4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
40 #include "../include/lustre_dlm.h"
41 #include <linux/pagemap.h>
42 #include <linux/file.h>
43 #include <linux/sched.h>
44 #include <linux/mount.h>
45 #include "../include/lustre/ll_fiemap.h"
46 #include "../include/lustre/lustre_ioctl.h"
47 #include "../include/lustre_swab.h"
49 #include "../include/cl_object.h"
50 #include "llite_internal.h"
53 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
55 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
58 static enum llioc_iter
59 ll_iocontrol_call(struct inode *inode, struct file *file,
60 unsigned int cmd, unsigned long arg, int *rcp);
62 static struct ll_file_data *ll_file_data_get(void)
64 struct ll_file_data *fd;
66 fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS);
69 fd->fd_write_failed = false;
73 static void ll_file_data_put(struct ll_file_data *fd)
76 kmem_cache_free(ll_file_data_slab, fd);
80 * Packs all the attributes into @op_data for the CLOSE rpc.
82 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
83 struct obd_client_handle *och)
85 struct ll_inode_info *lli = ll_i2info(inode);
87 ll_prep_md_op_data(op_data, inode, NULL, NULL,
88 0, 0, LUSTRE_OPC_ANY, NULL);
90 op_data->op_attr.ia_mode = inode->i_mode;
91 op_data->op_attr.ia_atime = inode->i_atime;
92 op_data->op_attr.ia_mtime = inode->i_mtime;
93 op_data->op_attr.ia_ctime = inode->i_ctime;
94 op_data->op_attr.ia_size = i_size_read(inode);
95 op_data->op_attr.ia_valid |= ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
96 ATTR_MTIME | ATTR_MTIME_SET |
97 ATTR_CTIME | ATTR_CTIME_SET;
98 op_data->op_attr_blocks = inode->i_blocks;
99 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
100 op_data->op_handle = och->och_fh;
103 * For HSM: if inode data has been modified, pack it so that
104 * MDT can set data dirty flag in the archive.
106 if (och->och_flags & FMODE_WRITE &&
107 test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags))
108 op_data->op_bias |= MDS_DATA_MODIFIED;
112 * Perform a close, possibly with a bias.
113 * The meaning of "data" depends on the value of "bias".
115 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
116 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
119 static int ll_close_inode_openhandle(struct obd_export *md_exp,
120 struct obd_client_handle *och,
122 enum mds_op_bias bias,
125 const struct ll_inode_info *lli = ll_i2info(inode);
126 struct md_op_data *op_data;
127 struct ptlrpc_request *req = NULL;
130 if (!class_exp2obd(md_exp)) {
131 CERROR("%s: invalid MDC connection handle closing " DFID "\n",
132 ll_get_fsname(inode->i_sb, NULL, 0),
133 PFID(&lli->lli_fid));
138 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
140 * We leak openhandle and request here on error, but not much to be
141 * done in OOM case since app won't retry close on error either.
148 ll_prepare_close(inode, op_data, och);
150 case MDS_CLOSE_LAYOUT_SWAP:
152 op_data->op_bias |= MDS_CLOSE_LAYOUT_SWAP;
153 op_data->op_data_version = 0;
154 op_data->op_lease_handle = och->och_lease_handle;
155 op_data->op_fid2 = *ll_inode2fid(data);
158 case MDS_HSM_RELEASE:
160 op_data->op_bias |= MDS_HSM_RELEASE;
161 op_data->op_data_version = *(__u64 *)data;
162 op_data->op_lease_handle = och->och_lease_handle;
163 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
171 rc = md_close(md_exp, op_data, och->och_mod, &req);
172 if (rc && rc != -EINTR) {
173 CERROR("%s: inode " DFID " mdc close failed: rc = %d\n",
174 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
177 if (op_data->op_bias & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP) &&
179 struct mdt_body *body;
181 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
182 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
186 ll_finish_md_op_data(op_data);
189 md_clear_open_replay_data(md_exp, och);
190 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
193 ptlrpc_req_finished(req);
197 int ll_md_real_close(struct inode *inode, fmode_t fmode)
199 struct ll_inode_info *lli = ll_i2info(inode);
200 struct obd_client_handle **och_p;
201 struct obd_client_handle *och;
205 if (fmode & FMODE_WRITE) {
206 och_p = &lli->lli_mds_write_och;
207 och_usecount = &lli->lli_open_fd_write_count;
208 } else if (fmode & FMODE_EXEC) {
209 och_p = &lli->lli_mds_exec_och;
210 och_usecount = &lli->lli_open_fd_exec_count;
212 LASSERT(fmode & FMODE_READ);
213 och_p = &lli->lli_mds_read_och;
214 och_usecount = &lli->lli_open_fd_read_count;
217 mutex_lock(&lli->lli_och_mutex);
218 if (*och_usecount > 0) {
219 /* There are still users of this handle, so skip
222 mutex_unlock(&lli->lli_och_mutex);
228 mutex_unlock(&lli->lli_och_mutex);
231 /* There might be a race and this handle may already
234 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
235 och, inode, 0, NULL);
241 static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
244 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
245 struct ll_inode_info *lli = ll_i2info(inode);
247 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
248 struct lustre_handle lockh;
249 union ldlm_policy_data policy = {
250 .l_inodebits = { MDS_INODELOCK_OPEN }
254 /* clear group lock, if present */
255 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
256 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
258 if (fd->fd_lease_och) {
261 /* Usually the lease is not released when the
262 * application crashed, we need to release here.
264 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
265 CDEBUG(rc ? D_ERROR : D_INODE,
266 "Clean up lease " DFID " %d/%d\n",
267 PFID(&lli->lli_fid), rc, lease_broken);
269 fd->fd_lease_och = NULL;
273 rc = ll_close_inode_openhandle(md_exp, fd->fd_och, inode, 0,
279 /* Let's see if we have good enough OPEN lock on the file and if
280 * we can skip talking to MDS
283 mutex_lock(&lli->lli_och_mutex);
284 if (fd->fd_omode & FMODE_WRITE) {
286 LASSERT(lli->lli_open_fd_write_count);
287 lli->lli_open_fd_write_count--;
288 } else if (fd->fd_omode & FMODE_EXEC) {
290 LASSERT(lli->lli_open_fd_exec_count);
291 lli->lli_open_fd_exec_count--;
294 LASSERT(lli->lli_open_fd_read_count);
295 lli->lli_open_fd_read_count--;
297 mutex_unlock(&lli->lli_och_mutex);
299 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
300 LDLM_IBITS, &policy, lockmode, &lockh))
301 rc = ll_md_real_close(inode, fd->fd_omode);
304 LUSTRE_FPRIVATE(file) = NULL;
305 ll_file_data_put(fd);
310 /* While this returns an error code, fput() the caller does not, so we need
311 * to make every effort to clean up all of our state here. Also, applications
312 * rarely check close errors and even if an error is returned they will not
313 * re-try the close call.
315 int ll_file_release(struct inode *inode, struct file *file)
317 struct ll_file_data *fd;
318 struct ll_sb_info *sbi = ll_i2sbi(inode);
319 struct ll_inode_info *lli = ll_i2info(inode);
322 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
323 PFID(ll_inode2fid(inode)), inode);
325 if (!is_root_inode(inode))
326 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
327 fd = LUSTRE_FPRIVATE(file);
330 /* The last ref on @file, maybe not be the owner pid of statahead,
331 * because parent and child process can share the same file handle.
333 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
334 ll_deauthorize_statahead(inode, fd);
336 if (is_root_inode(inode)) {
337 LUSTRE_FPRIVATE(file) = NULL;
338 ll_file_data_put(fd);
342 if (!S_ISDIR(inode->i_mode)) {
344 lov_read_and_clear_async_rc(lli->lli_clob);
345 lli->lli_async_rc = 0;
348 rc = ll_md_close(sbi->ll_md_exp, inode, file);
350 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
351 libcfs_debug_dumplog();
356 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
357 struct lookup_intent *itp)
359 struct inode *inode = d_inode(de);
360 struct ll_sb_info *sbi = ll_i2sbi(inode);
361 struct dentry *parent = de->d_parent;
362 const char *name = NULL;
363 struct md_op_data *op_data;
364 struct ptlrpc_request *req = NULL;
368 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
371 * if server supports open-by-fid, or file name is invalid, don't pack
372 * name in open request
374 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
375 lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
376 name = de->d_name.name;
377 len = de->d_name.len;
380 op_data = ll_prep_md_op_data(NULL, d_inode(parent), inode, name, len,
381 O_RDWR, LUSTRE_OPC_ANY, NULL);
383 return PTR_ERR(op_data);
384 op_data->op_data = lmm;
385 op_data->op_data_size = lmmsize;
387 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
388 &ll_md_blocking_ast, 0);
389 ll_finish_md_op_data(op_data);
391 /* reason for keep own exit path - don`t flood log
392 * with messages with -ESTALE errors.
394 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
395 it_open_error(DISP_OPEN_OPEN, itp))
397 ll_release_openhandle(inode, itp);
401 if (it_disposition(itp, DISP_LOOKUP_NEG)) {
406 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
407 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
408 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
412 rc = ll_prep_inode(&inode, req, NULL, itp);
413 if (!rc && itp->it_lock_mode)
414 ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL);
417 ptlrpc_req_finished(req);
418 ll_intent_drop_lock(itp);
421 * We did open by fid, but by the time we got to the server,
422 * the object disappeared. If this is a create, we cannot really
423 * tell the userspace that the file it was trying to create
424 * does not exist. Instead let's return -ESTALE, and the VFS will
425 * retry the create with LOOKUP_REVAL that we are going to catch
426 * in ll_revalidate_dentry() and use lookup then.
