio->vui_fd->fd_file;
+ struct cl_io_range *range = &io->u.ci_rw.rw_range;
+ bool lock_inode = !lli->lli_inode_locked &&
+ !IS_NOSEC(inode);
+ ssize_t result = 0;
+ ENTRY;
- ENTRY;
+ if (vio->vui_io_subtype == IO_NORMAL)
+ down_read(&lli->lli_trunc_sem);
if (!can_populate_pages(env, io, inode))
RETURN(0);
* out-of-order writes.
*/
ll_merge_attr(env, inode);
- pos = io->u.ci_wr.wr.crw_pos = i_size_read(inode);
- vio->vui_iocb->ki_pos = pos;
+ range->cir_pos = i_size_read(inode);
+ io->u.ci_rw.rw_iocb.ki_pos = range->cir_pos;
} else {
- LASSERT(vio->vui_iocb->ki_pos == pos);
+ LASSERTF(io->u.ci_rw.rw_iocb.ki_pos == range->cir_pos,
+ "ki_pos %lld [%lld, %lld)\n",
+ io->u.ci_rw.rw_iocb.ki_pos,
+ range->cir_pos, range->cir_pos + range->cir_count);
}
- CDEBUG(D_VFSTRACE, "write: [%lli, %lli)\n", pos, pos + (long long)cnt);
+ CDEBUG(D_VFSTRACE, "%s: write [%llu, %llu)\n",
+ file_dentry(file)->d_name.name,
+ range->cir_pos, range->cir_pos + range->cir_count);
+
+ /* The maximum Lustre file size is variable, based on the OST maximum
+ * object size and number of stripes. This needs another check in
+ * addition to the VFS checks earlier. */
+ if (range->cir_pos + range->cir_count > ll_file_maxbytes(inode)) {
+ CDEBUG(D_INODE,
+ "%s: file %s ("DFID") offset %llu > maxbytes %llu\n",
+ ll_get_fsname(inode->i_sb, NULL, 0),
+ file_dentry(file)->d_name.name,
+ PFID(ll_inode2fid(inode)),
+ range->cir_pos + range->cir_count,
+ ll_file_maxbytes(inode));
+ RETURN(-EFBIG);
+ }
- if (vio->vui_iov == NULL) {
- /* from a temp io in ll_cl_init(). */
- result = 0;
- } else {
- /*
- * When using the locked AIO function (generic_file_aio_write())
- * testing has shown the inode mutex to be a limiting factor
- * with multi-threaded single shared file performance. To get
- * around this, we now use the lockless version. To maintain
- * consistency, proper locking to protect against writes,
- * trucates, etc. is handled in the higher layers of lustre.
- */
- result = __generic_file_aio_write(vio->vui_iocb,
- vio->vui_iov, vio->vui_nrsegs,
- &vio->vui_iocb->ki_pos);
- if (result > 0 || result == -EIOCBQUEUED) {
- ssize_t err;
-
- err = generic_write_sync(vio->vui_iocb->ki_filp,
- pos, result);
- if (err < 0 && result > 0)
- result = err;
- }
+ /* Tests to verify we take the i_mutex correctly */
+ if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_IMUTEX_SEC) && !lock_inode)
+ RETURN(-EINVAL);
+ if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_IMUTEX_NOSEC) && lock_inode)
+ RETURN(-EINVAL);
+
+ /*
+ * When using the locked AIO function (generic_file_aio_write())
+ * testing has shown the inode mutex to be a limiting factor
+ * with multi-threaded single shared file performance. To get
+ * around this, we now use the lockless version. To maintain
+ * consistency, proper locking to protect against writes,
+ * trucates, etc. is handled in the higher layers of lustre.
