1 // SPDX-License-Identifier: GPL-2.0
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License version 2 for more details (a copy is included
15 * in the LICENSE file that accompanied this code).
17 * You should have received a copy of the GNU General Public License
18 * version 2 along with this program; If not, see
19 * http://www.gnu.org/licenses/gpl-2.0.html
24 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Use is subject to license terms.
27 * Copyright (c) 2011, 2015, Intel Corporation.
30 * This file is part of Lustre, http://www.lustre.org/
31 * Lustre is a trademark of Sun Microsystems, Inc.
35 * Author: Peter Braam <braam@clusterfs.com>
36 * Author: Phil Schwan <phil@clusterfs.com>
37 * Author: Andreas Dilger <adilger@clusterfs.com>
40 #define DEBUG_SUBSYSTEM S_LLITE
41 #include <lustre_dlm.h>
42 #include <linux/pagemap.h>
43 #include <linux/file.h>
44 #include <linux/sched.h>
45 #include <linux/mount.h>
46 #include <uapi/linux/lustre/lustre_fiemap.h>
47 #include <uapi/linux/lustre/lustre_ioctl.h>
48 #include <lustre_swab.h>
50 #include <cl_object.h>
51 #include "llite_internal.h"
54 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
56 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
59 static enum llioc_iter
60 ll_iocontrol_call(struct inode *inode, struct file *file,
61 unsigned int cmd, unsigned long arg, int *rcp);
63 static struct ll_file_data *ll_file_data_get(void)
65 struct ll_file_data *fd;
67 fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS);
70 fd->fd_write_failed = false;
74 static void ll_file_data_put(struct ll_file_data *fd)
77 kmem_cache_free(ll_file_data_slab, fd);
81 * Packs all the attributes into @op_data for the CLOSE rpc.
83 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
84 struct obd_client_handle *och)
86 struct ll_inode_info *lli = ll_i2info(inode);
88 ll_prep_md_op_data(op_data, inode, NULL, NULL,
89 0, 0, LUSTRE_OPC_ANY, NULL);
91 op_data->op_attr.ia_mode = inode->i_mode;
92 op_data->op_attr.ia_atime = inode->i_atime;
93 op_data->op_attr.ia_mtime = inode->i_mtime;
94 op_data->op_attr.ia_ctime = inode->i_ctime;
95 op_data->op_attr.ia_size = i_size_read(inode);
96 op_data->op_attr.ia_valid |= ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
97 ATTR_MTIME | ATTR_MTIME_SET |
98 ATTR_CTIME | ATTR_CTIME_SET;
99 op_data->op_attr_blocks = inode->i_blocks;
100 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
101 op_data->op_handle = och->och_fh;
104 * For HSM: if inode data has been modified, pack it so that
105 * MDT can set data dirty flag in the archive.
107 if (och->och_flags & FMODE_WRITE &&
108 test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags))
109 op_data->op_bias |= MDS_DATA_MODIFIED;
113 * Perform a close, possibly with a bias.
114 * The meaning of "data" depends on the value of "bias".
116 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
117 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
120 static int ll_close_inode_openhandle(struct inode *inode,
121 struct obd_client_handle *och,
122 enum mds_op_bias bias,
125 const struct ll_inode_info *lli = ll_i2info(inode);
126 struct obd_export *md_exp = ll_i2mdexp(inode);
127 struct md_op_data *op_data;
128 struct ptlrpc_request *req = NULL;
131 if (!class_exp2obd(md_exp)) {
132 CERROR("%s: invalid MDC connection handle closing " DFID "\n",
133 ll_get_fsname(inode->i_sb, NULL, 0),
134 PFID(&lli->lli_fid));
139 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
141 * We leak openhandle and request here on error, but not much to be
142 * done in OOM case since app won't retry close on error either.
149 ll_prepare_close(inode, op_data, och);
151 case MDS_CLOSE_LAYOUT_SWAP:
153 op_data->op_bias |= MDS_CLOSE_LAYOUT_SWAP;
154 op_data->op_data_version = 0;
155 op_data->op_lease_handle = och->och_lease_handle;
156 op_data->op_fid2 = *ll_inode2fid(data);
159 case MDS_HSM_RELEASE:
161 op_data->op_bias |= MDS_HSM_RELEASE;
162 op_data->op_data_version = *(__u64 *)data;
163 op_data->op_lease_handle = och->och_lease_handle;
164 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
172 rc = md_close(md_exp, op_data, och->och_mod, &req);
173 if (rc && rc != -EINTR) {
174 CERROR("%s: inode " DFID " mdc close failed: rc = %d\n",
175 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
178 if (op_data->op_bias & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP) &&
180 struct mdt_body *body;
182 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
183 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
187 ll_finish_md_op_data(op_data);
190 md_clear_open_replay_data(md_exp, och);
191 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
194 ptlrpc_req_finished(req);
198 int ll_md_real_close(struct inode *inode, fmode_t fmode)
200 struct ll_inode_info *lli = ll_i2info(inode);
201 struct obd_client_handle **och_p;
202 struct obd_client_handle *och;
206 if (fmode & FMODE_WRITE) {
207 och_p = &lli->lli_mds_write_och;
208 och_usecount = &lli->lli_open_fd_write_count;
209 } else if (fmode & FMODE_EXEC) {
210 och_p = &lli->lli_mds_exec_och;
211 och_usecount = &lli->lli_open_fd_exec_count;
213 LASSERT(fmode & FMODE_READ);
214 och_p = &lli->lli_mds_read_och;
215 och_usecount = &lli->lli_open_fd_read_count;
218 mutex_lock(&lli->lli_och_mutex);
219 if (*och_usecount > 0) {
220 /* There are still users of this handle, so skip
223 mutex_unlock(&lli->lli_och_mutex);
229 mutex_unlock(&lli->lli_och_mutex);
232 /* There might be a race and this handle may already
235 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
241 static int ll_md_close(struct inode *inode, struct file *file)
243 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
244 struct ll_inode_info *lli = ll_i2info(inode);
246 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
247 struct lustre_handle lockh;
248 union ldlm_policy_data policy = {
249 .l_inodebits = { MDS_INODELOCK_OPEN }
253 /* clear group lock, if present */
254 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
255 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
257 if (fd->fd_lease_och) {
260 /* Usually the lease is not released when the
261 * application crashed, we need to release here.
263 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
264 CDEBUG(rc ? D_ERROR : D_INODE,
265 "Clean up lease " DFID " %d/%d\n",
266 PFID(&lli->lli_fid), rc, lease_broken);
268 fd->fd_lease_och = NULL;
272 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
277 /* Let's see if we have good enough OPEN lock on the file and if
278 * we can skip talking to MDS
281 mutex_lock(&lli->lli_och_mutex);
282 if (fd->fd_omode & FMODE_WRITE) {
284 LASSERT(lli->lli_open_fd_write_count);
285 lli->lli_open_fd_write_count--;
286 } else if (fd->fd_omode & FMODE_EXEC) {
288 LASSERT(lli->lli_open_fd_exec_count);
289 lli->lli_open_fd_exec_count--;
292 LASSERT(lli->lli_open_fd_read_count);
293 lli->lli_open_fd_read_count--;
295 mutex_unlock(&lli->lli_och_mutex);
297 if (!md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
298 LDLM_IBITS, &policy, lockmode, &lockh))
299 rc = ll_md_real_close(inode, fd->fd_omode);
302 LUSTRE_FPRIVATE(file) = NULL;
303 ll_file_data_put(fd);
308 /* While this returns an error code, fput() the caller does not, so we need
309 * to make every effort to clean up all of our state here. Also, applications
310 * rarely check close errors and even if an error is returned they will not
311 * re-try the close call.
313 int ll_file_release(struct inode *inode, struct file *file)
315 struct ll_file_data *fd;
316 struct ll_sb_info *sbi = ll_i2sbi(inode);
317 struct ll_inode_info *lli = ll_i2info(inode);
320 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p)\n",
321 PFID(ll_inode2fid(inode)), inode);
323 if (!is_root_inode(inode))
324 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
325 fd = LUSTRE_FPRIVATE(file);
328 /* The last ref on @file, maybe not be the owner pid of statahead,
329 * because parent and child process can share the same file handle.
331 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
332 ll_deauthorize_statahead(inode, fd);
334 if (is_root_inode(inode)) {
335 LUSTRE_FPRIVATE(file) = NULL;
336 ll_file_data_put(fd);
340 if (!S_ISDIR(inode->i_mode)) {
342 lov_read_and_clear_async_rc(lli->lli_clob);
343 lli->lli_async_rc = 0;
346 rc = ll_md_close(inode, file);
348 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
349 libcfs_debug_dumplog();
354 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
355 struct lookup_intent *itp)
357 struct inode *inode = d_inode(de);
358 struct ll_sb_info *sbi = ll_i2sbi(inode);
359 struct dentry *parent = de->d_parent;
360 const char *name = NULL;
361 struct md_op_data *op_data;
362 struct ptlrpc_request *req = NULL;
366 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
369 * if server supports open-by-fid, or file name is invalid, don't pack
370 * name in open request
372 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
373 lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
374 name = de->d_name.name;
375 len = de->d_name.len;
378 op_data = ll_prep_md_op_data(NULL, d_inode(parent), inode, name, len,
379 O_RDWR, LUSTRE_OPC_ANY, NULL);
381 return PTR_ERR(op_data);
382 op_data->op_data = lmm;
383 op_data->op_data_size = lmmsize;
385 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
386 &ll_md_blocking_ast, 0);
387 ll_finish_md_op_data(op_data);
389 /* reason for keep own exit path - don`t flood log
390 * with messages with -ESTALE errors.
392 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
393 it_open_error(DISP_OPEN_OPEN, itp))
395 ll_release_openhandle(inode, itp);
399 if (it_disposition(itp, DISP_LOOKUP_NEG)) {
404 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
405 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
406 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
410 rc = ll_prep_inode(&inode, req, NULL, itp);
411 if (!rc && itp->it_lock_mode)
412 ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL);
415 ptlrpc_req_finished(req);
416 ll_intent_drop_lock(itp);
419 * We did open by fid, but by the time we got to the server,
420 * the object disappeared. If this is a create, we cannot really
421 * tell the userspace that the file it was trying to create
422 * does not exist. Instead let's return -ESTALE, and the VFS will
423 * retry the create with LOOKUP_REVAL that we are going to catch
424 * in ll_revalidate_dentry() and use lookup then.