428 if (rc == -ENOENT && itp->it_op & IT_CREAT)
434 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
435 struct obd_client_handle *och)
437 struct mdt_body *body;
439 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
440 och->och_fh = body->mbo_handle;
441 och->och_fid = body->mbo_fid1;
442 och->och_lease_handle.cookie = it->it_lock_handle;
443 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
444 och->och_flags = it->it_flags;
446 return md_set_open_replay_data(md_exp, och, it);
449 static int ll_local_open(struct file *file, struct lookup_intent *it,
450 struct ll_file_data *fd, struct obd_client_handle *och)
452 struct inode *inode = file_inode(file);
454 LASSERT(!LUSTRE_FPRIVATE(file));
461 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
466 LUSTRE_FPRIVATE(file) = fd;
467 ll_readahead_init(inode, &fd->fd_ras);
468 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
470 /* ll_cl_context initialize */
471 rwlock_init(&fd->fd_lock);
472 INIT_LIST_HEAD(&fd->fd_lccs);
477 /* Open a file, and (for the very first open) create objects on the OSTs at
478 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
479 * creation or open until ll_lov_setstripe() ioctl is called.
481 * If we already have the stripe MD locally then we don't request it in
482 * md_open(), by passing a lmm_size = 0.
484 * It is up to the application to ensure no other processes open this file
485 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
486 * used. We might be able to avoid races of that sort by getting lli_open_sem
487 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
488 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
490 int ll_file_open(struct inode *inode, struct file *file)
492 struct ll_inode_info *lli = ll_i2info(inode);
493 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
494 .it_flags = file->f_flags };
495 struct obd_client_handle **och_p = NULL;
496 __u64 *och_usecount = NULL;
497 struct ll_file_data *fd;
500 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
501 PFID(ll_inode2fid(inode)), inode, file->f_flags);
503 it = file->private_data; /* XXX: compat macro */
504 file->private_data = NULL; /* prevent ll_local_open assertion */
506 fd = ll_file_data_get();
513 if (S_ISDIR(inode->i_mode))
514 ll_authorize_statahead(inode, fd);
516 if (is_root_inode(inode)) {
517 LUSTRE_FPRIVATE(file) = fd;
521 if (!it || !it->it_disposition) {
522 /* Convert f_flags into access mode. We cannot use file->f_mode,
523 * because everything but O_ACCMODE mask was stripped from
526 if ((oit.it_flags + 1) & O_ACCMODE)
528 if (file->f_flags & O_TRUNC)
529 oit.it_flags |= FMODE_WRITE;
531 /* kernel only call f_op->open in dentry_open. filp_open calls
532 * dentry_open after call to open_namei that checks permissions.
533 * Only nfsd_open call dentry_open directly without checking
534 * permissions and because of that this code below is safe.
536 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
537 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
539 /* We do not want O_EXCL here, presumably we opened the file
540 * already? XXX - NFS implications?
542 oit.it_flags &= ~O_EXCL;
544 /* bug20584, if "it_flags" contains O_CREAT, the file will be
545 * created if necessary, then "IT_CREAT" should be set to keep
548 if (oit.it_flags & O_CREAT)
549 oit.it_op |= IT_CREAT;
555 /* Let's see if we have file open on MDS already. */
556 if (it->it_flags & FMODE_WRITE) {
557 och_p = &lli->lli_mds_write_och;
558 och_usecount = &lli->lli_open_fd_write_count;
559 } else if (it->it_flags & FMODE_EXEC) {
560 och_p = &lli->lli_mds_exec_och;
561 och_usecount = &lli->lli_open_fd_exec_count;
563 och_p = &lli->lli_mds_read_och;
564 och_usecount = &lli->lli_open_fd_read_count;
567 mutex_lock(&lli->lli_och_mutex);
568 if (*och_p) { /* Open handle is present */
569 if (it_disposition(it, DISP_OPEN_OPEN)) {
570 /* Well, there's extra open request that we do not need,
571 * let's close it somehow. This will decref request.
573 rc = it_open_error(DISP_OPEN_OPEN, it);
575 mutex_unlock(&lli->lli_och_mutex);
579 ll_release_openhandle(inode, it);
583 rc = ll_local_open(file, it, fd, NULL);
586 mutex_unlock(&lli->lli_och_mutex);
590 LASSERT(*och_usecount == 0);
591 if (!it->it_disposition) {
592 /* We cannot just request lock handle now, new ELC code
593 * means that one of other OPEN locks for this file
594 * could be cancelled, and since blocking ast handler
595 * would attempt to grab och_mutex as well, that would
596 * result in a deadlock
598 mutex_unlock(&lli->lli_och_mutex);
600 * Normally called under two situations:
602 * 2. revalidate with IT_OPEN (revalidate doesn't
603 * execute this intent any more).
605 * Always fetch MDS_OPEN_LOCK if this is not setstripe.
607 * Always specify MDS_OPEN_BY_FID because we don't want
608 * to get file with different fid.
610 it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
611 rc = ll_intent_file_open(file->f_path.dentry, NULL, 0, it);
617 *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS);
625 /* md_intent_lock() didn't get a request ref if there was an
626 * open error, so don't do cleanup on the request here
629 /* XXX (green): Should not we bail out on any error here, not
632 rc = it_open_error(DISP_OPEN_OPEN, it);
636 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
637 "inode %p: disposition %x, status %d\n", inode,
638 it_disposition(it, ~0), it->it_status);
640 rc = ll_local_open(file, it, fd, *och_p);
644 mutex_unlock(&lli->lli_och_mutex);
647 /* Must do this outside lli_och_mutex lock to prevent deadlock where
648 * different kind of OPEN lock for this same inode gets cancelled
651 if (!S_ISREG(inode->i_mode))
654 cl_lov_delay_create_clear(&file->f_flags);
659 if (och_p && *och_p) {
664 mutex_unlock(&lli->lli_och_mutex);
667 if (lli->lli_opendir_key == fd)
668 ll_deauthorize_statahead(inode, fd);
670 ll_file_data_put(fd);
672 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
675 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
676 ptlrpc_req_finished(it->it_request);
677 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
683 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
684 struct ldlm_lock_desc *desc,
685 void *data, int flag)
688 struct lustre_handle lockh;
691 case LDLM_CB_BLOCKING:
692 ldlm_lock2handle(lock, &lockh);
693 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
695 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
699 case LDLM_CB_CANCELING:
707 * Acquire a lease and open the file.
709 static struct obd_client_handle *
710 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
713 struct lookup_intent it = { .it_op = IT_OPEN };
714 struct ll_sb_info *sbi = ll_i2sbi(inode);
715 struct md_op_data *op_data;
716 struct ptlrpc_request *req = NULL;
717 struct lustre_handle old_handle = { 0 };
718 struct obd_client_handle *och = NULL;
722 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
723 return ERR_PTR(-EINVAL);
726 struct ll_inode_info *lli = ll_i2info(inode);
727 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
728 struct obd_client_handle **och_p;
731 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
732 return ERR_PTR(-EPERM);
734 /* Get the openhandle of the file */
736 mutex_lock(&lli->lli_och_mutex);
737 if (fd->fd_lease_och) {
738 mutex_unlock(&lli->lli_och_mutex);
743 if (file->f_mode & FMODE_WRITE) {
744 LASSERT(lli->lli_mds_write_och);
745 och_p = &lli->lli_mds_write_och;
746 och_usecount = &lli->lli_open_fd_write_count;
748 LASSERT(lli->lli_mds_read_och);
749 och_p = &lli->lli_mds_read_och;
750 och_usecount = &lli->lli_open_fd_read_count;
752 if (*och_usecount == 1) {
759 mutex_unlock(&lli->lli_och_mutex);
760 if (rc < 0) /* more than 1 opener */
764 old_handle = fd->fd_och->och_fh;
767 och = kzalloc(sizeof(*och), GFP_NOFS);
769 return ERR_PTR(-ENOMEM);
771 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
772 LUSTRE_OPC_ANY, NULL);
773 if (IS_ERR(op_data)) {
774 rc = PTR_ERR(op_data);
778 /* To tell the MDT this openhandle is from the same owner */
779 op_data->op_handle = old_handle;
781 it.it_flags = fmode | open_flags;
782 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
783 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
784 &ll_md_blocking_lease_ast,
785 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
786 * it can be cancelled which may mislead applications that the lease is
788 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
789 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
790 * doesn't deal with openhandle, so normal openhandle will be leaked.
792 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
793 ll_finish_md_op_data(op_data);
794 ptlrpc_req_finished(req);
798 if (it_disposition(&it, DISP_LOOKUP_NEG)) {
803 rc = it_open_error(DISP_OPEN_OPEN, &it);
807 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
808 ll_och_fill(sbi->ll_md_exp, &it, och);
810 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ {
815 /* already get lease, handle lease lock */
816 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
817 if (it.it_lock_mode == 0 ||
818 it.it_lock_bits != MDS_INODELOCK_OPEN) {
819 /* open lock must return for lease */
820 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
821 PFID(ll_inode2fid(inode)), it.it_lock_mode,
827 ll_intent_release(&it);
831 /* Cancel open lock */
832 if (it.it_lock_mode != 0) {
833 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
836 och->och_lease_handle.cookie = 0ULL;
838 rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, och, inode, 0, NULL);
840 CERROR("%s: error closing file "DFID": %d\n",
841 ll_get_fsname(inode->i_sb, NULL, 0),
842 PFID(&ll_i2info(inode)->lli_fid), rc2);
843 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
845 ll_intent_release(&it);
852 * Check whether a layout swap can be done between two inodes.
854 * \param[in] inode1 First inode to check
855 * \param[in] inode2 Second inode to check
857 * \retval 0 on success, layout swap can be performed between both inodes
858 * \retval negative error code if requirements are not met
860 static int ll_check_swap_layouts_validity(struct inode *inode1,
861 struct inode *inode2)
863 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
866 if (inode_permission(inode1, MAY_WRITE) ||
867 inode_permission(inode2, MAY_WRITE))
870 if (inode1->i_sb != inode2->i_sb)
876 static int ll_swap_layouts_close(struct obd_client_handle *och,
877 struct inode *inode, struct inode *inode2)
879 const struct lu_fid *fid1 = ll_inode2fid(inode);
880 const struct lu_fid *fid2;
883 CDEBUG(D_INODE, "%s: biased close of file " DFID "\n",
884 ll_get_fsname(inode->i_sb, NULL, 0), PFID(fid1));
886 rc = ll_check_swap_layouts_validity(inode, inode2);
890 /* We now know that inode2 is a lustre inode */
891 fid2 = ll_inode2fid(inode2);
893 rc = lu_fid_cmp(fid1, fid2);
900 * Close the file and swap layouts between inode & inode2.
901 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
902 * because we still need it to pack l_remote_handle to MDT.
904 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, och, inode,
905 MDS_CLOSE_LAYOUT_SWAP, inode2);
907 och = NULL; /* freed in ll_close_inode_openhandle() */
915 * Release lease and close the file.
916 * It will check if the lease has ever broken.
918 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
921 struct ldlm_lock *lock;
922 bool cancelled = true;
924 lock = ldlm_handle2lock(&och->och_lease_handle);
926 lock_res_and_lock(lock);
927 cancelled = ldlm_is_cancel(lock);
928 unlock_res_and_lock(lock);
932 CDEBUG(D_INODE, "lease for " DFID " broken? %d\n",
933 PFID(&ll_i2info(inode)->lli_fid), cancelled);
936 ldlm_cli_cancel(&och->och_lease_handle, 0);
938 *lease_broken = cancelled;
940 return ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
941 och, inode, 0, NULL);
944 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
946 struct ll_inode_info *lli = ll_i2info(inode);
947 struct cl_object *obj = lli->lli_clob;
948 struct cl_attr *attr = vvp_env_thread_attr(env);
954 ll_inode_size_lock(inode);
956 /* merge timestamps the most recently obtained from mds with
957 * timestamps obtained from osts
959 LTIME_S(inode->i_atime) = lli->lli_atime;
960 LTIME_S(inode->i_mtime) = lli->lli_mtime;
961 LTIME_S(inode->i_ctime) = lli->lli_ctime;
963 mtime = LTIME_S(inode->i_mtime);
964 atime = LTIME_S(inode->i_atime);
965 ctime = LTIME_S(inode->i_ctime);
967 cl_object_attr_lock(obj);
968 rc = cl_object_attr_get(env, obj, attr);
969 cl_object_attr_unlock(obj);
972 goto out_size_unlock;
974 if (atime < attr->cat_atime)
975 atime = attr->cat_atime;
977 if (ctime < attr->cat_ctime)
978 ctime = attr->cat_ctime;
980 if (mtime < attr->cat_mtime)
981 mtime = attr->cat_mtime;
983 CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n",
984 PFID(&lli->lli_fid), attr->cat_size);
986 i_size_write(inode, attr->cat_size);
988 inode->i_blocks = attr->cat_blocks;
990 LTIME_S(inode->i_mtime) = mtime;
991 LTIME_S(inode->i_atime) = atime;
992 LTIME_S(inode->i_ctime) = ctime;
995 ll_inode_size_unlock(inode);
1000 static bool file_is_noatime(const struct file *file)
1002 const struct vfsmount *mnt = file->f_path.mnt;
1003 const struct inode *inode = file_inode(file);
1005 /* Adapted from file_accessed() and touch_atime().*/
1006 if (file->f_flags & O_NOATIME)
1009 if (inode->i_flags & S_NOATIME)
1012 if (IS_NOATIME(inode))
1015 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1018 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1021 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1027 static void ll_io_init(struct cl_io *io, const struct file *file, int write)
1029 struct inode *inode = file_inode(file);
1031 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1033 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1034 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1035 file->f_flags & O_DIRECT ||
1038 io->ci_obj = ll_i2info(inode)->lli_clob;
1039 io->ci_lockreq = CILR_MAYBE;
1040 if (ll_file_nolock(file)) {
1041 io->ci_lockreq = CILR_NEVER;
1042 io->ci_no_srvlock = 1;
1043 } else if (file->f_flags & O_APPEND) {
1044 io->ci_lockreq = CILR_MANDATORY;
1047 io->ci_noatime = file_is_noatime(file);
1051 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1052 struct file *file, enum cl_io_type iot,
1053 loff_t *ppos, size_t count)
1055 struct ll_inode_info *lli = ll_i2info(file_inode(file));
1056 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1057 struct vvp_io *vio = vvp_env_io(env);
1058 struct range_lock range;
1063 CDEBUG(D_VFSTRACE, "file: %pD, type: %d ppos: %llu, count: %zu\n",
1064 file, iot, *ppos, count);
1067 io = vvp_env_thread_io(env);
1068 ll_io_init(io, file, iot == CIT_WRITE);
1070 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1071 struct vvp_io *vio = vvp_env_io(env);
1072 bool range_locked = false;
1074 if (file->f_flags & O_APPEND)
1075 range_lock_init(&range, 0, LUSTRE_EOF);
1077 range_lock_init(&range, *ppos, *ppos + count - 1);
1079 vio->vui_fd = LUSTRE_FPRIVATE(file);
1080 vio->vui_iter = args->u.normal.via_iter;
1081 vio->vui_iocb = args->u.normal.via_iocb;
1083 * Direct IO reads must also take range lock,
1084 * or multiple reads will try to work on the same pages
1085 * See LU-6227 for details.
1087 if (((iot == CIT_WRITE) ||
1088 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1089 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1090 CDEBUG(D_VFSTRACE, "Range lock [%llu, %llu]\n",
1091 range.rl_node.in_extent.start,
1092 range.rl_node.in_extent.end);
1093 rc = range_lock(&lli->lli_write_tree, &range);
1097 range_locked = true;
1099 ll_cl_add(file, env, io);
1100 rc = cl_io_loop(env, io);
1101 ll_cl_remove(file, env);
1103 CDEBUG(D_VFSTRACE, "Range unlock [%llu, %llu]\n",
1104 range.rl_node.in_extent.start,
1105 range.rl_node.in_extent.end);
1106 range_unlock(&lli->lli_write_tree, &range);
1109 /* cl_io_rw_init() handled IO */
1113 if (io->ci_nob > 0) {
1114 result = io->ci_nob;
1115 count -= io->ci_nob;
1116 *ppos = io->u.ci_wr.wr.crw_pos;
1118 /* prepare IO restart */
1120 args->u.normal.via_iter = vio->vui_iter;
1123 cl_io_fini(env, io);
1125 if ((!rc || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
1126 CDEBUG(D_VFSTRACE, "%s: restart %s from %lld, count:%zu, result: %zd\n",
1127 file_dentry(file)->d_name.name,
1128 iot == CIT_READ ? "read" : "write",
1129 *ppos, count, result);
1133 if (iot == CIT_READ) {
1135 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1136 LPROC_LL_READ_BYTES, result);
1137 } else if (iot == CIT_WRITE) {
1139 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1140 LPROC_LL_WRITE_BYTES, result);
1141 fd->fd_write_failed = false;
1142 } else if (!result && !rc) {
1145 fd->fd_write_failed = true;
1147 fd->fd_write_failed = false;
1148 } else if (rc != -ERESTARTSYS) {
1149 fd->fd_write_failed = true;
1152 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1154 return result > 0 ? result : rc;
1157 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1160 struct vvp_io_args *args;
1164 env = cl_env_get(&refcheck);
1166 return PTR_ERR(env);
1168 args = ll_env_args(env);
1169 args->u.normal.via_iter = to;
1170 args->u.normal.via_iocb = iocb;
1172 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1173 &iocb->ki_pos, iov_iter_count(to));
1174 cl_env_put(env, &refcheck);
1179 * Write to a file (through the page cache).
1181 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1184 struct vvp_io_args *args;
1188 env = cl_env_get(&refcheck);
1190 return PTR_ERR(env);
1192 args = ll_env_args(env);
1193 args->u.normal.via_iter = from;
1194 args->u.normal.via_iocb = iocb;
1196 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1197 &iocb->ki_pos, iov_iter_count(from));
1198 cl_env_put(env, &refcheck);
1202 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1203 __u64 flags, struct lov_user_md *lum,
1206 struct lookup_intent oit = {
1208 .it_flags = flags | MDS_OPEN_BY_FID,
1212 ll_inode_size_lock(inode);
1213 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
1217 ll_release_openhandle(inode, &oit);
1220 ll_inode_size_unlock(inode);
1221 ll_intent_release(&oit);
1225 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1226 struct lov_mds_md **lmmp, int *lmm_size,
1227 struct ptlrpc_request **request)
1229 struct ll_sb_info *sbi = ll_i2sbi(inode);
1230 struct mdt_body *body;
1231 struct lov_mds_md *lmm = NULL;
1232 struct ptlrpc_request *req = NULL;
1233 struct md_op_data *op_data;
1236 rc = ll_get_default_mdsize(sbi, &lmmsize);
1240 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1241 strlen(filename), lmmsize,
1242 LUSTRE_OPC_ANY, NULL);
1243 if (IS_ERR(op_data))
1244 return PTR_ERR(op_data);
1246 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1247 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1248 ll_finish_md_op_data(op_data);
1250 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
1255 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1257 lmmsize = body->mbo_eadatasize;
1259 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1265 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1267 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1268 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1274 * This is coming from the MDS, so is probably in
1275 * little endian. We convert it to host endian before
1276 * passing it to userspace.
1278 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
1281 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1282 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1285 /* if function called for directory - we should
1286 * avoid swab not existent lsm objects
1288 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1289 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1290 if (S_ISREG(body->mbo_mode))
1291 lustre_swab_lov_user_md_objects(
1292 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1294 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1295 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1296 if (S_ISREG(body->mbo_mode))
1297 lustre_swab_lov_user_md_objects(
1298 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1305 *lmm_size = lmmsize;
1310 static int ll_lov_setea(struct inode *inode, struct file *file,
1313 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1314 struct lov_user_md *lump;
1315 int lum_size = sizeof(struct lov_user_md) +
1316 sizeof(struct lov_user_ost_data);
1319 if (!capable(CFS_CAP_SYS_ADMIN))
1322 lump = libcfs_kvzalloc(lum_size, GFP_NOFS);
1326 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) {
1331 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump,
1333 cl_lov_delay_create_clear(&file->f_flags);
1339 static int ll_file_getstripe(struct inode *inode,
1340 struct lov_user_md __user *lum)
1346 env = cl_env_get(&refcheck);
1348 return PTR_ERR(env);
1350 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum);
1351 cl_env_put(env, &refcheck);
1355 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1358 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
1359 struct lov_user_md *klum;
1361 __u64 flags = FMODE_WRITE;
1363 rc = ll_copy_user_md(lum, &klum);
1368 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, klum,
1370 cl_lov_delay_create_clear(&file->f_flags);
1374 put_user(0, &lum->lmm_stripe_count);
1376 ll_layout_refresh(inode, &gen);
1377 rc = ll_file_getstripe(inode, (struct lov_user_md __user *)arg);
1385 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1387 struct ll_inode_info *lli = ll_i2info(inode);
1388 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1389 struct ll_grouplock grouplock;
1393 CWARN("group id for group lock must not be 0\n");
1397 if (ll_file_nolock(file))
1400 spin_lock(&lli->lli_lock);
1401 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1402 CWARN("group lock already existed with gid %lu\n",
1403 fd->fd_grouplock.lg_gid);
1404 spin_unlock(&lli->lli_lock);
1407 LASSERT(!fd->fd_grouplock.lg_lock);
1408 spin_unlock(&lli->lli_lock);
1410 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1411 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1415 spin_lock(&lli->lli_lock);
1416 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1417 spin_unlock(&lli->lli_lock);
1418 CERROR("another thread just won the race\n");
1419 cl_put_grouplock(&grouplock);
1423 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1424 fd->fd_grouplock = grouplock;
1425 spin_unlock(&lli->lli_lock);
1427 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1431 static int ll_put_grouplock(struct inode *inode, struct file *file,
1434 struct ll_inode_info *lli = ll_i2info(inode);
1435 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1436 struct ll_grouplock grouplock;
1438 spin_lock(&lli->lli_lock);
1439 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1440 spin_unlock(&lli->lli_lock);
1441 CWARN("no group lock held\n");
1444 LASSERT(fd->fd_grouplock.lg_lock);
1446 if (fd->fd_grouplock.lg_gid != arg) {
1447 CWARN("group lock %lu doesn't match current id %lu\n",
1448 arg, fd->fd_grouplock.lg_gid);
1449 spin_unlock(&lli->lli_lock);
1453 grouplock = fd->fd_grouplock;
1454 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1455 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1456 spin_unlock(&lli->lli_lock);
1458 cl_put_grouplock(&grouplock);
1459 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1464 * Close inode open handle
1466 * \param inode [in] inode in question
1467 * \param it [in,out] intent which contains open info and result
1470 * \retval <0 failure
1472 int ll_release_openhandle(struct inode *inode, struct lookup_intent *it)
1474 struct obd_client_handle *och;
1479 /* Root ? Do nothing. */
1480 if (is_root_inode(inode))
1483 /* No open handle to close? Move away */
1484 if (!it_disposition(it, DISP_OPEN_OPEN))
1487 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1489 och = kzalloc(sizeof(*och), GFP_NOFS);
1495 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1497 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1498 och, inode, 0, NULL);
1500 /* this one is in place of ll_file_open */
1501 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1502 ptlrpc_req_finished(it->it_request);
1503 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1509 * Get size for inode for which FIEMAP mapping is requested.
1510 * Make the FIEMAP get_info call and returns the result.
1512 * \param fiemap kernel buffer to hold extens
1513 * \param num_bytes kernel buffer size
1515 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
1518 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
1523 /* Checks for fiemap flags */
1524 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1525 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1529 /* Check for FIEMAP_FLAG_SYNC */
1530 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1531 rc = filemap_fdatawrite(inode->i_mapping);
1536 env = cl_env_get(&refcheck);
1538 return PTR_ERR(env);
1540 if (i_size_read(inode) == 0) {
1541 rc = ll_glimpse_size(inode);
1546 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1547 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
1548 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
1550 /* If filesize is 0, then there would be no objects for mapping */
1551 if (fmkey.lfik_oa.o_size == 0) {
1552 fiemap->fm_mapped_extents = 0;
1557 memcpy(&fmkey.lfik_fiemap, fiemap, sizeof(*fiemap));
1559 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
1560 &fmkey, fiemap, &num_bytes);
1562 cl_env_put(env, &refcheck);
1566 int ll_fid2path(struct inode *inode, void __user *arg)
1568 struct obd_export *exp = ll_i2mdexp(inode);
1569 const struct getinfo_fid2path __user *gfin = arg;
1570 struct getinfo_fid2path *gfout;
1575 if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
1576 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1579 /* Only need to get the buflen */
1580 if (get_user(pathlen, &gfin->gf_pathlen))
1583 if (pathlen > PATH_MAX)
1586 outsize = sizeof(*gfout) + pathlen;
1588 gfout = kzalloc(outsize, GFP_NOFS);
1592 if (copy_from_user(gfout, arg, sizeof(*gfout))) {
1597 /* Call mdc_iocontrol */
1598 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1602 if (copy_to_user(arg, gfout, outsize))
1611 * Read the data_version for inode.
1613 * This value is computed using stripe object version on OST.
1614 * Version is computed using server side locking.
1616 * @param flags if do sync on the OST side;
1618 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1619 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1621 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1623 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1629 /* If no file object initialized, we consider its version is 0. */
1635 env = cl_env_get(&refcheck);
1637 return PTR_ERR(env);
1639 io = vvp_env_thread_io(env);
1641 io->u.ci_data_version.dv_data_version = 0;
1642 io->u.ci_data_version.dv_flags = flags;
1645 if (!cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj))
1646 result = cl_io_loop(env, io);
1648 result = io->ci_result;
1650 *data_version = io->u.ci_data_version.dv_data_version;
1652 cl_io_fini(env, io);
1654 if (unlikely(io->ci_need_restart))
1657 cl_env_put(env, &refcheck);
1663 * Trigger a HSM release request for the provided inode.
1665 int ll_hsm_release(struct inode *inode)
1668 struct obd_client_handle *och = NULL;
1669 __u64 data_version = 0;
1673 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
1674 ll_get_fsname(inode->i_sb, NULL, 0),
1675 PFID(&ll_i2info(inode)->lli_fid));
1677 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
1683 /* Grab latest data_version and [am]time values */
1684 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
1688 env = cl_env_get(&refcheck);
1694 ll_merge_attr(env, inode);
1695 cl_env_put(env, &refcheck);
1697 /* Release the file.
1698 * NB: lease lock handle is released in mdc_hsm_release_pack() because
1699 * we still need it to pack l_remote_handle to MDT.
1701 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, och, inode,
1702 MDS_HSM_RELEASE, &data_version);
1706 if (och && !IS_ERR(och)) /* close the file */
1707 ll_lease_close(och, inode, NULL);
1712 struct ll_swap_stack {
1715 struct inode *inode1;
1716 struct inode *inode2;
1721 static int ll_swap_layouts(struct file *file1, struct file *file2,
1722 struct lustre_swap_layouts *lsl)
1724 struct mdc_swap_layouts msl;
1725 struct md_op_data *op_data;
1728 struct ll_swap_stack *llss = NULL;
1731 llss = kzalloc(sizeof(*llss), GFP_NOFS);
1735 llss->inode1 = file_inode(file1);
1736 llss->inode2 = file_inode(file2);
1738 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
1742 /* we use 2 bool because it is easier to swap than 2 bits */
1743 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
1744 llss->check_dv1 = true;
1746 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
1747 llss->check_dv2 = true;
1749 /* we cannot use lsl->sl_dvX directly because we may swap them */
1750 llss->dv1 = lsl->sl_dv1;
1751 llss->dv2 = lsl->sl_dv2;
1753 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
1754 if (!rc) /* same file, done! */
1757 if (rc < 0) { /* sequentialize it */
1758 swap(llss->inode1, llss->inode2);
1760 swap(llss->dv1, llss->dv2);
1761 swap(llss->check_dv1, llss->check_dv2);
1765 if (gid != 0) { /* application asks to flush dirty cache */
1766 rc = ll_get_grouplock(llss->inode1, file1, gid);
1770 rc = ll_get_grouplock(llss->inode2, file2, gid);
1772 ll_put_grouplock(llss->inode1, file1, gid);
1777 /* ultimate check, before swapping the layouts we check if
1778 * dataversion has changed (if requested)
1780 if (llss->check_dv1) {
1781 rc = ll_data_version(llss->inode1, &dv, 0);
1784 if (dv != llss->dv1) {
1790 if (llss->check_dv2) {
1791 rc = ll_data_version(llss->inode2, &dv, 0);
1794 if (dv != llss->dv2) {
1800 /* struct md_op_data is used to send the swap args to the mdt
1801 * only flags is missing, so we use struct mdc_swap_layouts
1802 * through the md_op_data->op_data
1804 /* flags from user space have to be converted before they are send to
1805 * server, no flag is sent today, they are only used on the client
1809 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
1810 0, LUSTRE_OPC_ANY, &msl);
1811 if (IS_ERR(op_data)) {
1812 rc = PTR_ERR(op_data);
1816 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
1817 sizeof(*op_data), op_data, NULL);
1818 ll_finish_md_op_data(op_data);
1822 ll_put_grouplock(llss->inode2, file2, gid);
1823 ll_put_grouplock(llss->inode1, file1, gid);
1832 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
1834 struct md_op_data *op_data;
1837 /* Detect out-of range masks */
1838 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
1841 /* Non-root users are forbidden to set or clear flags which are
1842 * NOT defined in HSM_USER_MASK.
1844 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
1845 !capable(CFS_CAP_SYS_ADMIN))
1848 /* Detect out-of range archive id */
1849 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
1850 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
1853 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1854 LUSTRE_OPC_ANY, hss);
1855 if (IS_ERR(op_data))
1856 return PTR_ERR(op_data);
1858 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
1859 sizeof(*op_data), op_data, NULL);
1861 ll_finish_md_op_data(op_data);
1866 static int ll_hsm_import(struct inode *inode, struct file *file,
1867 struct hsm_user_import *hui)
1869 struct hsm_state_set *hss = NULL;
1870 struct iattr *attr = NULL;
1873 if (!S_ISREG(inode->i_mode))
1877 hss = kzalloc(sizeof(*hss), GFP_NOFS);
1881 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
1882 hss->hss_archive_id = hui->hui_archive_id;
1883 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
1884 rc = ll_hsm_state_set(inode, hss);
1888 attr = kzalloc(sizeof(*attr), GFP_NOFS);
1894 attr->ia_mode = hui->hui_mode & 0777;
1895 attr->ia_mode |= S_IFREG;
1896 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
1897 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
1898 attr->ia_size = hui->hui_size;
1899 attr->ia_mtime.tv_sec = hui->hui_mtime;
1900 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
1901 attr->ia_atime.tv_sec = hui->hui_atime;
1902 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
1904 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
1905 ATTR_UID | ATTR_GID |
1906 ATTR_MTIME | ATTR_MTIME_SET |
1907 ATTR_ATIME | ATTR_ATIME_SET;
1911 rc = ll_setattr_raw(file->f_path.dentry, attr, true);
1915 inode_unlock(inode);
1923 static inline long ll_lease_type_from_fmode(fmode_t fmode)
1925 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
1926 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
1930 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1932 struct inode *inode = file_inode(file);
1933 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1936 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),cmd=%x\n",
1937 PFID(ll_inode2fid(inode)), inode, cmd);
1938 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1940 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1941 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1945 case LL_IOC_GETFLAGS:
1946 /* Get the current value of the file flags */
1947 return put_user(fd->fd_flags, (int __user *)arg);
1948 case LL_IOC_SETFLAGS:
1949 case LL_IOC_CLRFLAGS:
1950 /* Set or clear specific file flags */
1951 /* XXX This probably needs checks to ensure the flags are
1952 * not abused, and to handle any flag side effects.
1954 if (get_user(flags, (int __user *)arg))
1957 if (cmd == LL_IOC_SETFLAGS) {
1958 if ((flags & LL_FILE_IGNORE_LOCK) &&
1959 !(file->f_flags & O_DIRECT)) {
1960 CERROR("%s: unable to disable locking on non-O_DIRECT file\n",
1965 fd->fd_flags |= flags;
1967 fd->fd_flags &= ~flags;
1970 case LL_IOC_LOV_SETSTRIPE:
1971 return ll_lov_setstripe(inode, file, arg);
1972 case LL_IOC_LOV_SETEA:
1973 return ll_lov_setea(inode, file, arg);
1974 case LL_IOC_LOV_SWAP_LAYOUTS: {
1976 struct lustre_swap_layouts lsl;
1978 if (copy_from_user(&lsl, (char __user *)arg,
1979 sizeof(struct lustre_swap_layouts)))
1982 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
1985 file2 = fget(lsl.sl_fd);
1989 /* O_WRONLY or O_RDWR */
1990 if ((file2->f_flags & O_ACCMODE) == O_RDONLY) {
1995 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
1996 struct obd_client_handle *och = NULL;
1997 struct ll_inode_info *lli;
1998 struct inode *inode2;
2000 if (lsl.sl_flags != SWAP_LAYOUTS_CLOSE) {
2005 lli = ll_i2info(inode);
2006 mutex_lock(&lli->lli_och_mutex);
2007 if (fd->fd_lease_och) {
2008 och = fd->fd_lease_och;
2009 fd->fd_lease_och = NULL;
2011 mutex_unlock(&lli->lli_och_mutex);
2016 inode2 = file_inode(file2);
2017 rc = ll_swap_layouts_close(och, inode, inode2);
2019 rc = ll_swap_layouts(file, file2, &lsl);
2025 case LL_IOC_LOV_GETSTRIPE:
2026 return ll_file_getstripe(inode,
2027 (struct lov_user_md __user *)arg);
2028 case FSFILT_IOC_GETFLAGS:
2029 case FSFILT_IOC_SETFLAGS:
2030 return ll_iocontrol(inode, file, cmd, arg);
2031 case FSFILT_IOC_GETVERSION_OLD:
2032 case FSFILT_IOC_GETVERSION:
2033 return put_user(inode->i_generation, (int __user *)arg);
2034 case LL_IOC_GROUP_LOCK:
2035 return ll_get_grouplock(inode, file, arg);
2036 case LL_IOC_GROUP_UNLOCK:
2037 return ll_put_grouplock(inode, file, arg);
2038 case IOC_OBD_STATFS:
2039 return ll_obd_statfs(inode, (void __user *)arg);
2041 /* We need to special case any other ioctls we want to handle,
2042 * to send them to the MDS/OST as appropriate and to properly
2043 * network encode the arg field.
2044 case FSFILT_IOC_SETVERSION_OLD:
2045 case FSFILT_IOC_SETVERSION:
2047 case LL_IOC_FLUSHCTX:
2048 return ll_flush_ctx(inode);
2049 case LL_IOC_PATH2FID: {
2050 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2051 sizeof(struct lu_fid)))
2056 case LL_IOC_GETPARENT:
2057 return ll_getparent(file, (struct getparent __user *)arg);
2058 case OBD_IOC_FID2PATH:
2059 return ll_fid2path(inode, (void __user *)arg);
2060 case LL_IOC_DATA_VERSION: {
2061 struct ioc_data_version idv;
2064 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2067 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2068 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2069 if (rc == 0 && copy_to_user((char __user *)arg, &idv,
2076 case LL_IOC_GET_MDTIDX: {
2079 mdtidx = ll_get_mdt_idx(inode);
2083 if (put_user(mdtidx, (int __user *)arg))
2088 case OBD_IOC_GETDTNAME:
2089 case OBD_IOC_GETMDNAME:
2090 return ll_get_obd_name(inode, cmd, arg);
2091 case LL_IOC_HSM_STATE_GET: {
2092 struct md_op_data *op_data;
2093 struct hsm_user_state *hus;
2096 hus = kzalloc(sizeof(*hus), GFP_NOFS);
2100 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2101 LUSTRE_OPC_ANY, hus);
2102 if (IS_ERR(op_data)) {
2104 return PTR_ERR(op_data);
2107 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2110 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2113 ll_finish_md_op_data(op_data);
2117 case LL_IOC_HSM_STATE_SET: {
2118 struct hsm_state_set *hss;
2121 hss = memdup_user((char __user *)arg, sizeof(*hss));
2123 return PTR_ERR(hss);
2125 rc = ll_hsm_state_set(inode, hss);
2130 case LL_IOC_HSM_ACTION: {
2131 struct md_op_data *op_data;
2132 struct hsm_current_action *hca;
2135 hca = kzalloc(sizeof(*hca), GFP_NOFS);
2139 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2140 LUSTRE_OPC_ANY, hca);
2141 if (IS_ERR(op_data)) {
2143 return PTR_ERR(op_data);
2146 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2149 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2152 ll_finish_md_op_data(op_data);
2156 case LL_IOC_SET_LEASE: {
2157 struct ll_inode_info *lli = ll_i2info(inode);
2158 struct obd_client_handle *och = NULL;
2163 case LL_LEASE_WRLCK:
2164 if (!(file->f_mode & FMODE_WRITE))
2166 fmode = FMODE_WRITE;
2168 case LL_LEASE_RDLCK:
2169 if (!(file->f_mode & FMODE_READ))
2173 case LL_LEASE_UNLCK:
2174 mutex_lock(&lli->lli_och_mutex);
2175 if (fd->fd_lease_och) {
2176 och = fd->fd_lease_och;
2177 fd->fd_lease_och = NULL;
2179 mutex_unlock(&lli->lli_och_mutex);
2184 fmode = och->och_flags;
2185 rc = ll_lease_close(och, inode, &lease_broken);
2192 return ll_lease_type_from_fmode(fmode);
2197 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
2199 /* apply for lease */
2200 och = ll_lease_open(inode, file, fmode, 0);
2202 return PTR_ERR(och);
2205 mutex_lock(&lli->lli_och_mutex);
2206 if (!fd->fd_lease_och) {
2207 fd->fd_lease_och = och;
2210 mutex_unlock(&lli->lli_och_mutex);
2212 /* impossible now that only excl is supported for now */
2213 ll_lease_close(och, inode, &lease_broken);
2218 case LL_IOC_GET_LEASE: {
2219 struct ll_inode_info *lli = ll_i2info(inode);
2220 struct ldlm_lock *lock = NULL;
2223 mutex_lock(&lli->lli_och_mutex);
2224 if (fd->fd_lease_och) {
2225 struct obd_client_handle *och = fd->fd_lease_och;
2227 lock = ldlm_handle2lock(&och->och_lease_handle);
2229 lock_res_and_lock(lock);
2230 if (!ldlm_is_cancel(lock))
2231 fmode = och->och_flags;
2232 unlock_res_and_lock(lock);
2233 LDLM_LOCK_PUT(lock);
2236 mutex_unlock(&lli->lli_och_mutex);
2237 return ll_lease_type_from_fmode(fmode);
2239 case LL_IOC_HSM_IMPORT: {
2240 struct hsm_user_import *hui;
2242 hui = memdup_user((void __user *)arg, sizeof(*hui));
2244 return PTR_ERR(hui);
2246 rc = ll_hsm_import(inode, file, hui);
2254 if (ll_iocontrol_call(inode, file, cmd, arg, &err) ==
2258 return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2259 (void __user *)arg);
2264 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2266 struct inode *inode = file_inode(file);
2267 loff_t retval, eof = 0;
2269 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2270 (origin == SEEK_CUR) ? file->f_pos : 0);
2271 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2272 PFID(ll_inode2fid(inode)), inode, retval, retval, origin);
2273 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2275 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2276 retval = ll_glimpse_size(inode);
2279 eof = i_size_read(inode);
2282 return generic_file_llseek_size(file, offset, origin,
2283 ll_file_maxbytes(inode), eof);
2286 static int ll_flush(struct file *file, fl_owner_t id)
2288 struct inode *inode = file_inode(file);
2289 struct ll_inode_info *lli = ll_i2info(inode);
2290 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2293 LASSERT(!S_ISDIR(inode->i_mode));
2295 /* catch async errors that were recorded back when async writeback
2296 * failed for pages in this mapping.
2298 rc = lli->lli_async_rc;
2299 lli->lli_async_rc = 0;
2300 if (lli->lli_clob) {
2301 err = lov_read_and_clear_async_rc(lli->lli_clob);
2306 /* The application has been told about write failure already.
2307 * Do not report failure again.
2309 if (fd->fd_write_failed)
2311 return rc ? -EIO : 0;
2315 * Called to make sure a portion of file has been written out.
2316 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2318 * Return how many pages have been written.
2320 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2321 enum cl_fsync_mode mode, int ignore_layout)
2325 struct cl_fsync_io *fio;
2329 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2330 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2333 env = cl_env_get(&refcheck);
2335 return PTR_ERR(env);
2337 io = vvp_env_thread_io(env);
2338 io->ci_obj = ll_i2info(inode)->lli_clob;
2339 io->ci_ignore_layout = ignore_layout;
2341 /* initialize parameters for sync */
2342 fio = &io->u.ci_fsync;
2343 fio->fi_start = start;
2345 fio->fi_fid = ll_inode2fid(inode);
2346 fio->fi_mode = mode;
2347 fio->fi_nr_written = 0;
2349 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2350 result = cl_io_loop(env, io);
2352 result = io->ci_result;
2354 result = fio->fi_nr_written;
2355 cl_io_fini(env, io);
2356 cl_env_put(env, &refcheck);
2361 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2363 struct inode *inode = file_inode(file);
2364 struct ll_inode_info *lli = ll_i2info(inode);
2365 struct ptlrpc_request *req;
2368 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2369 PFID(ll_inode2fid(inode)), inode);
2370 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2372 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2375 /* catch async errors that were recorded back when async writeback
2376 * failed for pages in this mapping.
2378 if (!S_ISDIR(inode->i_mode)) {
2379 err = lli->lli_async_rc;
2380 lli->lli_async_rc = 0;
2383 if (lli->lli_clob) {
2384 err = lov_read_and_clear_async_rc(lli->lli_clob);
2390 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
2394 ptlrpc_req_finished(req);
2396 if (S_ISREG(inode->i_mode)) {
2397 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2399 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2400 if (rc == 0 && err < 0)
2403 fd->fd_write_failed = true;
2405 fd->fd_write_failed = false;
2408 inode_unlock(inode);
2413 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2415 struct inode *inode = file_inode(file);
2416 struct ll_sb_info *sbi = ll_i2sbi(inode);
2417 struct ldlm_enqueue_info einfo = {
2418 .ei_type = LDLM_FLOCK,
2419 .ei_cb_cp = ldlm_flock_completion_ast,
2420 .ei_cbdata = file_lock,
2422 struct md_op_data *op_data;
2423 struct lustre_handle lockh = {0};
2424 union ldlm_policy_data flock = { { 0 } };
2425 int fl_type = file_lock->fl_type;
2430 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2431 PFID(ll_inode2fid(inode)), file_lock);
2433 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2435 if (file_lock->fl_flags & FL_FLOCK)
2436 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2437 else if (!(file_lock->fl_flags & FL_POSIX))
2440 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2441 flock.l_flock.pid = file_lock->fl_pid;
2442 flock.l_flock.start = file_lock->fl_start;
2443 flock.l_flock.end = file_lock->fl_end;
2445 /* Somewhat ugly workaround for svc lockd.
2446 * lockd installs custom fl_lmops->lm_compare_owner that checks
2447 * for the fl_owner to be the same (which it always is on local node
2448 * I guess between lockd processes) and then compares pid.
2449 * As such we assign pid to the owner field to make it all work,
2450 * conflict with normal locks is unlikely since pid space and
2451 * pointer space for current->files are not intersecting
2453 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2454 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2458 einfo.ei_mode = LCK_PR;
2461 /* An unlock request may or may not have any relation to
2462 * existing locks so we may not be able to pass a lock handle
2463 * via a normal ldlm_lock_cancel() request. The request may even
2464 * unlock a byte range in the middle of an existing lock. In
2465 * order to process an unlock request we need all of the same
2466 * information that is given with a normal read or write record
2467 * lock request. To avoid creating another ldlm unlock (cancel)
2468 * message we'll treat a LCK_NL flock request as an unlock.
2470 einfo.ei_mode = LCK_NL;
2473 einfo.ei_mode = LCK_PW;
2476 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
2491 flags = LDLM_FL_BLOCK_NOWAIT;
2497 flags = LDLM_FL_TEST_LOCK;
2500 CERROR("unknown fcntl lock command: %d\n", cmd);
2505 * Save the old mode so that if the mode in the lock changes we
2506 * can decrement the appropriate reader or writer refcount.
2508 file_lock->fl_type = einfo.ei_mode;
2510 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2511 LUSTRE_OPC_ANY, NULL);
2512 if (IS_ERR(op_data))
2513 return PTR_ERR(op_data);
2515 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n",
2516 PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags,
2517 einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2519 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
2522 /* Restore the file lock type if not TEST lock. */
2523 if (!(flags & LDLM_FL_TEST_LOCK))
2524 file_lock->fl_type = fl_type;
2526 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
2527 !(flags & LDLM_FL_TEST_LOCK))
2528 rc2 = locks_lock_file_wait(file, file_lock);
2530 if (rc2 && file_lock->fl_type != F_UNLCK) {
2531 einfo.ei_mode = LCK_NL;
2532 md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
2537 ll_finish_md_op_data(op_data);
2542 int ll_get_fid_by_name(struct inode *parent, const char *name,
2543 int namelen, struct lu_fid *fid,
2544 struct inode **inode)
2546 struct md_op_data *op_data = NULL;
2547 struct ptlrpc_request *req;
2548 struct mdt_body *body;
2551 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
2552 LUSTRE_OPC_ANY, NULL);
2553 if (IS_ERR(op_data))
2554 return PTR_ERR(op_data);
2556 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
2557 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
2558 ll_finish_md_op_data(op_data);
2562 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2568 *fid = body->mbo_fid1;
2571 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
2573 ptlrpc_req_finished(req);
2577 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
2578 const char *name, int namelen)
2580 struct ptlrpc_request *request = NULL;
2581 struct obd_client_handle *och = NULL;
2582 struct inode *child_inode = NULL;
2583 struct dentry *dchild = NULL;
2584 struct md_op_data *op_data;
2585 struct mdt_body *body;
2586 u64 data_version = 0;
2590 CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%d\n",
2591 name, PFID(ll_inode2fid(parent)), mdtidx);
2593 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
2594 0, LUSTRE_OPC_ANY, NULL);
2595 if (IS_ERR(op_data))
2596 return PTR_ERR(op_data);
2598 /* Get child FID first */
2599 qstr.hash = full_name_hash(parent, name, namelen);
2602 dchild = d_lookup(file_dentry(file), &qstr);
2604 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2605 if (dchild->d_inode)
2606 child_inode = igrab(dchild->d_inode);
2611 rc = ll_get_fid_by_name(parent, name, namelen,
2612 &op_data->op_fid3, &child_inode);
2622 inode_lock(child_inode);
2623 op_data->op_fid3 = *ll_inode2fid(child_inode);
2624 if (!fid_is_sane(&op_data->op_fid3)) {
2625 CERROR("%s: migrate %s, but fid "DFID" is insane\n",
2626 ll_get_fsname(parent->i_sb, NULL, 0), name,
2627 PFID(&op_data->op_fid3));
2632 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
2637 CDEBUG(D_INFO, "%s:"DFID" is already on MDT%d.\n", name,
2638 PFID(&op_data->op_fid3), mdtidx);
2643 if (S_ISREG(child_inode->i_mode)) {
2644 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
2651 rc = ll_data_version(child_inode, &data_version,
2656 op_data->op_handle = och->och_fh;
2657 op_data->op_data = och->och_mod;
2658 op_data->op_data_version = data_version;
2659 op_data->op_lease_handle = och->och_lease_handle;
2660 op_data->op_bias |= MDS_RENAME_MIGRATE;
2663 op_data->op_mds = mdtidx;
2664 op_data->op_cli_flags = CLI_MIGRATE;
2665 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
2666 namelen, name, namelen, &request);
2668 ll_update_times(request, parent);
2671 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2678 * If the server does release layout lock, then we cleanup
2679 * the client och here, otherwise release it in out_close:
2681 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
2682 obd_mod_put(och->och_mod);
2683 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
2685 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
2690 ptlrpc_req_finished(request);
2692 /* Try again if the file layout has changed. */
2693 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode)) {
2698 if (och) /* close the file */
2699 ll_lease_close(och, child_inode, NULL);
2701 clear_nlink(child_inode);
2703 inode_unlock(child_inode);
2706 ll_finish_md_op_data(op_data);
2711 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2717 * test if some locks matching bits and l_req_mode are acquired
2718 * - bits can be in different locks
2719 * - if found clear the common lock bits in *bits
2720 * - the bits not found, are kept in *bits
2722 * \param bits [IN] searched lock bits [IN]
2723 * \param l_req_mode [IN] searched lock mode
2724 * \retval boolean, true iff all bits are found
2726 int ll_have_md_lock(struct inode *inode, __u64 *bits,
2727 enum ldlm_mode l_req_mode)
2729 struct lustre_handle lockh;
2730 union ldlm_policy_data policy;
2731 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
2732 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
2740 fid = &ll_i2info(inode)->lli_fid;
2741 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
2742 ldlm_lockname[mode]);
2744 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2745 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2746 policy.l_inodebits.bits = *bits & (1 << i);
2747 if (policy.l_inodebits.bits == 0)
2750 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2751 &policy, mode, &lockh)) {
2752 struct ldlm_lock *lock;
2754 lock = ldlm_handle2lock(&lockh);
2757 ~(lock->l_policy_data.l_inodebits.bits);
2758 LDLM_LOCK_PUT(lock);
2760 *bits &= ~policy.l_inodebits.bits;
2767 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
2768 struct lustre_handle *lockh, __u64 flags,
2769 enum ldlm_mode mode)
2771 union ldlm_policy_data policy = { .l_inodebits = { bits } };
2774 fid = &ll_i2info(inode)->lli_fid;
2775 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2777 return md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED,
2778 fid, LDLM_IBITS, &policy, mode, lockh);
2781 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2783 /* Already unlinked. Just update nlink and return success */
2784 if (rc == -ENOENT) {
2786 /* If it is striped directory, and there is bad stripe
2787 * Let's revalidate the dentry again, instead of returning
2790 if (S_ISDIR(inode->i_mode) && ll_i2info(inode)->lli_lsm_md)
2793 /* This path cannot be hit for regular files unless in
2794 * case of obscure races, so no need to validate size.
2796 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2798 } else if (rc != 0) {
2799 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
2800 "%s: revalidate FID "DFID" error: rc = %d\n",
2801 ll_get_fsname(inode->i_sb, NULL, 0),
2802 PFID(ll_inode2fid(inode)), rc);
2808 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2810 struct inode *inode = d_inode(dentry);
2811 struct ptlrpc_request *req = NULL;
2812 struct obd_export *exp;
2815 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%pd\n",
2816 PFID(ll_inode2fid(inode)), inode, dentry);
2818 exp = ll_i2mdexp(inode);
2820 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2821 * But under CMD case, it caused some lock issues, should be fixed
2822 * with new CMD ibits lock. See bug 12718
2824 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
2825 struct lookup_intent oit = { .it_op = IT_GETATTR };
2826 struct md_op_data *op_data;
2828 if (ibits == MDS_INODELOCK_LOOKUP)
2829 oit.it_op = IT_LOOKUP;
2831 /* Call getattr by fid, so do not provide name at all. */
2832 op_data = ll_prep_md_op_data(NULL, inode,
2834 LUSTRE_OPC_ANY, NULL);
2835 if (IS_ERR(op_data))
2836 return PTR_ERR(op_data);
2838 rc = md_intent_lock(exp, op_data, &oit, &req,
2839 &ll_md_blocking_ast, 0);
2840 ll_finish_md_op_data(op_data);
2842 rc = ll_inode_revalidate_fini(inode, rc);
2846 rc = ll_revalidate_it_finish(req, &oit, inode);
2848 ll_intent_release(&oit);
2852 /* Unlinked? Unhash dentry, so it is not picked up later by
2853 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2854 * here to preserve get_cwd functionality on 2.6.
2857 if (!d_inode(dentry)->i_nlink) {
2858 spin_lock(&inode->i_lock);
2859 d_lustre_invalidate(dentry, 0);
2860 spin_unlock(&inode->i_lock);
2863 ll_lookup_finish_locks(&oit, inode);
2864 } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) {
2865 struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry));
2866 u64 valid = OBD_MD_FLGETATTR;
2867 struct md_op_data *op_data;
2870 if (S_ISREG(inode->i_mode)) {
2871 rc = ll_get_default_mdsize(sbi, &ealen);
2874 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2877 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2878 0, ealen, LUSTRE_OPC_ANY,
2880 if (IS_ERR(op_data))
2881 return PTR_ERR(op_data);
2883 op_data->op_valid = valid;
2884 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2885 ll_finish_md_op_data(op_data);
2887 return ll_inode_revalidate_fini(inode, rc);
2889 rc = ll_prep_inode(&inode, req, NULL, NULL);
2892 ptlrpc_req_finished(req);
2896 static int ll_merge_md_attr(struct inode *inode)
2898 struct cl_attr attr = { 0 };
2901 LASSERT(ll_i2info(inode)->lli_lsm_md);
2902 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
2903 &attr, ll_md_blocking_ast);
2907 set_nlink(inode, attr.cat_nlink);
2908 inode->i_blocks = attr.cat_blocks;
2909 i_size_write(inode, attr.cat_size);
2911 ll_i2info(inode)->lli_atime = attr.cat_atime;
2912 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
2913 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
2918 static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2920 struct inode *inode = d_inode(dentry);
2923 rc = __ll_inode_revalidate(dentry, ibits);
2927 /* if object isn't regular file, don't validate size */
2928 if (!S_ISREG(inode->i_mode)) {
2929 if (S_ISDIR(inode->i_mode) &&
2930 ll_i2info(inode)->lli_lsm_md) {
2931 rc = ll_merge_md_attr(inode);
2936 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
2937 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
2938 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
2940 struct ll_inode_info *lli = ll_i2info(inode);
2942 /* In case of restore, the MDT has the right size and has
2943 * already send it back without granting the layout lock,
2944 * inode is up-to-date so glimpse is useless.
2945 * Also to glimpse we need the layout, in case of a running
2946 * restore the MDT holds the layout lock so the glimpse will
2947 * block up to the end of restore (getattr will block)
2949 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags))
2950 rc = ll_glimpse_size(inode);
2955 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
2957 struct inode *inode = d_inode(de);
2958 struct ll_sb_info *sbi = ll_i2sbi(inode);
2959 struct ll_inode_info *lli = ll_i2info(inode);
2962 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
2963 MDS_INODELOCK_LOOKUP);
2964 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
2969 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
2971 stat->dev = inode->i_sb->s_dev;
2972 if (ll_need_32bit_api(sbi))
2973 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
2975 stat->ino = inode->i_ino;
2976 stat->mode = inode->i_mode;
2977 stat->uid = inode->i_uid;
2978 stat->gid = inode->i_gid;
2979 stat->rdev = inode->i_rdev;
2980 stat->atime = inode->i_atime;
2981 stat->mtime = inode->i_mtime;
2982 stat->ctime = inode->i_ctime;
2983 stat->blksize = 1 << inode->i_blkbits;
2985 stat->nlink = inode->i_nlink;
2986 stat->size = i_size_read(inode);
2987 stat->blocks = inode->i_blocks;
2992 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2993 __u64 start, __u64 len)
2997 struct fiemap *fiemap;
2998 unsigned int extent_count = fieinfo->fi_extents_max;
3000 num_bytes = sizeof(*fiemap) + (extent_count *
3001 sizeof(struct fiemap_extent));
3002 fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS);
3006 fiemap->fm_flags = fieinfo->fi_flags;
3007 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3008 fiemap->fm_start = start;
3009 fiemap->fm_length = len;
3011 if (extent_count > 0 &&
3012 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3013 sizeof(struct fiemap_extent))) {
3018 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3020 fieinfo->fi_flags = fiemap->fm_flags;
3021 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3022 if (extent_count > 0 &&
3023 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3024 fiemap->fm_mapped_extents *
3025 sizeof(struct fiemap_extent))) {
3034 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3036 struct ll_inode_info *lli = ll_i2info(inode);
3037 struct posix_acl *acl = NULL;
3039 spin_lock(&lli->lli_lock);
3040 /* VFS' acl_permission_check->check_acl will release the refcount */
3041 acl = posix_acl_dup(lli->lli_posix_acl);
3042 #ifdef CONFIG_FS_POSIX_ACL
3043 forget_cached_acl(inode, type);
3045 spin_unlock(&lli->lli_lock);
3050 int ll_inode_permission(struct inode *inode, int mask)
3052 struct ll_sb_info *sbi;
3053 struct root_squash_info *squash;
3054 const struct cred *old_cred = NULL;
3055 struct cred *cred = NULL;
3056 bool squash_id = false;
3060 if (mask & MAY_NOT_BLOCK)
3063 /* as root inode are NOT getting validated in lookup operation,
3064 * need to do it before permission check.
3067 if (is_root_inode(inode)) {
3068 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3069 MDS_INODELOCK_LOOKUP);
3074 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3075 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3077 /* squash fsuid/fsgid if needed */
3078 sbi = ll_i2sbi(inode);
3079 squash = &sbi->ll_squash;
3080 if (unlikely(squash->rsi_uid &&
3081 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3082 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3087 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3088 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3089 squash->rsi_uid, squash->rsi_gid);
3092 * update current process's credentials
3095 cred = prepare_creds();
3099 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3100 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3101 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3102 if ((1 << cap) & CFS_CAP_FS_MASK)
3103 cap_lower(cred->cap_effective, cap);
3105 old_cred = override_creds(cred);
3108 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
3109 rc = generic_permission(inode, mask);
3111 /* restore current process's credentials and FS capability */
3113 revert_creds(old_cred);
3120 /* -o localflock - only provides locally consistent flock locks */
3121 struct file_operations ll_file_operations = {
3122 .read_iter = ll_file_read_iter,
3123 .write_iter = ll_file_write_iter,
3124 .unlocked_ioctl = ll_file_ioctl,
3125 .open = ll_file_open,
3126 .release = ll_file_release,
3127 .mmap = ll_file_mmap,
3128 .llseek = ll_file_seek,
3129 .splice_read = generic_file_splice_read,
3134 struct file_operations ll_file_operations_flock = {
3135 .read_iter = ll_file_read_iter,
3136 .write_iter = ll_file_write_iter,
3137 .unlocked_ioctl = ll_file_ioctl,
3138 .open = ll_file_open,
3139 .release = ll_file_release,
3140 .mmap = ll_file_mmap,
3141 .llseek = ll_file_seek,
3142 .splice_read = generic_file_splice_read,
3145 .flock = ll_file_flock,
3146 .lock = ll_file_flock
3149 /* These are for -o noflock - to return ENOSYS on flock calls */
3150 struct file_operations ll_file_operations_noflock = {
3151 .read_iter = ll_file_read_iter,
3152 .write_iter = ll_file_write_iter,
3153 .unlocked_ioctl = ll_file_ioctl,
3154 .open = ll_file_open,
3155 .release = ll_file_release,
3156 .mmap = ll_file_mmap,
3157 .llseek = ll_file_seek,
3158 .splice_read = generic_file_splice_read,
3161 .flock = ll_file_noflock,
3162 .lock = ll_file_noflock
3165 const struct inode_operations ll_file_inode_operations = {
3166 .setattr = ll_setattr,
3167 .getattr = ll_getattr,
3168 .permission = ll_inode_permission,
3169 .listxattr = ll_listxattr,
3170 .fiemap = ll_fiemap,
3171 .get_acl = ll_get_acl,
3174 /* dynamic ioctl number support routines */
3175 static struct llioc_ctl_data {
3176 struct rw_semaphore ioc_sem;
3177 struct list_head ioc_head;
3179 __RWSEM_INITIALIZER(llioc.ioc_sem),
3180 LIST_HEAD_INIT(llioc.ioc_head)
3184 struct list_head iocd_list;
3185 unsigned int iocd_size;
3186 llioc_callback_t iocd_cb;
3187 unsigned int iocd_count;
3188 unsigned int iocd_cmd[0];
3191 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3194 struct llioc_data *in_data = NULL;
3196 if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0)
3199 size = sizeof(*in_data) + count * sizeof(unsigned int);
3200 in_data = kzalloc(size, GFP_NOFS);
3204 in_data->iocd_size = size;
3205 in_data->iocd_cb = cb;
3206 in_data->iocd_count = count;
3207 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3209 down_write(&llioc.ioc_sem);
3210 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3211 up_write(&llioc.ioc_sem);
3215 EXPORT_SYMBOL(ll_iocontrol_register);
3217 void ll_iocontrol_unregister(void *magic)
3219 struct llioc_data *tmp;
3224 down_write(&llioc.ioc_sem);
3225 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3227 list_del(&tmp->iocd_list);
3228 up_write(&llioc.ioc_sem);
3234 up_write(&llioc.ioc_sem);
3236 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3238 EXPORT_SYMBOL(ll_iocontrol_unregister);
3240 static enum llioc_iter
3241 ll_iocontrol_call(struct inode *inode, struct file *file,
3242 unsigned int cmd, unsigned long arg, int *rcp)
3244 enum llioc_iter ret = LLIOC_CONT;
3245 struct llioc_data *data;
3246 int rc = -EINVAL, i;
3248 down_read(&llioc.ioc_sem);
3249 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3250 for (i = 0; i < data->iocd_count; i++) {
3251 if (cmd != data->iocd_cmd[i])
3254 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3258 if (ret == LLIOC_STOP)
3261 up_read(&llioc.ioc_sem);
3268 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3270 struct ll_inode_info *lli = ll_i2info(inode);
3271 struct cl_object *obj = lli->lli_clob;
3279 env = cl_env_get(&refcheck);
3281 return PTR_ERR(env);
3283 rc = cl_conf_set(env, obj, conf);
3287 if (conf->coc_opc == OBJECT_CONF_SET) {
3288 struct ldlm_lock *lock = conf->coc_lock;
3289 struct cl_layout cl = {
3294 LASSERT(ldlm_has_layout(lock));
3296 /* it can only be allowed to match after layout is
3297 * applied to inode otherwise false layout would be
3298 * seen. Applying layout should happen before dropping
3301 ldlm_lock_allow_match(lock);
3303 rc = cl_object_layout_get(env, obj, &cl);
3307 CDEBUG(D_VFSTRACE, DFID ": layout version change: %u -> %u\n",
3308 PFID(&lli->lli_fid), ll_layout_version_get(lli),
3310 ll_layout_version_set(lli, cl.cl_layout_gen);
3313 cl_env_put(env, &refcheck);
3317 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3318 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3321 struct ll_sb_info *sbi = ll_i2sbi(inode);
3322 struct ptlrpc_request *req;
3323 struct mdt_body *body;
3329 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3330 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3331 lock->l_lvb_data, lock->l_lvb_len);
3333 if (lock->l_lvb_data && ldlm_is_lvb_ready(lock))
3336 /* if layout lock was granted right away, the layout is returned
3337 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3338 * blocked and then granted via completion ast, we have to fetch
3339 * layout here. Please note that we can't use the LVB buffer in
3340 * completion AST because it doesn't have a large enough buffer
3342 rc = ll_get_default_mdsize(sbi, &lmmsize);
3344 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
3345 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3350 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3356 lmmsize = body->mbo_eadatasize;
3357 if (lmmsize == 0) /* empty layout */ {
3362 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3368 lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS);
3374 memcpy(lvbdata, lmm, lmmsize);
3375 lock_res_and_lock(lock);
3376 if (lock->l_lvb_data)
3377 kvfree(lock->l_lvb_data);
3379 lock->l_lvb_data = lvbdata;
3380 lock->l_lvb_len = lmmsize;
3381 unlock_res_and_lock(lock);
3384 ptlrpc_req_finished(req);
3389 * Apply the layout to the inode. Layout lock is held and will be released
3392 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
3393 struct inode *inode)
3395 struct ll_inode_info *lli = ll_i2info(inode);
3396 struct ll_sb_info *sbi = ll_i2sbi(inode);
3397 struct ldlm_lock *lock;
3398 struct cl_object_conf conf;
3401 bool wait_layout = false;
3403 LASSERT(lustre_handle_is_used(lockh));
3405 lock = ldlm_handle2lock(lockh);
3407 LASSERT(ldlm_has_layout(lock));
3409 LDLM_DEBUG(lock, "File " DFID "(%p) being reconfigured",
3410 PFID(&lli->lli_fid), inode);
3412 /* in case this is a caching lock and reinstate with new inode */
3413 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
3415 lock_res_and_lock(lock);
3416 lvb_ready = ldlm_is_lvb_ready(lock);
3417 unlock_res_and_lock(lock);
3418 /* checking lvb_ready is racy but this is okay. The worst case is
3419 * that multi processes may configure the file on the same time.
3426 rc = ll_layout_fetch(inode, lock);
3430 /* for layout lock, lmm is returned in lock's lvb.
3431 * lvb_data is immutable if the lock is held so it's safe to access it
3434 * set layout to file. Unlikely this will fail as old layout was
3437 memset(&conf, 0, sizeof(conf));
3438 conf.coc_opc = OBJECT_CONF_SET;
3439 conf.coc_inode = inode;
3440 conf.coc_lock = lock;
3441 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
3442 conf.u.coc_layout.lb_len = lock->l_lvb_len;
3443 rc = ll_layout_conf(inode, &conf);
3445 /* refresh layout failed, need to wait */
3446 wait_layout = rc == -EBUSY;
3449 LDLM_LOCK_PUT(lock);
3450 ldlm_lock_decref(lockh, mode);
3452 /* wait for IO to complete if it's still being used. */
3454 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3455 ll_get_fsname(inode->i_sb, NULL, 0),
3456 PFID(&lli->lli_fid), inode);
3458 memset(&conf, 0, sizeof(conf));
3459 conf.coc_opc = OBJECT_CONF_WAIT;
3460 conf.coc_inode = inode;
3461 rc = ll_layout_conf(inode, &conf);
3465 CDEBUG(D_INODE, "%s: file="DFID" waiting layout return: %d.\n",
3466 ll_get_fsname(inode->i_sb, NULL, 0),
3467 PFID(&lli->lli_fid), rc);
3472 static int ll_layout_refresh_locked(struct inode *inode)
3474 struct ll_inode_info *lli = ll_i2info(inode);
3475 struct ll_sb_info *sbi = ll_i2sbi(inode);
3476 struct md_op_data *op_data;
3477 struct lookup_intent it;
3478 struct lustre_handle lockh;
3479 enum ldlm_mode mode;
3480 struct ldlm_enqueue_info einfo = {
3481 .ei_type = LDLM_IBITS,
3483 .ei_cb_bl = &ll_md_blocking_ast,
3484 .ei_cb_cp = &ldlm_completion_ast,
3489 /* mostly layout lock is caching on the local side, so try to match
3490 * it before grabbing layout lock mutex.
3492 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
3493 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
3494 if (mode != 0) { /* hit cached lock */
3495 rc = ll_layout_lock_set(&lockh, mode, inode);
3501 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3502 0, 0, LUSTRE_OPC_ANY, NULL);
3503 if (IS_ERR(op_data))
3504 return PTR_ERR(op_data);
3506 /* have to enqueue one */
3507 memset(&it, 0, sizeof(it));
3508 it.it_op = IT_LAYOUT;
3509 lockh.cookie = 0ULL;
3511 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
3512 ll_get_fsname(inode->i_sb, NULL, 0),
3513 PFID(&lli->lli_fid), inode);
3515 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
3516 ptlrpc_req_finished(it.it_request);
3517 it.it_request = NULL;
3519 ll_finish_md_op_data(op_data);
3521 mode = it.it_lock_mode;
3522 it.it_lock_mode = 0;
3523 ll_intent_drop_lock(&it);
3526 /* set lock data in case this is a new lock */
3527 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3528 rc = ll_layout_lock_set(&lockh, mode, inode);
3537 * This function checks if there exists a LAYOUT lock on the client side,
3538 * or enqueues it if it doesn't have one in cache.
3540 * This function will not hold layout lock so it may be revoked any time after
3541 * this function returns. Any operations depend on layout should be redone
3544 * This function should be called before lov_io_init() to get an uptodate
3545 * layout version, the caller should save the version number and after IO
3546 * is finished, this function should be called again to verify that layout
3547 * is not changed during IO time.
3549 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3551 struct ll_inode_info *lli = ll_i2info(inode);
3552 struct ll_sb_info *sbi = ll_i2sbi(inode);
3555 *gen = ll_layout_version_get(lli);
3556 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
3560 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3561 LASSERT(S_ISREG(inode->i_mode));
3563 /* take layout lock mutex to enqueue layout lock exclusively. */
3564 mutex_lock(&lli->lli_layout_mutex);
3566 rc = ll_layout_refresh_locked(inode);
3570 *gen = ll_layout_version_get(lli);
3572 mutex_unlock(&lli->lli_layout_mutex);
3578 * This function send a restore request to the MDT
3580 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
3582 struct hsm_user_request *hur;
3585 len = sizeof(struct hsm_user_request) +
3586 sizeof(struct hsm_user_item);
3587 hur = kzalloc(len, GFP_NOFS);
3591 hur->hur_request.hr_action = HUA_RESTORE;
3592 hur->hur_request.hr_archive_id = 0;
3593 hur->hur_request.hr_flags = 0;
3594 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
3595 sizeof(hur->hur_user_item[0].hui_fid));
3596 hur->hur_user_item[0].hui_extent.offset = offset;
3597 hur->hur_user_item[0].hui_extent.length = length;
3598 hur->hur_request.hr_itemcount = 1;
3599 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,