+ */
+ if (lock_inode)
+ inode_lock(inode);
+ result = __generic_file_write_iter(&io->u.ci_rw.rw_iocb,
+ &io->u.ci_rw.rw_iter);
+ if (lock_inode)
+ inode_unlock(inode);
+
+ if (result > 0 || result == -EIOCBQUEUED)
+#ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
+ result = generic_write_sync(&io->u.ci_rw.rw_iocb, result);
+#else
+ {
+ ssize_t err;
+
+ err = generic_write_sync(io->u.ci_rw.rw_iocb.ki_filp,
+ range->cir_pos, result);
+ if (err < 0 && result > 0)
+ result = err;
}
+#endif
+
if (result > 0) {
result = vvp_io_write_commit(env, io);
if (vio->u.write.vui_written > 0) {
result = vio->u.write.vui_written;
- io->ci_nob += result;
-
- CDEBUG(D_VFSTRACE, "write: nob %zd, result: %zd\n",
+ CDEBUG(D_VFSTRACE, "%s: write nob %zd, result: %zd\n",
+ file_dentry(file)->d_name.name,
io->ci_nob, result);
+ io->ci_nob += result;
}
}
if (result > 0) {
- struct ll_inode_info *lli = ll_i2info(inode);
-
- spin_lock(&lli->lli_lock);
- lli->lli_flags |= LLIF_DATA_MODIFIED;
- spin_unlock(&lli->lli_lock);
+ ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED);
- if (result < cnt)
+ if (result < range->cir_count)
io->ci_continue = 0;
ll_rw_stats_tally(ll_i2sbi(inode), current->pid,
- vio->vui_fd, pos, result, WRITE);
+ vio->vui_fd, range->cir_pos, result, WRITE);
result = 0;
}
RETURN(result);
}
+static void vvp_io_rw_end(const struct lu_env *env,
+ const struct cl_io_slice *ios)
+{
+ struct vvp_io *vio = cl2vvp_io(env, ios);
+ struct inode *inode = vvp_object_inode(ios->cis_obj);
+ struct ll_inode_info *lli = ll_i2info(inode);
+
+ if (vio->vui_io_subtype == IO_NORMAL)
+ up_read(&lli->lli_trunc_sem);
+}
+
static int vvp_io_kernel_fault(struct vvp_fault_io *cfio)
{
struct vm_fault *vmf = cfio->ft_vmf;
- cfio->ft_flags = filemap_fault(cfio->ft_vma, vmf);
+ cfio->ft_flags = ll_filemap_fault(cfio->ft_vma, vmf);
cfio->ft_flags_valid = 1;
if (vmf->page) {
LL_CDEBUG_PAGE(D_PAGE, vmf->page, "got addr %p type NOPAGE\n",
- vmf->virtual_address);
+ get_vmf_address(vmf));
if (unlikely(!(cfio->ft_flags & VM_FAULT_LOCKED))) {
lock_page(vmf->page);
cfio->ft_flags |= VM_FAULT_LOCKED;
}
if (cfio->ft_flags & VM_FAULT_SIGBUS) {
- CDEBUG(D_PAGE, "got addr %p - SIGBUS\n", vmf->virtual_address);
+ CDEBUG(D_PAGE, "got addr %p - SIGBUS\n", get_vmf_address(vmf));
return -EFAULT;
}
if (cfio->ft_flags & VM_FAULT_OOM) {
- CDEBUG(D_PAGE, "got addr %p - OOM\n", vmf->virtual_address);
+ CDEBUG(D_PAGE, "got addr %p - OOM\n", get_vmf_address(vmf));
return -ENOMEM;
}
static int vvp_io_fault_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
- struct vvp_io *vio = cl2vvp_io(env, ios);
- struct cl_io *io = ios->cis_io;
- struct cl_object *obj = io->ci_obj;
- struct inode *inode = vvp_object_inode(obj);
- struct cl_fault_io *fio = &io->u.ci_fault;
- struct vvp_fault_io *cfio = &vio->u.fault;
- loff_t offset;
- int result = 0;
- struct page *vmpage = NULL;
- struct cl_page *page;
- loff_t size;
- pgoff_t last_index;
+ struct vvp_io *vio = cl2vvp_io(env, ios);
+ struct cl_io *io = ios->cis_io;
+ struct cl_object *obj = io->ci_obj;
+ struct inode *inode = vvp_object_inode(obj);
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct cl_fault_io *fio = &io->u.ci_fault;
+ struct vvp_fault_io *cfio = &vio->u.fault;