426 if (rc == -ENOENT && itp->it_op & IT_CREAT)
432 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
433 struct obd_client_handle *och)
435 struct mdt_body *body;
437 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
438 och->och_fh = body->mbo_handle;
439 och->och_fid = body->mbo_fid1;
440 och->och_lease_handle.cookie = it->it_lock_handle;
441 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
442 och->och_flags = it->it_flags;
444 return md_set_open_replay_data(md_exp, och, it);
447 static int ll_local_open(struct file *file, struct lookup_intent *it,
448 struct ll_file_data *fd, struct obd_client_handle *och)
450 struct inode *inode = file_inode(file);
452 LASSERT(!LUSTRE_FPRIVATE(file));
459 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
464 LUSTRE_FPRIVATE(file) = fd;
465 ll_readahead_init(inode, &fd->fd_ras);
466 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
468 /* ll_cl_context initialize */
469 rwlock_init(&fd->fd_lock);
470 INIT_LIST_HEAD(&fd->fd_lccs);
475 /* Open a file, and (for the very first open) create objects on the OSTs at
476 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
477 * creation or open until ll_lov_setstripe() ioctl is called.
479 * If we already have the stripe MD locally then we don't request it in
480 * md_open(), by passing a lmm_size = 0.
482 * It is up to the application to ensure no other processes open this file
483 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
484 * used. We might be able to avoid races of that sort by getting lli_open_sem
485 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
486 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
488 int ll_file_open(struct inode *inode, struct file *file)
490 struct ll_inode_info *lli = ll_i2info(inode);
491 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
492 .it_flags = file->f_flags };
493 struct obd_client_handle **och_p = NULL;
494 __u64 *och_usecount = NULL;
495 struct ll_file_data *fd;
498 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p), flags %o\n",
499 PFID(ll_inode2fid(inode)), inode, file->f_flags);
501 it = file->private_data; /* XXX: compat macro */
502 file->private_data = NULL; /* prevent ll_local_open assertion */
504 fd = ll_file_data_get();
511 if (S_ISDIR(inode->i_mode))
512 ll_authorize_statahead(inode, fd);
514 if (is_root_inode(inode)) {
515 LUSTRE_FPRIVATE(file) = fd;
519 if (!it || !it->it_disposition) {
520 /* Convert f_flags into access mode. We cannot use file->f_mode,
521 * because everything but O_ACCMODE mask was stripped from
524 if ((oit.it_flags + 1) & O_ACCMODE)
526 if (file->f_flags & O_TRUNC)
527 oit.it_flags |= FMODE_WRITE;
529 /* kernel only call f_op->open in dentry_open. filp_open calls
530 * dentry_open after call to open_namei that checks permissions.
531 * Only nfsd_open call dentry_open directly without checking
532 * permissions and because of that this code below is safe.
534 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
535 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
537 /* We do not want O_EXCL here, presumably we opened the file
538 * already? XXX - NFS implications?
540 oit.it_flags &= ~O_EXCL;
542 /* bug20584, if "it_flags" contains O_CREAT, the file will be
543 * created if necessary, then "IT_CREAT" should be set to keep
546 if (oit.it_flags & O_CREAT)
547 oit.it_op |= IT_CREAT;
553 /* Let's see if we have file open on MDS already. */
554 if (it->it_flags & FMODE_WRITE) {
555 och_p = &lli->lli_mds_write_och;
556 och_usecount = &lli->lli_open_fd_write_count;
557 } else if (it->it_flags & FMODE_EXEC) {
558 och_p = &lli->lli_mds_exec_och;
559 och_usecount = &lli->lli_open_fd_exec_count;
561 och_p = &lli->lli_mds_read_och;
562 och_usecount = &lli->lli_open_fd_read_count;
565 mutex_lock(&lli->lli_och_mutex);
566 if (*och_p) { /* Open handle is present */
567 if (it_disposition(it, DISP_OPEN_OPEN)) {
568 /* Well, there's extra open request that we do not need,
569 * let's close it somehow. This will decref request.
571 rc = it_open_error(DISP_OPEN_OPEN, it);
573 mutex_unlock(&lli->lli_och_mutex);
577 ll_release_openhandle(inode, it);
581 rc = ll_local_open(file, it, fd, NULL);
584 mutex_unlock(&lli->lli_och_mutex);
588 LASSERT(*och_usecount == 0);
589 if (!it->it_disposition) {
590 /* We cannot just request lock handle now, new ELC code
591 * means that one of other OPEN locks for this file
592 * could be cancelled, and since blocking ast handler
593 * would attempt to grab och_mutex as well, that would
594 * result in a deadlock
596 mutex_unlock(&lli->lli_och_mutex);
598 * Normally called under two situations:
600 * 2. revalidate with IT_OPEN (revalidate doesn't
601 * execute this intent any more).
603 * Always fetch MDS_OPEN_LOCK if this is not setstripe.
605 * Always specify MDS_OPEN_BY_FID because we don't want
606 * to get file with different fid.
608 it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
609 rc = ll_intent_file_open(file->f_path.dentry,
616 *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS);
624 /* md_intent_lock() didn't get a request ref if there was an
625 * open error, so don't do cleanup on the request here
628 /* XXX (green): Should not we bail out on any error here, not
631 rc = it_open_error(DISP_OPEN_OPEN, it);
635 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
636 "inode %p: disposition %x, status %d\n", inode,
637 it_disposition(it, ~0), it->it_status);
639 rc = ll_local_open(file, it, fd, *och_p);
643 mutex_unlock(&lli->lli_och_mutex);
646 /* Must do this outside lli_och_mutex lock to prevent deadlock where
647 * different kind of OPEN lock for this same inode gets cancelled
650 if (!S_ISREG(inode->i_mode))
653 cl_lov_delay_create_clear(&file->f_flags);
658 if (och_p && *och_p) {
663 mutex_unlock(&lli->lli_och_mutex);
666 if (lli->lli_opendir_key == fd)
667 ll_deauthorize_statahead(inode, fd);
669 ll_file_data_put(fd);
671 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
674 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
675 ptlrpc_req_finished(it->it_request);
676 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
682 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
683 struct ldlm_lock_desc *desc,
684 void *data, int flag)
687 struct lustre_handle lockh;
690 case LDLM_CB_BLOCKING:
691 ldlm_lock2handle(lock, &lockh);
692 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
694 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
698 case LDLM_CB_CANCELING:
706 * Acquire a lease and open the file.
708 static struct obd_client_handle *
709 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
712 struct lookup_intent it = { .it_op = IT_OPEN };
713 struct ll_sb_info *sbi = ll_i2sbi(inode);
714 struct md_op_data *op_data;
715 struct ptlrpc_request *req = NULL;
716 struct lustre_handle old_handle = { 0 };
717 struct obd_client_handle *och = NULL;
721 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
722 return ERR_PTR(-EINVAL);
725 struct ll_inode_info *lli = ll_i2info(inode);
726 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
727 struct obd_client_handle **och_p;
730 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
731 return ERR_PTR(-EPERM);
733 /* Get the openhandle of the file */
735 mutex_lock(&lli->lli_och_mutex);
736 if (fd->fd_lease_och) {
737 mutex_unlock(&lli->lli_och_mutex);
742 if (file->f_mode & FMODE_WRITE) {
743 LASSERT(lli->lli_mds_write_och);
744 och_p = &lli->lli_mds_write_och;
745 och_usecount = &lli->lli_open_fd_write_count;
747 LASSERT(lli->lli_mds_read_och);
748 och_p = &lli->lli_mds_read_och;
749 och_usecount = &lli->lli_open_fd_read_count;
751 if (*och_usecount == 1) {
758 mutex_unlock(&lli->lli_och_mutex);
759 if (rc < 0) /* more than 1 opener */
763 old_handle = fd->fd_och->och_fh;
766 och = kzalloc(sizeof(*och), GFP_NOFS);
768 return ERR_PTR(-ENOMEM);
770 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
771 LUSTRE_OPC_ANY, NULL);
772 if (IS_ERR(op_data)) {
773 rc = PTR_ERR(op_data);
777 /* To tell the MDT this openhandle is from the same owner */
778 op_data->op_handle = old_handle;
780 it.it_flags = fmode | open_flags;
781 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
782 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
783 &ll_md_blocking_lease_ast,
784 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
785 * it can be cancelled which may mislead applications that the lease is
787 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
788 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
789 * doesn't deal with openhandle, so normal openhandle will be leaked.
791 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
792 ll_finish_md_op_data(op_data);
793 ptlrpc_req_finished(req);
797 if (it_disposition(&it, DISP_LOOKUP_NEG)) {
802 rc = it_open_error(DISP_OPEN_OPEN, &it);
806 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
807 ll_och_fill(sbi->ll_md_exp, &it, och);
809 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ {
814 /* already get lease, handle lease lock */
815 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
816 if (it.it_lock_mode == 0 ||
817 it.it_lock_bits != MDS_INODELOCK_OPEN) {
818 /* open lock must return for lease */
819 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
820 PFID(ll_inode2fid(inode)), it.it_lock_mode,
826 ll_intent_release(&it);
830 /* Cancel open lock */
831 if (it.it_lock_mode != 0) {
832 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
835 och->och_lease_handle.cookie = 0ULL;
837 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
839 CERROR("%s: error closing file " DFID ": %d\n",
840 ll_get_fsname(inode->i_sb, NULL, 0),
841 PFID(&ll_i2info(inode)->lli_fid), rc2);
842 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
844 ll_intent_release(&it);
851 * Check whether a layout swap can be done between two inodes.
853 * \param[in] inode1 First inode to check
854 * \param[in] inode2 Second inode to check
856 * \retval 0 on success, layout swap can be performed between both inodes
857 * \retval negative error code if requirements are not met
859 static int ll_check_swap_layouts_validity(struct inode *inode1,
860 struct inode *inode2)
862 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
865 if (inode_permission(inode1, MAY_WRITE) ||
866 inode_permission(inode2, MAY_WRITE))
869 if (inode1->i_sb != inode2->i_sb)
875 static int ll_swap_layouts_close(struct obd_client_handle *och,
876 struct inode *inode, struct inode *inode2)
878 const struct lu_fid *fid1 = ll_inode2fid(inode);
879 const struct lu_fid *fid2;
882 CDEBUG(D_INODE, "%s: biased close of file " DFID "\n",
883 ll_get_fsname(inode->i_sb, NULL, 0), PFID(fid1));
885 rc = ll_check_swap_layouts_validity(inode, inode2);
889 /* We now know that inode2 is a lustre inode */
890 fid2 = ll_inode2fid(inode2);
892 rc = lu_fid_cmp(fid1, fid2);
899 * Close the file and swap layouts between inode & inode2.
900 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
901 * because we still need it to pack l_remote_handle to MDT.
903 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
906 och = NULL; /* freed in ll_close_inode_openhandle() */
914 * Release lease and close the file.
915 * It will check if the lease has ever broken.
917 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
920 struct ldlm_lock *lock;
921 bool cancelled = true;
923 lock = ldlm_handle2lock(&och->och_lease_handle);
925 lock_res_and_lock(lock);
926 cancelled = ldlm_is_cancel(lock);
927 unlock_res_and_lock(lock);
931 CDEBUG(D_INODE, "lease for " DFID " broken? %d\n",
932 PFID(&ll_i2info(inode)->lli_fid), cancelled);
935 ldlm_cli_cancel(&och->och_lease_handle, 0);
937 *lease_broken = cancelled;
939 return ll_close_inode_openhandle(inode, och, 0, NULL);
942 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
944 struct ll_inode_info *lli = ll_i2info(inode);
945 struct cl_object *obj = lli->lli_clob;
946 struct cl_attr *attr = vvp_env_thread_attr(env);
952 ll_inode_size_lock(inode);
954 /* merge timestamps the most recently obtained from mds with
955 * timestamps obtained from osts
957 LTIME_S(inode->i_atime) = lli->lli_atime;
958 LTIME_S(inode->i_mtime) = lli->lli_mtime;
959 LTIME_S(inode->i_ctime) = lli->lli_ctime;
961 mtime = LTIME_S(inode->i_mtime);
962 atime = LTIME_S(inode->i_atime);
963 ctime = LTIME_S(inode->i_ctime);
965 cl_object_attr_lock(obj);
966 rc = cl_object_attr_get(env, obj, attr);
967 cl_object_attr_unlock(obj);
970 goto out_size_unlock;
972 if (atime < attr->cat_atime)
973 atime = attr->cat_atime;
975 if (ctime < attr->cat_ctime)
976 ctime = attr->cat_ctime;
978 if (mtime < attr->cat_mtime)
979 mtime = attr->cat_mtime;
981 CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n",
982 PFID(&lli->lli_fid), attr->cat_size);
984 i_size_write(inode, attr->cat_size);
986 inode->i_blocks = attr->cat_blocks;
988 LTIME_S(inode->i_mtime) = mtime;
989 LTIME_S(inode->i_atime) = atime;
990 LTIME_S(inode->i_ctime) = ctime;
993 ll_inode_size_unlock(inode);
998 static bool file_is_noatime(const struct file *file)
1000 const struct vfsmount *mnt = file->f_path.mnt;
1001 const struct inode *inode = file_inode(file);
1003 /* Adapted from file_accessed() and touch_atime().*/
1004 if (file->f_flags & O_NOATIME)
1007 if (inode->i_flags & S_NOATIME)
1010 if (IS_NOATIME(inode))
1013 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1016 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1019 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1025 static void ll_io_init(struct cl_io *io, const struct file *file, int write)
1027 struct inode *inode = file_inode(file);
1029 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1031 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1032 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1033 file->f_flags & O_DIRECT ||
1036 io->ci_obj = ll_i2info(inode)->lli_clob;
1037 io->ci_lockreq = CILR_MAYBE;
1038 if (ll_file_nolock(file)) {
1039 io->ci_lockreq = CILR_NEVER;
1040 io->ci_no_srvlock = 1;
1041 } else if (file->f_flags & O_APPEND) {
1042 io->ci_lockreq = CILR_MANDATORY;
1045 io->ci_noatime = file_is_noatime(file);
1049 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1050 struct file *file, enum cl_io_type iot,
1051 loff_t *ppos, size_t count)
1053 struct ll_inode_info *lli = ll_i2info(file_inode(file));
1054 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1055 struct vvp_io *vio = vvp_env_io(env);
1056 struct range_lock range;
1061 CDEBUG(D_VFSTRACE, "file: %pD, type: %d ppos: %llu, count: %zu\n",
1062 file, iot, *ppos, count);
1065 io = vvp_env_thread_io(env);
1066 ll_io_init(io, file, iot == CIT_WRITE);
1068 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1069 struct vvp_io *vio = vvp_env_io(env);
1070 bool range_locked = false;
1072 if (file->f_flags & O_APPEND)
1073 range_lock_init(&range, 0, LUSTRE_EOF);
1075 range_lock_init(&range, *ppos, *ppos + count - 1);
1077 vio->vui_fd = LUSTRE_FPRIVATE(file);
1078 vio->vui_iter = args->u.normal.via_iter;
1079 vio->vui_iocb = args->u.normal.via_iocb;
1081 * Direct IO reads must also take range lock,
1082 * or multiple reads will try to work on the same pages
1083 * See LU-6227 for details.
1085 if (((iot == CIT_WRITE) ||
1086 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1087 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1088 CDEBUG(D_VFSTRACE, "Range lock [%llu, %llu]\n",
1089 range.rl_node.in_extent.start,
1090 range.rl_node.in_extent.end);
1091 rc = range_lock(&lli->lli_write_tree, &range);
1095 range_locked = true;
1097 ll_cl_add(file, env, io);
1098 rc = cl_io_loop(env, io);
1099 ll_cl_remove(file, env);
1101 CDEBUG(D_VFSTRACE, "Range unlock [%llu, %llu]\n",
1102 range.rl_node.in_extent.start,
1103 range.rl_node.in_extent.end);
1104 range_unlock(&lli->lli_write_tree, &range);
1107 /* cl_io_rw_init() handled IO */
1111 if (io->ci_nob > 0) {
1112 result = io->ci_nob;
1113 count -= io->ci_nob;
1114 *ppos = io->u.ci_wr.wr.crw_pos;
1116 /* prepare IO restart */
1118 args->u.normal.via_iter = vio->vui_iter;
1121 cl_io_fini(env, io);
1123 if ((!rc || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
1125 "%s: restart %s from %lld, count:%zu, result: %zd\n",
1126 file_dentry(file)->d_name.name,
1127 iot == CIT_READ ? "read" : "write",
1128 *ppos, count, result);
1132 if (iot == CIT_READ) {
1134 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1135 LPROC_LL_READ_BYTES, result);
1136 } else if (iot == CIT_WRITE) {
1138 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1139 LPROC_LL_WRITE_BYTES, result);
1140 fd->fd_write_failed = false;
1141 } else if (!result && !rc) {
1144 fd->fd_write_failed = true;
1146 fd->fd_write_failed = false;
1147 } else if (rc != -ERESTARTSYS) {
1148 fd->fd_write_failed = true;
1151 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1153 return result > 0 ? result : rc;
1156 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1159 struct vvp_io_args *args;
1163 env = cl_env_get(&refcheck);
1165 return PTR_ERR(env);
1167 args = ll_env_args(env);
1168 args->u.normal.via_iter = to;
1169 args->u.normal.via_iocb = iocb;
1171 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1172 &iocb->ki_pos, iov_iter_count(to));
1173 cl_env_put(env, &refcheck);
1178 * Write to a file (through the page cache).
1180 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1183 struct vvp_io_args *args;
1187 env = cl_env_get(&refcheck);
1189 return PTR_ERR(env);
1191 args = ll_env_args(env);
1192 args->u.normal.via_iter = from;
1193 args->u.normal.via_iocb = iocb;
1195 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1196 &iocb->ki_pos, iov_iter_count(from));
1197 cl_env_put(env, &refcheck);
1201 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1202 __u64 flags, struct lov_user_md *lum,
1205 struct lookup_intent oit = {
1207 .it_flags = flags | MDS_OPEN_BY_FID,
1211 ll_inode_size_lock(inode);
1212 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
1216 ll_release_openhandle(inode, &oit);
1219 ll_inode_size_unlock(inode);
1220 ll_intent_release(&oit);
1224 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1225 struct lov_mds_md **lmmp, int *lmm_size,
1226 struct ptlrpc_request **request)
1228 struct ll_sb_info *sbi = ll_i2sbi(inode);
1229 struct mdt_body *body;
1230 struct lov_mds_md *lmm = NULL;
1231 struct ptlrpc_request *req = NULL;
1232 struct md_op_data *op_data;
1235 rc = ll_get_default_mdsize(sbi, &lmmsize);
1239 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1240 strlen(filename), lmmsize,
1241 LUSTRE_OPC_ANY, NULL);
1242 if (IS_ERR(op_data))
1243 return PTR_ERR(op_data);
1245 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1246 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1247 ll_finish_md_op_data(op_data);
1249 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
1254 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1256 lmmsize = body->mbo_eadatasize;
1258 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1264 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1266 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1267 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1273 * This is coming from the MDS, so is probably in
1274 * little endian. We convert it to host endian before
1275 * passing it to userspace.
1277 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
1280 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1281 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1284 /* if function called for directory - we should
1285 * avoid swab not existent lsm objects
1287 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1288 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1289 if (S_ISREG(body->mbo_mode))
1290 lustre_swab_lov_user_md_objects(
1291 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1293 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1294 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1295 if (S_ISREG(body->mbo_mode))
1296 lustre_swab_lov_user_md_objects(
1297 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1304 *lmm_size = lmmsize;
1309 static int ll_lov_setea(struct inode *inode, struct file *file,
1312 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1313 struct lov_user_md *lump;
1314 int lum_size = sizeof(struct lov_user_md) +
1315 sizeof(struct lov_user_ost_data);
1318 if (!capable(CFS_CAP_SYS_ADMIN))
1321 lump = libcfs_kvzalloc(lum_size, GFP_NOFS);
1325 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) {
1330 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump,
1332 cl_lov_delay_create_clear(&file->f_flags);
1338 static int ll_file_getstripe(struct inode *inode,
1339 struct lov_user_md __user *lum)
1345 env = cl_env_get(&refcheck);
1347 return PTR_ERR(env);
1349 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum);
1350 cl_env_put(env, &refcheck);
1354 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1357 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
1358 struct lov_user_md *klum;
1360 __u64 flags = FMODE_WRITE;
1362 rc = ll_copy_user_md(lum, &klum);
1367 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, klum,
1369 cl_lov_delay_create_clear(&file->f_flags);
1373 put_user(0, &lum->lmm_stripe_count);
1375 ll_layout_refresh(inode, &gen);
1376 rc = ll_file_getstripe(inode, (struct lov_user_md __user *)arg);
1384 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1386 struct ll_inode_info *lli = ll_i2info(inode);
1387 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1388 struct ll_grouplock grouplock;
1392 CWARN("group id for group lock must not be 0\n");
1396 if (ll_file_nolock(file))
1399 spin_lock(&lli->lli_lock);
1400 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1401 CWARN("group lock already existed with gid %lu\n",
1402 fd->fd_grouplock.lg_gid);
1403 spin_unlock(&lli->lli_lock);
1406 LASSERT(!fd->fd_grouplock.lg_lock);
1407 spin_unlock(&lli->lli_lock);
1409 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1410 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1414 spin_lock(&lli->lli_lock);
1415 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1416 spin_unlock(&lli->lli_lock);
1417 CERROR("another thread just won the race\n");
1418 cl_put_grouplock(&grouplock);
1422 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1423 fd->fd_grouplock = grouplock;
1424 spin_unlock(&lli->lli_lock);
1426 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1430 static int ll_put_grouplock(struct inode *inode, struct file *file,
1433 struct ll_inode_info *lli = ll_i2info(inode);
1434 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1435 struct ll_grouplock grouplock;
1437 spin_lock(&lli->lli_lock);
1438 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1439 spin_unlock(&lli->lli_lock);
1440 CWARN("no group lock held\n");
1443 LASSERT(fd->fd_grouplock.lg_lock);
1445 if (fd->fd_grouplock.lg_gid != arg) {
1446 CWARN("group lock %lu doesn't match current id %lu\n",
1447 arg, fd->fd_grouplock.lg_gid);
1448 spin_unlock(&lli->lli_lock);
1452 grouplock = fd->fd_grouplock;
1453 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1454 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1455 spin_unlock(&lli->lli_lock);
1457 cl_put_grouplock(&grouplock);
1458 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1463 * Close inode open handle
1465 * \param inode [in] inode in question
1466 * \param it [in,out] intent which contains open info and result
1469 * \retval <0 failure
1471 int ll_release_openhandle(struct inode *inode, struct lookup_intent *it)
1473 struct obd_client_handle *och;
1478 /* Root ? Do nothing. */
1479 if (is_root_inode(inode))
1482 /* No open handle to close? Move away */
1483 if (!it_disposition(it, DISP_OPEN_OPEN))
1486 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1488 och = kzalloc(sizeof(*och), GFP_NOFS);
1494 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1496 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
1498 /* this one is in place of ll_file_open */
1499 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1500 ptlrpc_req_finished(it->it_request);
1501 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1507 * Get size for inode for which FIEMAP mapping is requested.
1508 * Make the FIEMAP get_info call and returns the result.
1510 * \param fiemap kernel buffer to hold extens
1511 * \param num_bytes kernel buffer size
1513 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
1516 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
1521 /* Checks for fiemap flags */
1522 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1523 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1527 /* Check for FIEMAP_FLAG_SYNC */
1528 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1529 rc = filemap_fdatawrite(inode->i_mapping);
1534 env = cl_env_get(&refcheck);
1536 return PTR_ERR(env);
1538 if (i_size_read(inode) == 0) {
1539 rc = ll_glimpse_size(inode);
1544 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1545 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
1546 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
1548 /* If filesize is 0, then there would be no objects for mapping */
1549 if (fmkey.lfik_oa.o_size == 0) {
1550 fiemap->fm_mapped_extents = 0;
1555 memcpy(&fmkey.lfik_fiemap, fiemap, sizeof(*fiemap));
1557 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
1558 &fmkey, fiemap, &num_bytes);
1560 cl_env_put(env, &refcheck);
1564 int ll_fid2path(struct inode *inode, void __user *arg)
1566 struct obd_export *exp = ll_i2mdexp(inode);
1567 const struct getinfo_fid2path __user *gfin = arg;
1568 struct getinfo_fid2path *gfout;
1573 if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
1574 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1577 /* Only need to get the buflen */
1578 if (get_user(pathlen, &gfin->gf_pathlen))
1581 if (pathlen > PATH_MAX)
1584 outsize = sizeof(*gfout) + pathlen;
1586 gfout = kzalloc(outsize, GFP_NOFS);
1590 if (copy_from_user(gfout, arg, sizeof(*gfout))) {
1595 /* Call mdc_iocontrol */
1596 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1600 if (copy_to_user(arg, gfout, outsize))
1609 * Read the data_version for inode.
1611 * This value is computed using stripe object version on OST.
1612 * Version is computed using server side locking.
1614 * @param flags if do sync on the OST side;
1616 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1617 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1619 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1621 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1627 /* If no file object initialized, we consider its version is 0. */
1633 env = cl_env_get(&refcheck);
1635 return PTR_ERR(env);
1637 io = vvp_env_thread_io(env);
1639 io->u.ci_data_version.dv_data_version = 0;
1640 io->u.ci_data_version.dv_flags = flags;
1643 if (!cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj))
1644 result = cl_io_loop(env, io);
1646 result = io->ci_result;
1648 *data_version = io->u.ci_data_version.dv_data_version;
1650 cl_io_fini(env, io);
1652 if (unlikely(io->ci_need_restart))
1655 cl_env_put(env, &refcheck);
1661 * Trigger a HSM release request for the provided inode.
1663 int ll_hsm_release(struct inode *inode)
1666 struct obd_client_handle *och = NULL;
1667 __u64 data_version = 0;
1671 CDEBUG(D_INODE, "%s: Releasing file " DFID ".\n",
1672 ll_get_fsname(inode->i_sb, NULL, 0),
1673 PFID(&ll_i2info(inode)->lli_fid));
1675 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
1681 /* Grab latest data_version and [am]time values */
1682 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
1686 env = cl_env_get(&refcheck);
1692 ll_merge_attr(env, inode);
1693 cl_env_put(env, &refcheck);
1695 /* Release the file.
1696 * NB: lease lock handle is released in mdc_hsm_release_pack() because
1697 * we still need it to pack l_remote_handle to MDT.
1699 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
1704 if (och && !IS_ERR(och)) /* close the file */
1705 ll_lease_close(och, inode, NULL);
1710 struct ll_swap_stack {
1713 struct inode *inode1;
1714 struct inode *inode2;
1719 static int ll_swap_layouts(struct file *file1, struct file *file2,
1720 struct lustre_swap_layouts *lsl)
1722 struct mdc_swap_layouts msl;
1723 struct md_op_data *op_data;
1726 struct ll_swap_stack *llss = NULL;
1729 llss = kzalloc(sizeof(*llss), GFP_NOFS);
1733 llss->inode1 = file_inode(file1);
1734 llss->inode2 = file_inode(file2);
1736 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
1740 /* we use 2 bool because it is easier to swap than 2 bits */
1741 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
1742 llss->check_dv1 = true;
1744 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
1745 llss->check_dv2 = true;
1747 /* we cannot use lsl->sl_dvX directly because we may swap them */
1748 llss->dv1 = lsl->sl_dv1;
1749 llss->dv2 = lsl->sl_dv2;
1751 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
1752 if (!rc) /* same file, done! */
1755 if (rc < 0) { /* sequentialize it */
1756 swap(llss->inode1, llss->inode2);
1758 swap(llss->dv1, llss->dv2);
1759 swap(llss->check_dv1, llss->check_dv2);
1763 if (gid != 0) { /* application asks to flush dirty cache */
1764 rc = ll_get_grouplock(llss->inode1, file1, gid);
1768 rc = ll_get_grouplock(llss->inode2, file2, gid);
1770 ll_put_grouplock(llss->inode1, file1, gid);
1775 /* ultimate check, before swapping the layouts we check if
1776 * dataversion has changed (if requested)
1778 if (llss->check_dv1) {
1779 rc = ll_data_version(llss->inode1, &dv, 0);
1782 if (dv != llss->dv1) {
1788 if (llss->check_dv2) {
1789 rc = ll_data_version(llss->inode2, &dv, 0);
1792 if (dv != llss->dv2) {
1798 /* struct md_op_data is used to send the swap args to the mdt
1799 * only flags is missing, so we use struct mdc_swap_layouts
1800 * through the md_op_data->op_data
1802 /* flags from user space have to be converted before they are send to
1803 * server, no flag is sent today, they are only used on the client
1807 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
1808 0, LUSTRE_OPC_ANY, &msl);
1809 if (IS_ERR(op_data)) {
1810 rc = PTR_ERR(op_data);
1814 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
1815 sizeof(*op_data), op_data, NULL);
1816 ll_finish_md_op_data(op_data);
1820 ll_put_grouplock(llss->inode2, file2, gid);
1821 ll_put_grouplock(llss->inode1, file1, gid);
1830 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
1832 struct md_op_data *op_data;
1835 /* Detect out-of range masks */
1836 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
1839 /* Non-root users are forbidden to set or clear flags which are
1840 * NOT defined in HSM_USER_MASK.
1842 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
1843 !capable(CFS_CAP_SYS_ADMIN))
1846 /* Detect out-of range archive id */
1847 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
1848 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
1851 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1852 LUSTRE_OPC_ANY, hss);
1853 if (IS_ERR(op_data))
1854 return PTR_ERR(op_data);
1856 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
1857 sizeof(*op_data), op_data, NULL);
1859 ll_finish_md_op_data(op_data);
1864 static int ll_hsm_import(struct inode *inode, struct file *file,
1865 struct hsm_user_import *hui)
1867 struct hsm_state_set *hss = NULL;
1868 struct iattr *attr = NULL;
1871 if (!S_ISREG(inode->i_mode))
1875 hss = kzalloc(sizeof(*hss), GFP_NOFS);
1879 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
1880 hss->hss_archive_id = hui->hui_archive_id;
1881 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
1882 rc = ll_hsm_state_set(inode, hss);
1886 attr = kzalloc(sizeof(*attr), GFP_NOFS);
1892 attr->ia_mode = hui->hui_mode & 0777;
1893 attr->ia_mode |= S_IFREG;
1894 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
1895 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
1896 attr->ia_size = hui->hui_size;
1897 attr->ia_mtime.tv_sec = hui->hui_mtime;
1898 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
1899 attr->ia_atime.tv_sec = hui->hui_atime;
1900 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
1902 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
1903 ATTR_UID | ATTR_GID |
1904 ATTR_MTIME | ATTR_MTIME_SET |
1905 ATTR_ATIME | ATTR_ATIME_SET;
1909 rc = ll_setattr_raw(file->f_path.dentry, attr, true);
1913 inode_unlock(inode);
1921 static inline long ll_lease_type_from_fmode(fmode_t fmode)
1923 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
1924 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
1928 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1930 struct inode *inode = file_inode(file);
1931 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1934 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p),cmd=%x\n",
1935 PFID(ll_inode2fid(inode)), inode, cmd);
1936 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1938 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1939 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1943 case LL_IOC_GETFLAGS:
1944 /* Get the current value of the file flags */
1945 return put_user(fd->fd_flags, (int __user *)arg);
1946 case LL_IOC_SETFLAGS:
1947 case LL_IOC_CLRFLAGS:
1948 /* Set or clear specific file flags */
1949 /* XXX This probably needs checks to ensure the flags are
1950 * not abused, and to handle any flag side effects.
1952 if (get_user(flags, (int __user *)arg))
1955 if (cmd == LL_IOC_SETFLAGS) {
1956 if ((flags & LL_FILE_IGNORE_LOCK) &&
1957 !(file->f_flags & O_DIRECT)) {
1958 CERROR("%s: unable to disable locking on non-O_DIRECT file\n",
1963 fd->fd_flags |= flags;
1965 fd->fd_flags &= ~flags;
1968 case LL_IOC_LOV_SETSTRIPE:
1969 return ll_lov_setstripe(inode, file, arg);
1970 case LL_IOC_LOV_SETEA:
1971 return ll_lov_setea(inode, file, arg);
1972 case LL_IOC_LOV_SWAP_LAYOUTS: {
1974 struct lustre_swap_layouts lsl;
1976 if (copy_from_user(&lsl, (char __user *)arg,
1977 sizeof(struct lustre_swap_layouts)))
1980 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
1983 file2 = fget(lsl.sl_fd);
1987 /* O_WRONLY or O_RDWR */
1988 if ((file2->f_flags & O_ACCMODE) == O_RDONLY) {
1993 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
1994 struct obd_client_handle *och = NULL;
1995 struct ll_inode_info *lli;
1996 struct inode *inode2;
1998 if (lsl.sl_flags != SWAP_LAYOUTS_CLOSE) {
2003 lli = ll_i2info(inode);
2004 mutex_lock(&lli->lli_och_mutex);
2005 if (fd->fd_lease_och) {
2006 och = fd->fd_lease_och;
2007 fd->fd_lease_och = NULL;
2009 mutex_unlock(&lli->lli_och_mutex);
2014 inode2 = file_inode(file2);
2015 rc = ll_swap_layouts_close(och, inode, inode2);
2017 rc = ll_swap_layouts(file, file2, &lsl);
2023 case LL_IOC_LOV_GETSTRIPE:
2024 return ll_file_getstripe(inode,
2025 (struct lov_user_md __user *)arg);
2026 case FSFILT_IOC_GETFLAGS:
2027 case FSFILT_IOC_SETFLAGS:
2028 return ll_iocontrol(inode, file, cmd, arg);
2029 case FSFILT_IOC_GETVERSION_OLD:
2030 case FSFILT_IOC_GETVERSION:
2031 return put_user(inode->i_generation, (int __user *)arg);
2032 case LL_IOC_GROUP_LOCK:
2033 return ll_get_grouplock(inode, file, arg);
2034 case LL_IOC_GROUP_UNLOCK:
2035 return ll_put_grouplock(inode, file, arg);
2036 case IOC_OBD_STATFS:
2037 return ll_obd_statfs(inode, (void __user *)arg);
2039 /* We need to special case any other ioctls we want to handle,
2040 * to send them to the MDS/OST as appropriate and to properly
2041 * network encode the arg field.
2042 case FSFILT_IOC_SETVERSION_OLD:
2043 case FSFILT_IOC_SETVERSION:
2045 case LL_IOC_FLUSHCTX:
2046 return ll_flush_ctx(inode);
2047 case LL_IOC_PATH2FID: {
2048 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2049 sizeof(struct lu_fid)))
2054 case LL_IOC_GETPARENT:
2055 return ll_getparent(file, (struct getparent __user *)arg);
2056 case OBD_IOC_FID2PATH:
2057 return ll_fid2path(inode, (void __user *)arg);
2058 case LL_IOC_DATA_VERSION: {
2059 struct ioc_data_version idv;
2062 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2065 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2066 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2067 if (rc == 0 && copy_to_user((char __user *)arg, &idv,
2074 case LL_IOC_GET_MDTIDX: {
2077 mdtidx = ll_get_mdt_idx(inode);
2081 if (put_user(mdtidx, (int __user *)arg))
2086 case OBD_IOC_GETDTNAME:
2087 case OBD_IOC_GETMDNAME:
2088 return ll_get_obd_name(inode, cmd, arg);
2089 case LL_IOC_HSM_STATE_GET: {
2090 struct md_op_data *op_data;
2091 struct hsm_user_state *hus;
2094 hus = kzalloc(sizeof(*hus), GFP_NOFS);
2098 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2099 LUSTRE_OPC_ANY, hus);
2100 if (IS_ERR(op_data)) {
2102 return PTR_ERR(op_data);
2105 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2108 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2111 ll_finish_md_op_data(op_data);
2115 case LL_IOC_HSM_STATE_SET: {
2116 struct hsm_state_set *hss;
2119 hss = memdup_user((char __user *)arg, sizeof(*hss));
2121 return PTR_ERR(hss);
2123 rc = ll_hsm_state_set(inode, hss);
2128 case LL_IOC_HSM_ACTION: {
2129 struct md_op_data *op_data;
2130 struct hsm_current_action *hca;
2133 hca = kzalloc(sizeof(*hca), GFP_NOFS);
2137 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2138 LUSTRE_OPC_ANY, hca);
2139 if (IS_ERR(op_data)) {
2141 return PTR_ERR(op_data);
2144 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2147 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2150 ll_finish_md_op_data(op_data);
2154 case LL_IOC_SET_LEASE: {
2155 struct ll_inode_info *lli = ll_i2info(inode);
2156 struct obd_client_handle *och = NULL;
2161 case LL_LEASE_WRLCK:
2162 if (!(file->f_mode & FMODE_WRITE))
2164 fmode = FMODE_WRITE;
2166 case LL_LEASE_RDLCK:
2167 if (!(file->f_mode & FMODE_READ))
2171 case LL_LEASE_UNLCK:
2172 mutex_lock(&lli->lli_och_mutex);
2173 if (fd->fd_lease_och) {
2174 och = fd->fd_lease_och;
2175 fd->fd_lease_och = NULL;
2177 mutex_unlock(&lli->lli_och_mutex);
2182 fmode = och->och_flags;
2183 rc = ll_lease_close(och, inode, &lease_broken);
2190 return ll_lease_type_from_fmode(fmode);
2195 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
2197 /* apply for lease */
2198 och = ll_lease_open(inode, file, fmode, 0);
2200 return PTR_ERR(och);
2203 mutex_lock(&lli->lli_och_mutex);
2204 if (!fd->fd_lease_och) {
2205 fd->fd_lease_och = och;
2208 mutex_unlock(&lli->lli_och_mutex);
2210 /* impossible now that only excl is supported for now */
2211 ll_lease_close(och, inode, &lease_broken);
2216 case LL_IOC_GET_LEASE: {
2217 struct ll_inode_info *lli = ll_i2info(inode);
2218 struct ldlm_lock *lock = NULL;
2221 mutex_lock(&lli->lli_och_mutex);
2222 if (fd->fd_lease_och) {
2223 struct obd_client_handle *och = fd->fd_lease_och;
2225 lock = ldlm_handle2lock(&och->och_lease_handle);
2227 lock_res_and_lock(lock);
2228 if (!ldlm_is_cancel(lock))
2229 fmode = och->och_flags;
2230 unlock_res_and_lock(lock);
2231 LDLM_LOCK_PUT(lock);
2234 mutex_unlock(&lli->lli_och_mutex);
2235 return ll_lease_type_from_fmode(fmode);
2237 case LL_IOC_HSM_IMPORT: {
2238 struct hsm_user_import *hui;
2240 hui = memdup_user((void __user *)arg, sizeof(*hui));
2242 return PTR_ERR(hui);
2244 rc = ll_hsm_import(inode, file, hui);
2252 if (ll_iocontrol_call(inode, file, cmd, arg, &err) ==
2256 return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2257 (void __user *)arg);
2262 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2264 struct inode *inode = file_inode(file);
2265 loff_t retval, eof = 0;
2267 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2268 (origin == SEEK_CUR) ? file->f_pos : 0);
2269 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p), to=%llu=%#llx(%d)\n",
2270 PFID(ll_inode2fid(inode)), inode, retval, retval, origin);
2271 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2273 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2274 retval = ll_glimpse_size(inode);
2277 eof = i_size_read(inode);
2280 return generic_file_llseek_size(file, offset, origin,
2281 ll_file_maxbytes(inode), eof);
2284 static int ll_flush(struct file *file, fl_owner_t id)
2286 struct inode *inode = file_inode(file);
2287 struct ll_inode_info *lli = ll_i2info(inode);
2288 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2291 LASSERT(!S_ISDIR(inode->i_mode));
2293 /* catch async errors that were recorded back when async writeback
2294 * failed for pages in this mapping.
2296 rc = lli->lli_async_rc;
2297 lli->lli_async_rc = 0;
2298 if (lli->lli_clob) {
2299 err = lov_read_and_clear_async_rc(lli->lli_clob);
2304 /* The application has been told about write failure already.
2305 * Do not report failure again.
2307 if (fd->fd_write_failed)
2309 return rc ? -EIO : 0;
2313 * Called to make sure a portion of file has been written out.
2314 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2316 * Return how many pages have been written.
2318 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2319 enum cl_fsync_mode mode, int ignore_layout)
2323 struct cl_fsync_io *fio;
2327 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2328 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2331 env = cl_env_get(&refcheck);
2333 return PTR_ERR(env);
2335 io = vvp_env_thread_io(env);
2336 io->ci_obj = ll_i2info(inode)->lli_clob;
2337 io->ci_ignore_layout = ignore_layout;
2339 /* initialize parameters for sync */
2340 fio = &io->u.ci_fsync;
2341 fio->fi_start = start;
2343 fio->fi_fid = ll_inode2fid(inode);
2344 fio->fi_mode = mode;
2345 fio->fi_nr_written = 0;
2347 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2348 result = cl_io_loop(env, io);
2350 result = io->ci_result;
2352 result = fio->fi_nr_written;
2353 cl_io_fini(env, io);
2354 cl_env_put(env, &refcheck);
2359 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2361 struct inode *inode = file_inode(file);
2362 struct ll_inode_info *lli = ll_i2info(inode);
2363 struct ptlrpc_request *req;
2366 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p)\n",
2367 PFID(ll_inode2fid(inode)), inode);
2368 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2370 rc = file_write_and_wait_range(file, start, end);
2373 /* catch async errors that were recorded back when async writeback
2374 * failed for pages in this mapping.
2376 if (!S_ISDIR(inode->i_mode)) {
2377 err = lli->lli_async_rc;
2378 lli->lli_async_rc = 0;
2381 if (lli->lli_clob) {
2382 err = lov_read_and_clear_async_rc(lli->lli_clob);
2388 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
2392 ptlrpc_req_finished(req);
2394 if (S_ISREG(inode->i_mode)) {
2395 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2397 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2398 if (rc == 0 && err < 0)
2401 fd->fd_write_failed = true;
2403 fd->fd_write_failed = false;
2406 inode_unlock(inode);
2411 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2413 struct inode *inode = file_inode(file);
2414 struct ll_sb_info *sbi = ll_i2sbi(inode);
2415 struct ldlm_enqueue_info einfo = {
2416 .ei_type = LDLM_FLOCK,
2417 .ei_cb_cp = ldlm_flock_completion_ast,
2418 .ei_cbdata = file_lock,
2420 struct md_op_data *op_data;
2421 struct lustre_handle lockh = {0};
2422 union ldlm_policy_data flock = { { 0 } };
2423 int fl_type = file_lock->fl_type;
2428 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID " file_lock=%p\n",
2429 PFID(ll_inode2fid(inode)), file_lock);
2431 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2433 if (file_lock->fl_flags & FL_FLOCK)
2434 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2435 else if (!(file_lock->fl_flags & FL_POSIX))
2438 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2439 flock.l_flock.pid = file_lock->fl_pid;
2440 flock.l_flock.start = file_lock->fl_start;
2441 flock.l_flock.end = file_lock->fl_end;
2443 /* Somewhat ugly workaround for svc lockd.
2444 * lockd installs custom fl_lmops->lm_compare_owner that checks
2445 * for the fl_owner to be the same (which it always is on local node
2446 * I guess between lockd processes) and then compares pid.
2447 * As such we assign pid to the owner field to make it all work,
2448 * conflict with normal locks is unlikely since pid space and
2449 * pointer space for current->files are not intersecting
2451 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2452 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2456 einfo.ei_mode = LCK_PR;
2459 /* An unlock request may or may not have any relation to
2460 * existing locks so we may not be able to pass a lock handle
2461 * via a normal ldlm_lock_cancel() request. The request may even
2462 * unlock a byte range in the middle of an existing lock. In
2463 * order to process an unlock request we need all of the same
2464 * information that is given with a normal read or write record
2465 * lock request. To avoid creating another ldlm unlock (cancel)
2466 * message we'll treat a LCK_NL flock request as an unlock.
2468 einfo.ei_mode = LCK_NL;
2471 einfo.ei_mode = LCK_PW;
2474 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
2489 flags = LDLM_FL_BLOCK_NOWAIT;
2495 flags = LDLM_FL_TEST_LOCK;
2498 CERROR("unknown fcntl lock command: %d\n", cmd);
2503 * Save the old mode so that if the mode in the lock changes we
2504 * can decrement the appropriate reader or writer refcount.
2506 file_lock->fl_type = einfo.ei_mode;
2508 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2509 LUSTRE_OPC_ANY, NULL);
2510 if (IS_ERR(op_data))
2511 return PTR_ERR(op_data);
2513 CDEBUG(D_DLMTRACE, "inode=" DFID ", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n",
2514 PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags,
2515 einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2517 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
2520 /* Restore the file lock type if not TEST lock. */
2521 if (!(flags & LDLM_FL_TEST_LOCK))
2522 file_lock->fl_type = fl_type;
2524 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
2525 !(flags & LDLM_FL_TEST_LOCK))
2526 rc2 = locks_lock_file_wait(file, file_lock);
2528 if (rc2 && file_lock->fl_type != F_UNLCK) {
2529 einfo.ei_mode = LCK_NL;
2530 md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
2535 ll_finish_md_op_data(op_data);
2540 int ll_get_fid_by_name(struct inode *parent, const char *name,
2541 int namelen, struct lu_fid *fid,
2542 struct inode **inode)
2544 struct md_op_data *op_data = NULL;
2545 struct ptlrpc_request *req;
2546 struct mdt_body *body;
2549 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
2550 LUSTRE_OPC_ANY, NULL);
2551 if (IS_ERR(op_data))
2552 return PTR_ERR(op_data);
2554 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
2555 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
2556 ll_finish_md_op_data(op_data);
2560 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2566 *fid = body->mbo_fid1;
2569 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
2571 ptlrpc_req_finished(req);
2575 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
2576 const char *name, int namelen)
2578 struct ptlrpc_request *request = NULL;
2579 struct obd_client_handle *och = NULL;
2580 struct inode *child_inode = NULL;
2581 struct dentry *dchild = NULL;
2582 struct md_op_data *op_data;
2583 struct mdt_body *body;
2584 u64 data_version = 0;
2588 CDEBUG(D_VFSTRACE, "migrate %s under " DFID " to MDT%d\n",
2589 name, PFID(ll_inode2fid(parent)), mdtidx);
2591 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
2592 0, LUSTRE_OPC_ANY, NULL);
2593 if (IS_ERR(op_data))
2594 return PTR_ERR(op_data);
2596 /* Get child FID first */
2597 qstr.hash = full_name_hash(parent, name, namelen);
2600 dchild = d_lookup(file_dentry(file), &qstr);
2602 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2603 if (dchild->d_inode)
2604 child_inode = igrab(dchild->d_inode);
2609 rc = ll_get_fid_by_name(parent, name, namelen,
2610 &op_data->op_fid3, &child_inode);
2620 inode_lock(child_inode);
2621 op_data->op_fid3 = *ll_inode2fid(child_inode);
2622 if (!fid_is_sane(&op_data->op_fid3)) {
2623 CERROR("%s: migrate %s, but fid " DFID " is insane\n",
2624 ll_get_fsname(parent->i_sb, NULL, 0), name,
2625 PFID(&op_data->op_fid3));
2630 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
2635 CDEBUG(D_INFO, "%s: " DFID " is already on MDT%d.\n", name,
2636 PFID(&op_data->op_fid3), mdtidx);
2641 if (S_ISREG(child_inode->i_mode)) {
2642 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
2649 rc = ll_data_version(child_inode, &data_version,
2654 op_data->op_handle = och->och_fh;
2655 op_data->op_data = och->och_mod;
2656 op_data->op_data_version = data_version;
2657 op_data->op_lease_handle = och->och_lease_handle;
2658 op_data->op_bias |= MDS_RENAME_MIGRATE;
2661 op_data->op_mds = mdtidx;
2662 op_data->op_cli_flags = CLI_MIGRATE;
2663 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
2664 namelen, name, namelen, &request);
2667 ll_update_times(request, parent);
2669 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2673 * If the server does release layout lock, then we cleanup
2674 * the client och here, otherwise release it in out_close:
2676 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
2677 obd_mod_put(och->och_mod);
2678 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
2680 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
2687 ptlrpc_req_finished(request);
2691 /* Try again if the file layout has changed. */
2692 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
2696 if (och) /* close the file */
2697 ll_lease_close(och, child_inode, NULL);
2699 clear_nlink(child_inode);
2701 inode_unlock(child_inode);
2704 ll_finish_md_op_data(op_data);
2709 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2715 * test if some locks matching bits and l_req_mode are acquired
2716 * - bits can be in different locks
2717 * - if found clear the common lock bits in *bits
2718 * - the bits not found, are kept in *bits
2720 * \param bits [IN] searched lock bits [IN]
2721 * \param l_req_mode [IN] searched lock mode
2722 * \retval boolean, true iff all bits are found
2724 int ll_have_md_lock(struct inode *inode, __u64 *bits,
2725 enum ldlm_mode l_req_mode)
2727 struct lustre_handle lockh;
2728 union ldlm_policy_data policy;
2729 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
2730 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
2738 fid = &ll_i2info(inode)->lli_fid;
2739 CDEBUG(D_INFO, "trying to match res " DFID " mode %s\n", PFID(fid),
2740 ldlm_lockname[mode]);
2742 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2743 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2744 policy.l_inodebits.bits = *bits & (1 << i);
2745 if (policy.l_inodebits.bits == 0)
2748 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2749 &policy, mode, &lockh)) {
2750 struct ldlm_lock *lock;
2752 lock = ldlm_handle2lock(&lockh);
2755 ~(lock->l_policy_data.l_inodebits.bits);
2756 LDLM_LOCK_PUT(lock);
2758 *bits &= ~policy.l_inodebits.bits;
2765 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
2766 struct lustre_handle *lockh, __u64 flags,
2767 enum ldlm_mode mode)
2769 union ldlm_policy_data policy = { .l_inodebits = { bits } };
2772 fid = &ll_i2info(inode)->lli_fid;
2773 CDEBUG(D_INFO, "trying to match res " DFID "\n", PFID(fid));
2775 return md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED,
2776 fid, LDLM_IBITS, &policy, mode, lockh);
2779 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2781 /* Already unlinked. Just update nlink and return success */
2782 if (rc == -ENOENT) {
2784 /* If it is striped directory, and there is bad stripe
2785 * Let's revalidate the dentry again, instead of returning
2788 if (S_ISDIR(inode->i_mode) && ll_i2info(inode)->lli_lsm_md)
2791 /* This path cannot be hit for regular files unless in
2792 * case of obscure races, so no need to validate size.
2794 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2796 } else if (rc != 0) {
2797 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
2798 "%s: revalidate FID " DFID " error: rc = %d\n",
2799 ll_get_fsname(inode->i_sb, NULL, 0),
2800 PFID(ll_inode2fid(inode)), rc);
2806 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2808 struct inode *inode = d_inode(dentry);
2809 struct ptlrpc_request *req = NULL;
2810 struct obd_export *exp;
2813 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p),name=%pd\n",
2814 PFID(ll_inode2fid(inode)), inode, dentry);
2816 exp = ll_i2mdexp(inode);
2818 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2819 * But under CMD case, it caused some lock issues, should be fixed
2820 * with new CMD ibits lock. See bug 12718
2822 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
2823 struct lookup_intent oit = { .it_op = IT_GETATTR };
2824 struct md_op_data *op_data;
2826 if (ibits == MDS_INODELOCK_LOOKUP)
2827 oit.it_op = IT_LOOKUP;
2829 /* Call getattr by fid, so do not provide name at all. */
2830 op_data = ll_prep_md_op_data(NULL, inode,
2832 LUSTRE_OPC_ANY, NULL);
2833 if (IS_ERR(op_data))
2834 return PTR_ERR(op_data);
2836 rc = md_intent_lock(exp, op_data, &oit, &req,
2837 &ll_md_blocking_ast, 0);
2838 ll_finish_md_op_data(op_data);
2840 rc = ll_inode_revalidate_fini(inode, rc);
2844 rc = ll_revalidate_it_finish(req, &oit, inode);
2846 ll_intent_release(&oit);
2850 /* Unlinked? Unhash dentry, so it is not picked up later by
2851 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2852 * here to preserve get_cwd functionality on 2.6.
2855 if (!d_inode(dentry)->i_nlink) {
2856 spin_lock(&inode->i_lock);
2857 d_lustre_invalidate(dentry, 0);
2858 spin_unlock(&inode->i_lock);
2861 ll_lookup_finish_locks(&oit, inode);
2862 } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) {
2863 struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry));
2864 u64 valid = OBD_MD_FLGETATTR;
2865 struct md_op_data *op_data;
2868 if (S_ISREG(inode->i_mode)) {
2869 rc = ll_get_default_mdsize(sbi, &ealen);
2872 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2875 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2876 0, ealen, LUSTRE_OPC_ANY,
2878 if (IS_ERR(op_data))
2879 return PTR_ERR(op_data);
2881 op_data->op_valid = valid;
2882 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2883 ll_finish_md_op_data(op_data);
2885 return ll_inode_revalidate_fini(inode, rc);
2887 rc = ll_prep_inode(&inode, req, NULL, NULL);
2890 ptlrpc_req_finished(req);
2894 static int ll_merge_md_attr(struct inode *inode)
2896 struct cl_attr attr = { 0 };
2899 LASSERT(ll_i2info(inode)->lli_lsm_md);
2900 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
2901 &attr, ll_md_blocking_ast);
2905 set_nlink(inode, attr.cat_nlink);
2906 inode->i_blocks = attr.cat_blocks;
2907 i_size_write(inode, attr.cat_size);
2909 ll_i2info(inode)->lli_atime = attr.cat_atime;
2910 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
2911 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
2916 static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2918 struct inode *inode = d_inode(dentry);
2921 rc = __ll_inode_revalidate(dentry, ibits);
2925 /* if object isn't regular file, don't validate size */
2926 if (!S_ISREG(inode->i_mode)) {
2927 if (S_ISDIR(inode->i_mode) &&
2928 ll_i2info(inode)->lli_lsm_md) {
2929 rc = ll_merge_md_attr(inode);
2934 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
2935 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
2936 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
2938 struct ll_inode_info *lli = ll_i2info(inode);
2940 /* In case of restore, the MDT has the right size and has
2941 * already send it back without granting the layout lock,
2942 * inode is up-to-date so glimpse is useless.
2943 * Also to glimpse we need the layout, in case of a running
2944 * restore the MDT holds the layout lock so the glimpse will
2945 * block up to the end of restore (getattr will block)
2947 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags))
2948 rc = ll_glimpse_size(inode);
2953 int ll_getattr(const struct path *path, struct kstat *stat,
2954 u32 request_mask, unsigned int flags)
2956 struct inode *inode = d_inode(path->dentry);
2957 struct ll_sb_info *sbi = ll_i2sbi(inode);
2958 struct ll_inode_info *lli = ll_i2info(inode);
2961 res = ll_inode_revalidate(path->dentry,
2962 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP);
2963 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
2968 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
2970 stat->dev = inode->i_sb->s_dev;
2971 if (ll_need_32bit_api(sbi))
2972 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
2974 stat->ino = inode->i_ino;
2975 stat->mode = inode->i_mode;
2976 stat->uid = inode->i_uid;
2977 stat->gid = inode->i_gid;
2978 stat->rdev = inode->i_rdev;
2979 stat->atime = inode->i_atime;
2980 stat->mtime = inode->i_mtime;
2981 stat->ctime = inode->i_ctime;
2982 stat->blksize = 1 << inode->i_blkbits;
2984 stat->nlink = inode->i_nlink;
2985 stat->size = i_size_read(inode);
2986 stat->blocks = inode->i_blocks;
2991 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2992 __u64 start, __u64 len)
2996 struct fiemap *fiemap;
2997 unsigned int extent_count = fieinfo->fi_extents_max;
2999 num_bytes = sizeof(*fiemap) + (extent_count *
3000 sizeof(struct fiemap_extent));
3001 fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS);
3005 fiemap->fm_flags = fieinfo->fi_flags;
3006 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3007 fiemap->fm_start = start;
3008 fiemap->fm_length = len;
3010 if (extent_count > 0 &&
3011 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3012 sizeof(struct fiemap_extent))) {
3017 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3019 fieinfo->fi_flags = fiemap->fm_flags;
3020 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3021 if (extent_count > 0 &&
3022 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3023 fiemap->fm_mapped_extents *
3024 sizeof(struct fiemap_extent))) {
3033 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3035 struct ll_inode_info *lli = ll_i2info(inode);
3036 struct posix_acl *acl = NULL;
3038 spin_lock(&lli->lli_lock);
3039 /* VFS' acl_permission_check->check_acl will release the refcount */
3040 acl = posix_acl_dup(lli->lli_posix_acl);
3041 spin_unlock(&lli->lli_lock);
3046 int ll_inode_permission(struct inode *inode, int mask)
3048 struct ll_sb_info *sbi;
3049 struct root_squash_info *squash;
3050 const struct cred *old_cred = NULL;
3051 struct cred *cred = NULL;
3052 bool squash_id = false;
3056 if (mask & MAY_NOT_BLOCK)
3059 /* as root inode are NOT getting validated in lookup operation,
3060 * need to do it before permission check.
3063 if (is_root_inode(inode)) {
3064 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3065 MDS_INODELOCK_LOOKUP);
3070 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p), inode mode %x mask %o\n",
3071 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3073 /* squash fsuid/fsgid if needed */
3074 sbi = ll_i2sbi(inode);
3075 squash = &sbi->ll_squash;
3076 if (unlikely(squash->rsi_uid &&
3077 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3078 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3083 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3084 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3085 squash->rsi_uid, squash->rsi_gid);
3088 * update current process's credentials
3091 cred = prepare_creds();
3095 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3096 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3097 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3098 if ((1 << cap) & CFS_CAP_FS_MASK)
3099 cap_lower(cred->cap_effective, cap);
3101 old_cred = override_creds(cred);
3104 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
3105 rc = generic_permission(inode, mask);
3107 /* restore current process's credentials and FS capability */
3109 revert_creds(old_cred);
3116 /* -o localflock - only provides locally consistent flock locks */
3117 const struct file_operations ll_file_operations = {
3118 .read_iter = ll_file_read_iter,
3119 .write_iter = ll_file_write_iter,
3120 .unlocked_ioctl = ll_file_ioctl,
3121 .open = ll_file_open,
3122 .release = ll_file_release,
3123 .mmap = ll_file_mmap,
3124 .llseek = ll_file_seek,
3125 .splice_read = generic_file_splice_read,
3130 const struct file_operations ll_file_operations_flock = {
3131 .read_iter = ll_file_read_iter,
3132 .write_iter = ll_file_write_iter,
3133 .unlocked_ioctl = ll_file_ioctl,
3134 .open = ll_file_open,
3135 .release = ll_file_release,
3136 .mmap = ll_file_mmap,
3137 .llseek = ll_file_seek,
3138 .splice_read = generic_file_splice_read,
3141 .flock = ll_file_flock,
3142 .lock = ll_file_flock
3145 /* These are for -o noflock - to return ENOSYS on flock calls */
3146 const struct file_operations ll_file_operations_noflock = {
3147 .read_iter = ll_file_read_iter,
3148 .write_iter = ll_file_write_iter,
3149 .unlocked_ioctl = ll_file_ioctl,
3150 .open = ll_file_open,
3151 .release = ll_file_release,
3152 .mmap = ll_file_mmap,
3153 .llseek = ll_file_seek,
3154 .splice_read = generic_file_splice_read,
3157 .flock = ll_file_noflock,
3158 .lock = ll_file_noflock
3161 const struct inode_operations ll_file_inode_operations = {
3162 .setattr = ll_setattr,
3163 .getattr = ll_getattr,
3164 .permission = ll_inode_permission,
3165 .listxattr = ll_listxattr,
3166 .fiemap = ll_fiemap,
3167 .get_acl = ll_get_acl,
3170 /* dynamic ioctl number support routines */
3171 static struct llioc_ctl_data {
3172 struct rw_semaphore ioc_sem;
3173 struct list_head ioc_head;
3175 __RWSEM_INITIALIZER(llioc.ioc_sem),
3176 LIST_HEAD_INIT(llioc.ioc_head)
3180 struct list_head iocd_list;
3181 unsigned int iocd_size;
3182 llioc_callback_t iocd_cb;
3183 unsigned int iocd_count;
3184 unsigned int iocd_cmd[0];
3187 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3190 struct llioc_data *in_data = NULL;
3192 if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0)
3195 size = sizeof(*in_data) + count * sizeof(unsigned int);
3196 in_data = kzalloc(size, GFP_NOFS);
3200 in_data->iocd_size = size;
3201 in_data->iocd_cb = cb;
3202 in_data->iocd_count = count;
3203 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3205 down_write(&llioc.ioc_sem);
3206 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3207 up_write(&llioc.ioc_sem);
3211 EXPORT_SYMBOL(ll_iocontrol_register);
3213 void ll_iocontrol_unregister(void *magic)
3215 struct llioc_data *tmp;
3220 down_write(&llioc.ioc_sem);
3221 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3223 list_del(&tmp->iocd_list);
3224 up_write(&llioc.ioc_sem);
3230 up_write(&llioc.ioc_sem);
3232 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3234 EXPORT_SYMBOL(ll_iocontrol_unregister);
3236 static enum llioc_iter
3237 ll_iocontrol_call(struct inode *inode, struct file *file,
3238 unsigned int cmd, unsigned long arg, int *rcp)
3240 enum llioc_iter ret = LLIOC_CONT;
3241 struct llioc_data *data;
3242 int rc = -EINVAL, i;
3244 down_read(&llioc.ioc_sem);
3245 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3246 for (i = 0; i < data->iocd_count; i++) {
3247 if (cmd != data->iocd_cmd[i])
3250 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3254 if (ret == LLIOC_STOP)
3257 up_read(&llioc.ioc_sem);
3264 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3266 struct ll_inode_info *lli = ll_i2info(inode);
3267 struct cl_object *obj = lli->lli_clob;
3275 env = cl_env_get(&refcheck);
3277 return PTR_ERR(env);
3279 rc = cl_conf_set(env, obj, conf);
3283 if (conf->coc_opc == OBJECT_CONF_SET) {
3284 struct ldlm_lock *lock = conf->coc_lock;
3285 struct cl_layout cl = {
3290 LASSERT(ldlm_has_layout(lock));
3292 /* it can only be allowed to match after layout is
3293 * applied to inode otherwise false layout would be
3294 * seen. Applying layout should happen before dropping
3297 ldlm_lock_allow_match(lock);
3299 rc = cl_object_layout_get(env, obj, &cl);
3303 CDEBUG(D_VFSTRACE, DFID ": layout version change: %u -> %u\n",
3304 PFID(&lli->lli_fid), ll_layout_version_get(lli),
3306 ll_layout_version_set(lli, cl.cl_layout_gen);
3309 cl_env_put(env, &refcheck);
3313 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3314 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3317 struct ll_sb_info *sbi = ll_i2sbi(inode);
3318 struct ptlrpc_request *req;
3319 struct mdt_body *body;
3325 CDEBUG(D_INODE, DFID " LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3326 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3327 lock->l_lvb_data, lock->l_lvb_len);
3329 if (lock->l_lvb_data && ldlm_is_lvb_ready(lock))
3332 /* if layout lock was granted right away, the layout is returned
3333 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3334 * blocked and then granted via completion ast, we have to fetch
3335 * layout here. Please note that we can't use the LVB buffer in
3336 * completion AST because it doesn't have a large enough buffer
3338 rc = ll_get_default_mdsize(sbi, &lmmsize);
3340 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
3341 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3346 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3352 lmmsize = body->mbo_eadatasize;
3353 if (lmmsize == 0) /* empty layout */ {
3358 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3364 lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS);
3370 memcpy(lvbdata, lmm, lmmsize);
3371 lock_res_and_lock(lock);
3372 if (lock->l_lvb_data)
3373 kvfree(lock->l_lvb_data);
3375 lock->l_lvb_data = lvbdata;
3376 lock->l_lvb_len = lmmsize;
3377 unlock_res_and_lock(lock);
3380 ptlrpc_req_finished(req);
3385 * Apply the layout to the inode. Layout lock is held and will be released
3388 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
3389 struct inode *inode)
3391 struct ll_inode_info *lli = ll_i2info(inode);
3392 struct ll_sb_info *sbi = ll_i2sbi(inode);
3393 struct ldlm_lock *lock;
3394 struct cl_object_conf conf;
3397 bool wait_layout = false;
3399 LASSERT(lustre_handle_is_used(lockh));
3401 lock = ldlm_handle2lock(lockh);
3403 LASSERT(ldlm_has_layout(lock));
3405 LDLM_DEBUG(lock, "File " DFID "(%p) being reconfigured",
3406 PFID(&lli->lli_fid), inode);
3408 /* in case this is a caching lock and reinstate with new inode */
3409 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
3411 lock_res_and_lock(lock);
3412 lvb_ready = ldlm_is_lvb_ready(lock);
3413 unlock_res_and_lock(lock);
3414 /* checking lvb_ready is racy but this is okay. The worst case is
3415 * that multi processes may configure the file on the same time.
3422 rc = ll_layout_fetch(inode, lock);
3426 /* for layout lock, lmm is returned in lock's lvb.
3427 * lvb_data is immutable if the lock is held so it's safe to access it
3430 * set layout to file. Unlikely this will fail as old layout was
3433 memset(&conf, 0, sizeof(conf));
3434 conf.coc_opc = OBJECT_CONF_SET;
3435 conf.coc_inode = inode;
3436 conf.coc_lock = lock;
3437 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
3438 conf.u.coc_layout.lb_len = lock->l_lvb_len;
3439 rc = ll_layout_conf(inode, &conf);
3441 /* refresh layout failed, need to wait */
3442 wait_layout = rc == -EBUSY;
3445 LDLM_LOCK_PUT(lock);
3446 ldlm_lock_decref(lockh, mode);
3448 /* wait for IO to complete if it's still being used. */
3450 CDEBUG(D_INODE, "%s: " DFID "(%p) wait for layout reconf\n",
3451 ll_get_fsname(inode->i_sb, NULL, 0),
3452 PFID(&lli->lli_fid), inode);
3454 memset(&conf, 0, sizeof(conf));
3455 conf.coc_opc = OBJECT_CONF_WAIT;
3456 conf.coc_inode = inode;
3457 rc = ll_layout_conf(inode, &conf);
3462 "%s: file=" DFID " waiting layout return: %d.\n",
3463 ll_get_fsname(inode->i_sb, NULL, 0),
3464 PFID(&lli->lli_fid), rc);
3469 static int ll_layout_refresh_locked(struct inode *inode)
3471 struct ll_inode_info *lli = ll_i2info(inode);
3472 struct ll_sb_info *sbi = ll_i2sbi(inode);
3473 struct md_op_data *op_data;
3474 struct lookup_intent it;
3475 struct lustre_handle lockh;
3476 enum ldlm_mode mode;
3477 struct ldlm_enqueue_info einfo = {
3478 .ei_type = LDLM_IBITS,
3480 .ei_cb_bl = &ll_md_blocking_ast,
3481 .ei_cb_cp = &ldlm_completion_ast,
3486 /* mostly layout lock is caching on the local side, so try to match
3487 * it before grabbing layout lock mutex.
3489 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
3490 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
3491 if (mode != 0) { /* hit cached lock */
3492 rc = ll_layout_lock_set(&lockh, mode, inode);
3498 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3499 0, 0, LUSTRE_OPC_ANY, NULL);
3500 if (IS_ERR(op_data))
3501 return PTR_ERR(op_data);
3503 /* have to enqueue one */
3504 memset(&it, 0, sizeof(it));
3505 it.it_op = IT_LAYOUT;
3506 lockh.cookie = 0ULL;
3508 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file " DFID "(%p)",
3509 ll_get_fsname(inode->i_sb, NULL, 0),
3510 PFID(&lli->lli_fid), inode);
3512 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
3513 ptlrpc_req_finished(it.it_request);
3514 it.it_request = NULL;
3516 ll_finish_md_op_data(op_data);
3518 mode = it.it_lock_mode;
3519 it.it_lock_mode = 0;
3520 ll_intent_drop_lock(&it);
3523 /* set lock data in case this is a new lock */
3524 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3525 rc = ll_layout_lock_set(&lockh, mode, inode);
3534 * This function checks if there exists a LAYOUT lock on the client side,
3535 * or enqueues it if it doesn't have one in cache.
3537 * This function will not hold layout lock so it may be revoked any time after
3538 * this function returns. Any operations depend on layout should be redone
3541 * This function should be called before lov_io_init() to get an uptodate
3542 * layout version, the caller should save the version number and after IO
3543 * is finished, this function should be called again to verify that layout
3544 * is not changed during IO time.
3546 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3548 struct ll_inode_info *lli = ll_i2info(inode);
3549 struct ll_sb_info *sbi = ll_i2sbi(inode);
3552 *gen = ll_layout_version_get(lli);
3553 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
3557 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3558 LASSERT(S_ISREG(inode->i_mode));
3560 /* take layout lock mutex to enqueue layout lock exclusively. */
3561 mutex_lock(&lli->lli_layout_mutex);
3563 rc = ll_layout_refresh_locked(inode);
3567 *gen = ll_layout_version_get(lli);
3569 mutex_unlock(&lli->lli_layout_mutex);
3575 * This function send a restore request to the MDT
3577 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
3579 struct hsm_user_request *hur;
3582 len = sizeof(struct hsm_user_request) +
3583 sizeof(struct hsm_user_item);
3584 hur = kzalloc(len, GFP_NOFS);
3588 hur->hur_request.hr_action = HUA_RESTORE;
3589 hur->hur_request.hr_archive_id = 0;
3590 hur->hur_request.hr_flags = 0;
3591 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
3592 sizeof(hur->hur_user_item[0].hui_fid));
3593 hur->hur_user_item[0].hui_extent.offset = offset;
3594 hur->hur_user_item[0].hui_extent.length = length;
3595 hur->hur_request.hr_itemcount = 1;
3596 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,