+ loff_t offset;
+ int result = 0;
+ struct page *vmpage = NULL;
+ struct cl_page *page;
+ loff_t size;
+ pgoff_t last_index;
ENTRY;
- if (fio->ft_executable &&
- LTIME_S(inode->i_mtime) != vio->u.fault.ft_mtime)
- CWARN("binary "DFID
- " changed while waiting for the page fault lock\n",
- PFID(lu_object_fid(&obj->co_lu)));
+ down_read(&lli->lli_trunc_sem);
/* offset of the last byte on the page */
offset = cl_offset(obj, fio->ft_index + 1) - 1;
return result;
}
+static void vvp_io_fault_end(const struct lu_env *env,
+ const struct cl_io_slice *ios)
+{
+ struct inode *inode = vvp_object_inode(ios->cis_obj);
+ struct ll_inode_info *lli = ll_i2info(inode);
+
+ CLOBINVRNT(env, ios->cis_io->ci_obj,
+ vvp_object_invariant(ios->cis_io->ci_obj));
+ up_read(&lli->lli_trunc_sem);
+}
+
static int vvp_io_fsync_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
RETURN(result);
}
-static void vvp_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
-{
- CLOBINVRNT(env, ios->cis_io->ci_obj,
- vvp_object_invariant(ios->cis_io->ci_obj));
-}
-
static const struct cl_io_operations vvp_io_ops = {
.op = {
[CIT_READ] = {
.cio_fini = vvp_io_fini,
.cio_lock = vvp_io_read_lock,
.cio_start = vvp_io_read_start,
+ .cio_end = vvp_io_rw_end,
.cio_advance = vvp_io_advance,
},
[CIT_WRITE] = {
.cio_iter_fini = vvp_io_write_iter_fini,
.cio_lock = vvp_io_write_lock,
.cio_start = vvp_io_write_start,
+ .cio_end = vvp_io_rw_end,
.cio_advance = vvp_io_advance,
},
[CIT_SETATTR] = {
.cio_iter_init = vvp_io_fault_iter_init,
.cio_lock = vvp_io_fault_lock,
.cio_start = vvp_io_fault_start,
- .cio_end = vvp_io_end,
+ .cio_end = vvp_io_fault_end,
},
[CIT_FSYNC] = {
- .cio_start = vvp_io_fsync_start,
- .cio_fini = vvp_io_fini
+ .cio_start = vvp_io_fsync_start,
+ .cio_fini = vvp_io_fini
+ },
+ [CIT_MISC] = {
+ .cio_fini = vvp_io_fini
+ },
+ [CIT_LADVISE] = {
+ .cio_fini = vvp_io_fini
},
- [CIT_MISC] = {
- .cio_fini = vvp_io_fini
- }
- },
+ },
.cio_read_ahead = vvp_io_read_ahead
};
vio->vui_ra_valid = false;
result = 0;
if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE) {
- size_t count;
struct ll_inode_info *lli = ll_i2info(inode);
- count = io->u.ci_rw.crw_count;
- /* "If nbyte is 0, read() will return 0 and have no other
- * results." -- Single Unix Spec */
- if (count == 0)
- result = 1;
- else {
- vio->vui_tot_count = count;
- vio->vui_tot_nrsegs = 0;
- }
+ vio->vui_tot_count = io->u.ci_rw.rw_range.cir_count;
+ /* "If nbyte is 0, read() will return 0 and have no other
+ * results." -- Single Unix Spec */
+ if (vio->vui_tot_count == 0)
+ result = 1;
/* for read/write, we store the jobid in the inode, and
* it'll be fetched by osc when building RPC.
io->ci_lockreq = CILR_MANDATORY;
}
- /* ignore layout change for generic CIT_MISC but not for glimpse.
- * io context for glimpse must set ci_verify_layout to true,
- * see cl_glimpse_size0() for details. */
- if (io->ci_type == CIT_MISC && !io->ci_verify_layout)
- io->ci_ignore_layout = 1;
-
/* Enqueue layout lock and get layout version. We need to do this
* even for operations requiring to open file, such as read and write,
* because it might not grant layout lock in IT_OPEN. */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff