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, 2016, 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 <lustre_dlm.h>
41 #include <linux/pagemap.h>
42 #include <linux/file.h>
43 #include <linux/sched.h>
44 #include <linux/user_namespace.h>
45 #ifdef HAVE_UIDGID_HEADER
46 # include <linux/uidgid.h>
48 #include <lustre/ll_fiemap.h>
50 #include <lustre_ioctl.h>
51 #include <lustre_swab.h>
53 #include "cl_object.h"
54 #include "llite_internal.h"
55 #include "vvp_internal.h"
58 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
60 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
63 static enum llioc_iter
64 ll_iocontrol_call(struct inode *inode, struct file *file,
65 unsigned int cmd, unsigned long arg, int *rcp);
67 static struct ll_file_data *ll_file_data_get(void)
69 struct ll_file_data *fd;
71 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
75 fd->fd_write_failed = false;
80 static void ll_file_data_put(struct ll_file_data *fd)
83 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
87 * Packs all the attributes into @op_data for the CLOSE rpc.
89 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
90 struct obd_client_handle *och)
94 ll_prep_md_op_data(op_data, inode, NULL, NULL,
95 0, 0, LUSTRE_OPC_ANY, NULL);
97 op_data->op_attr.ia_mode = inode->i_mode;
98 op_data->op_attr.ia_atime = inode->i_atime;
99 op_data->op_attr.ia_mtime = inode->i_mtime;
100 op_data->op_attr.ia_ctime = inode->i_ctime;
101 op_data->op_attr.ia_size = i_size_read(inode);
102 op_data->op_attr.ia_valid |= ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
103 ATTR_MTIME | ATTR_MTIME_SET |
104 ATTR_CTIME | ATTR_CTIME_SET;
105 op_data->op_attr_blocks = inode->i_blocks;
106 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
107 op_data->op_handle = och->och_fh;
109 if (och->och_flags & FMODE_WRITE &&
110 ll_file_test_and_clear_flag(ll_i2info(inode), LLIF_DATA_MODIFIED))
111 /* For HSM: if inode data has been modified, pack it so that
112 * MDT can set data dirty flag in the archive. */
113 op_data->op_bias |= MDS_DATA_MODIFIED;
119 * Perform a close, possibly with a bias.
120 * The meaning of "data" depends on the value of "bias".
122 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
123 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
126 static int ll_close_inode_openhandle(struct inode *inode,
127 struct obd_client_handle *och,
128 enum mds_op_bias bias, void *data)
130 struct obd_export *md_exp = ll_i2mdexp(inode);
131 const struct ll_inode_info *lli = ll_i2info(inode);
132 struct md_op_data *op_data;
133 struct ptlrpc_request *req = NULL;
137 if (class_exp2obd(md_exp) == NULL) {
138 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
139 ll_get_fsname(inode->i_sb, NULL, 0),
140 PFID(&lli->lli_fid));
144 OBD_ALLOC_PTR(op_data);
145 /* We leak openhandle and request here on error, but not much to be
146 * done in OOM case since app won't retry close on error either. */
148 GOTO(out, rc = -ENOMEM);
150 ll_prepare_close(inode, op_data, och);
152 case MDS_CLOSE_LAYOUT_SWAP:
153 LASSERT(data != NULL);
154 op_data->op_bias |= MDS_CLOSE_LAYOUT_SWAP;
155 op_data->op_data_version = 0;
156 op_data->op_lease_handle = och->och_lease_handle;
157 op_data->op_fid2 = *ll_inode2fid(data);
160 case MDS_HSM_RELEASE:
161 LASSERT(data != NULL);
162 op_data->op_bias |= MDS_HSM_RELEASE;
163 op_data->op_data_version = *(__u64 *)data;
164 op_data->op_lease_handle = och->och_lease_handle;
165 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
169 LASSERT(data == NULL);
173 rc = md_close(md_exp, op_data, och->och_mod, &req);
174 if (rc != 0 && rc != -EINTR)
175 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
176 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
179 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);
191 md_clear_open_replay_data(md_exp, och);
192 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
195 ptlrpc_req_finished(req); /* This is close request */
199 int ll_md_real_close(struct inode *inode, fmode_t fmode)
201 struct ll_inode_info *lli = ll_i2info(inode);
202 struct obd_client_handle **och_p;
203 struct obd_client_handle *och;
208 if (fmode & FMODE_WRITE) {
209 och_p = &lli->lli_mds_write_och;
210 och_usecount = &lli->lli_open_fd_write_count;
211 } else if (fmode & FMODE_EXEC) {
212 och_p = &lli->lli_mds_exec_och;
213 och_usecount = &lli->lli_open_fd_exec_count;
215 LASSERT(fmode & FMODE_READ);
216 och_p = &lli->lli_mds_read_och;
217 och_usecount = &lli->lli_open_fd_read_count;
220 mutex_lock(&lli->lli_och_mutex);
221 if (*och_usecount > 0) {
222 /* There are still users of this handle, so skip
224 mutex_unlock(&lli->lli_och_mutex);
230 mutex_unlock(&lli->lli_och_mutex);
233 /* 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 union ldlm_policy_data policy = {
244 .l_inodebits = { MDS_INODELOCK_OPEN },
246 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
247 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
248 struct ll_inode_info *lli = ll_i2info(inode);
249 struct lustre_handle lockh;
250 enum ldlm_mode lockmode;
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 != NULL) {
261 /* Usually the lease is not released when the
262 * 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, "Clean up lease "DFID" %d/%d\n",
265 PFID(&lli->lli_fid), rc, lease_broken);
267 fd->fd_lease_och = NULL;
270 if (fd->fd_och != NULL) {
271 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
276 /* Let's see if we have good enough OPEN lock on the file and if
277 we can skip talking to MDS */
278 mutex_lock(&lli->lli_och_mutex);
279 if (fd->fd_omode & FMODE_WRITE) {
281 LASSERT(lli->lli_open_fd_write_count);
282 lli->lli_open_fd_write_count--;
283 } else if (fd->fd_omode & FMODE_EXEC) {
285 LASSERT(lli->lli_open_fd_exec_count);
286 lli->lli_open_fd_exec_count--;
289 LASSERT(lli->lli_open_fd_read_count);
290 lli->lli_open_fd_read_count--;
292 mutex_unlock(&lli->lli_och_mutex);
294 if (!md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
295 LDLM_IBITS, &policy, lockmode, &lockh))
296 rc = ll_md_real_close(inode, fd->fd_omode);
299 LUSTRE_FPRIVATE(file) = NULL;
300 ll_file_data_put(fd);
305 /* While this returns an error code, fput() the caller does not, so we need
306 * to make every effort to clean up all of our state here. Also, applications
307 * rarely check close errors and even if an error is returned they will not
308 * re-try the close call.
310 int ll_file_release(struct inode *inode, struct file *file)
312 struct ll_file_data *fd;
313 struct ll_sb_info *sbi = ll_i2sbi(inode);
314 struct ll_inode_info *lli = ll_i2info(inode);
318 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
319 PFID(ll_inode2fid(inode)), inode);
321 if (inode->i_sb->s_root != file_dentry(file))
322 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
323 fd = LUSTRE_FPRIVATE(file);
326 /* The last ref on @file, maybe not the the owner pid of statahead,
327 * because parent and child process can share the same file handle. */
328 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
329 ll_deauthorize_statahead(inode, fd);
331 if (inode->i_sb->s_root == file_dentry(file)) {
332 LUSTRE_FPRIVATE(file) = NULL;
333 ll_file_data_put(fd);
337 if (!S_ISDIR(inode->i_mode)) {
338 if (lli->lli_clob != NULL)
339 lov_read_and_clear_async_rc(lli->lli_clob);
340 lli->lli_async_rc = 0;
343 rc = ll_md_close(inode, file);
345 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
346 libcfs_debug_dumplog();
351 static int ll_intent_file_open(struct file *file, void *lmm, int lmmsize,
352 struct lookup_intent *itp)
354 struct dentry *de = file_dentry(file);
355 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
356 struct dentry *parent = de->d_parent;
357 const char *name = NULL;
359 struct md_op_data *op_data;
360 struct ptlrpc_request *req = NULL;
364 LASSERT(parent != NULL);
365 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
367 /* if server supports open-by-fid, or file name is invalid, don't pack
368 * name in open request */
369 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
370 lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
371 name = de->d_name.name;
372 len = de->d_name.len;
375 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
376 name, len, 0, LUSTRE_OPC_ANY, NULL);
378 RETURN(PTR_ERR(op_data));
379 op_data->op_data = lmm;
380 op_data->op_data_size = lmmsize;
382 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
383 &ll_md_blocking_ast, 0);
384 ll_finish_md_op_data(op_data);
386 /* reason for keep own exit path - don`t flood log
387 * with messages with -ESTALE errors.
389 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
390 it_open_error(DISP_OPEN_OPEN, itp))
392 ll_release_openhandle(de, itp);
396 if (it_disposition(itp, DISP_LOOKUP_NEG))
397 GOTO(out, rc = -ENOENT);
399 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
400 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
401 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
405 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
406 if (!rc && itp->it_lock_mode)
407 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
410 ptlrpc_req_finished(req);
411 ll_intent_drop_lock(itp);
413 /* We did open by fid, but by the time we got to the server,
414 * the object disappeared. If this is a create, we cannot really
415 * tell the userspace that the file it was trying to create
416 * does not exist. Instead let's return -ESTALE, and the VFS will
417 * retry the create with LOOKUP_REVAL that we are going to catch
418 * in ll_revalidate_dentry() and use lookup then.
420 if (rc == -ENOENT && itp->it_op & IT_CREAT)
426 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
427 struct obd_client_handle *och)
429 struct mdt_body *body;
431 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
432 och->och_fh = body->mbo_handle;
433 och->och_fid = body->mbo_fid1;
434 och->och_lease_handle.cookie = it->it_lock_handle;
435 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
436 och->och_flags = it->it_flags;
438 return md_set_open_replay_data(md_exp, och, it);
441 static int ll_local_open(struct file *file, struct lookup_intent *it,
442 struct ll_file_data *fd, struct obd_client_handle *och)
444 struct inode *inode = file_inode(file);
447 LASSERT(!LUSTRE_FPRIVATE(file));
454 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
459 LUSTRE_FPRIVATE(file) = fd;
460 ll_readahead_init(inode, &fd->fd_ras);
461 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
463 /* ll_cl_context initialize */
464 rwlock_init(&fd->fd_lock);
465 INIT_LIST_HEAD(&fd->fd_lccs);
470 /* Open a file, and (for the very first open) create objects on the OSTs at
471 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
472 * creation or open until ll_lov_setstripe() ioctl is called.
474 * If we already have the stripe MD locally then we don't request it in
475 * md_open(), by passing a lmm_size = 0.
477 * It is up to the application to ensure no other processes open this file
478 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
479 * used. We might be able to avoid races of that sort by getting lli_open_sem
480 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
481 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
483 int ll_file_open(struct inode *inode, struct file *file)
485 struct ll_inode_info *lli = ll_i2info(inode);
486 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
487 .it_flags = file->f_flags };
488 struct obd_client_handle **och_p = NULL;
489 __u64 *och_usecount = NULL;
490 struct ll_file_data *fd;
494 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
495 PFID(ll_inode2fid(inode)), inode, file->f_flags);
497 it = file->private_data; /* XXX: compat macro */
498 file->private_data = NULL; /* prevent ll_local_open assertion */
500 fd = ll_file_data_get();
502 GOTO(out_openerr, rc = -ENOMEM);
505 if (S_ISDIR(inode->i_mode))
506 ll_authorize_statahead(inode, fd);
508 if (inode->i_sb->s_root == file_dentry(file)) {
509 LUSTRE_FPRIVATE(file) = fd;
513 if (!it || !it->it_disposition) {
514 /* Convert f_flags into access mode. We cannot use file->f_mode,
515 * because everything but O_ACCMODE mask was stripped from
517 if ((oit.it_flags + 1) & O_ACCMODE)
519 if (file->f_flags & O_TRUNC)
520 oit.it_flags |= FMODE_WRITE;
522 /* kernel only call f_op->open in dentry_open. filp_open calls
523 * dentry_open after call to open_namei that checks permissions.
524 * Only nfsd_open call dentry_open directly without checking
525 * permissions and because of that this code below is safe. */
526 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
527 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
529 /* We do not want O_EXCL here, presumably we opened the file
530 * already? XXX - NFS implications? */
531 oit.it_flags &= ~O_EXCL;
533 /* bug20584, if "it_flags" contains O_CREAT, the file will be
534 * created if necessary, then "IT_CREAT" should be set to keep
535 * consistent with it */
536 if (oit.it_flags & O_CREAT)
537 oit.it_op |= IT_CREAT;
543 /* Let's see if we have file open on MDS already. */
544 if (it->it_flags & FMODE_WRITE) {
545 och_p = &lli->lli_mds_write_och;
546 och_usecount = &lli->lli_open_fd_write_count;
547 } else if (it->it_flags & FMODE_EXEC) {
548 och_p = &lli->lli_mds_exec_och;
549 och_usecount = &lli->lli_open_fd_exec_count;
551 och_p = &lli->lli_mds_read_och;
552 och_usecount = &lli->lli_open_fd_read_count;
555 mutex_lock(&lli->lli_och_mutex);
556 if (*och_p) { /* Open handle is present */
557 if (it_disposition(it, DISP_OPEN_OPEN)) {
558 /* Well, there's extra open request that we do not need,
559 let's close it somehow. This will decref request. */
560 rc = it_open_error(DISP_OPEN_OPEN, it);
562 mutex_unlock(&lli->lli_och_mutex);
563 GOTO(out_openerr, rc);
566 ll_release_openhandle(file_dentry(file), it);
570 rc = ll_local_open(file, it, fd, NULL);
573 mutex_unlock(&lli->lli_och_mutex);
574 GOTO(out_openerr, rc);
577 LASSERT(*och_usecount == 0);
578 if (!it->it_disposition) {
579 struct ll_dentry_data *ldd = ll_d2d(file->f_path.dentry);
580 /* We cannot just request lock handle now, new ELC code
581 means that one of other OPEN locks for this file
582 could be cancelled, and since blocking ast handler
583 would attempt to grab och_mutex as well, that would
584 result in a deadlock */
585 mutex_unlock(&lli->lli_och_mutex);
587 * Normally called under two situations:
589 * 2. A race/condition on MDS resulting in no open
590 * handle to be returned from LOOKUP|OPEN request,
591 * for example if the target entry was a symlink.
593 * Only fetch MDS_OPEN_LOCK if this is in NFS path,
594 * marked by a bit set in ll_iget_for_nfs. Clear the
595 * bit so that it's not confusing later callers.
597 * NB; when ldd is NULL, it must have come via normal
598 * lookup path only, since ll_iget_for_nfs always calls
601 if (ldd && ldd->lld_nfs_dentry) {
602 ldd->lld_nfs_dentry = 0;
603 it->it_flags |= MDS_OPEN_LOCK;
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_BY_FID;
611 rc = ll_intent_file_open(file, NULL, 0, it);
613 GOTO(out_openerr, rc);
617 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
619 GOTO(out_och_free, rc = -ENOMEM);
623 /* md_intent_lock() didn't get a request ref if there was an
624 * open error, so don't do cleanup on the request here
626 /* XXX (green): Should not we bail out on any error here, not
627 * just open error? */
628 rc = it_open_error(DISP_OPEN_OPEN, it);
630 GOTO(out_och_free, rc);
632 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
633 "inode %p: disposition %x, status %d\n", inode,
634 it_disposition(it, ~0), it->it_status);
636 rc = ll_local_open(file, it, fd, *och_p);
638 GOTO(out_och_free, rc);
640 mutex_unlock(&lli->lli_och_mutex);
643 /* Must do this outside lli_och_mutex lock to prevent deadlock where
644 different kind of OPEN lock for this same inode gets cancelled
645 by ldlm_cancel_lru */
646 if (!S_ISREG(inode->i_mode))
647 GOTO(out_och_free, rc);
649 cl_lov_delay_create_clear(&file->f_flags);
650 GOTO(out_och_free, rc);
654 if (och_p && *och_p) {
655 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
656 *och_p = NULL; /* OBD_FREE writes some magic there */
659 mutex_unlock(&lli->lli_och_mutex);
662 if (lli->lli_opendir_key == fd)
663 ll_deauthorize_statahead(inode, fd);
665 ll_file_data_put(fd);
667 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
670 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
671 ptlrpc_req_finished(it->it_request);
672 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
678 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
679 struct ldlm_lock_desc *desc, void *data, int flag)
682 struct lustre_handle lockh;
686 case LDLM_CB_BLOCKING:
687 ldlm_lock2handle(lock, &lockh);
688 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
690 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
694 case LDLM_CB_CANCELING:
702 * When setting a lease on a file, we take ownership of the lli_mds_*_och
703 * and save it as fd->fd_och so as to force client to reopen the file even
704 * if it has an open lock in cache already.
706 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
707 struct lustre_handle *old_handle)
709 struct ll_inode_info *lli = ll_i2info(inode);
710 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
711 struct obd_client_handle **och_p;
716 /* Get the openhandle of the file */
717 mutex_lock(&lli->lli_och_mutex);
718 if (fd->fd_lease_och != NULL)
719 GOTO(out_unlock, rc = -EBUSY);
721 if (fd->fd_och == NULL) {
722 if (file->f_mode & FMODE_WRITE) {
723 LASSERT(lli->lli_mds_write_och != NULL);
724 och_p = &lli->lli_mds_write_och;
725 och_usecount = &lli->lli_open_fd_write_count;
727 LASSERT(lli->lli_mds_read_och != NULL);
728 och_p = &lli->lli_mds_read_och;
729 och_usecount = &lli->lli_open_fd_read_count;
732 if (*och_usecount > 1)
733 GOTO(out_unlock, rc = -EBUSY);
740 *old_handle = fd->fd_och->och_fh;
744 mutex_unlock(&lli->lli_och_mutex);
749 * Release ownership on lli_mds_*_och when putting back a file lease.
751 static int ll_lease_och_release(struct inode *inode, struct file *file)
753 struct ll_inode_info *lli = ll_i2info(inode);
754 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
755 struct obd_client_handle **och_p;
756 struct obd_client_handle *old_och = NULL;
761 mutex_lock(&lli->lli_och_mutex);
762 if (file->f_mode & FMODE_WRITE) {
763 och_p = &lli->lli_mds_write_och;
764 och_usecount = &lli->lli_open_fd_write_count;
766 och_p = &lli->lli_mds_read_och;
767 och_usecount = &lli->lli_open_fd_read_count;
770 /* The file may have been open by another process (broken lease) so
771 * *och_p is not NULL. In this case we should simply increase usecount
774 if (*och_p != NULL) {
775 old_och = fd->fd_och;
782 mutex_unlock(&lli->lli_och_mutex);
785 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
791 * Acquire a lease and open the file.
793 static struct obd_client_handle *
794 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
797 struct lookup_intent it = { .it_op = IT_OPEN };
798 struct ll_sb_info *sbi = ll_i2sbi(inode);
799 struct md_op_data *op_data;
800 struct ptlrpc_request *req = NULL;
801 struct lustre_handle old_handle = { 0 };
802 struct obd_client_handle *och = NULL;
807 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
808 RETURN(ERR_PTR(-EINVAL));
811 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
812 RETURN(ERR_PTR(-EPERM));
814 rc = ll_lease_och_acquire(inode, file, &old_handle);
821 RETURN(ERR_PTR(-ENOMEM));
823 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
824 LUSTRE_OPC_ANY, NULL);
826 GOTO(out, rc = PTR_ERR(op_data));
828 /* To tell the MDT this openhandle is from the same owner */
829 op_data->op_handle = old_handle;
831 it.it_flags = fmode | open_flags;
832 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
833 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
834 &ll_md_blocking_lease_ast,
835 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
836 * it can be cancelled which may mislead applications that the lease is
838 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
839 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
840 * doesn't deal with openhandle, so normal openhandle will be leaked. */
841 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
842 ll_finish_md_op_data(op_data);
843 ptlrpc_req_finished(req);
845 GOTO(out_release_it, rc);
847 if (it_disposition(&it, DISP_LOOKUP_NEG))
848 GOTO(out_release_it, rc = -ENOENT);
850 rc = it_open_error(DISP_OPEN_OPEN, &it);
852 GOTO(out_release_it, rc);
854 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
855 ll_och_fill(sbi->ll_md_exp, &it, och);
857 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
858 GOTO(out_close, rc = -EOPNOTSUPP);
860 /* already get lease, handle lease lock */
861 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
862 if (it.it_lock_mode == 0 ||
863 it.it_lock_bits != MDS_INODELOCK_OPEN) {
864 /* open lock must return for lease */
865 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
866 PFID(ll_inode2fid(inode)), it.it_lock_mode,
868 GOTO(out_close, rc = -EPROTO);
871 ll_intent_release(&it);
875 /* Cancel open lock */
876 if (it.it_lock_mode != 0) {
877 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
880 och->och_lease_handle.cookie = 0ULL;
882 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
884 CERROR("%s: error closing file "DFID": %d\n",
885 ll_get_fsname(inode->i_sb, NULL, 0),
886 PFID(&ll_i2info(inode)->lli_fid), rc2);
887 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
889 ll_intent_release(&it);
897 * Check whether a layout swap can be done between two inodes.
899 * \param[in] inode1 First inode to check
900 * \param[in] inode2 Second inode to check
902 * \retval 0 on success, layout swap can be performed between both inodes
903 * \retval negative error code if requirements are not met
905 static int ll_check_swap_layouts_validity(struct inode *inode1,
906 struct inode *inode2)
908 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
911 if (inode_permission(inode1, MAY_WRITE) ||
912 inode_permission(inode2, MAY_WRITE))
915 if (inode1->i_sb != inode2->i_sb)
921 static int ll_swap_layouts_close(struct obd_client_handle *och,
922 struct inode *inode, struct inode *inode2)
924 const struct lu_fid *fid1 = ll_inode2fid(inode);
925 const struct lu_fid *fid2;
929 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
930 ll_get_fsname(inode->i_sb, NULL, 0), PFID(fid1));
932 rc = ll_check_swap_layouts_validity(inode, inode2);
934 GOTO(out_free_och, rc);
936 /* We now know that inode2 is a lustre inode */
937 fid2 = ll_inode2fid(inode2);
939 rc = lu_fid_cmp(fid1, fid2);
941 GOTO(out_free_och, rc = -EINVAL);
943 /* Close the file and swap layouts between inode & inode2.
944 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
945 * because we still need it to pack l_remote_handle to MDT. */
946 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
949 och = NULL; /* freed in ll_close_inode_openhandle() */
959 * Release lease and close the file.
960 * It will check if the lease has ever broken.
962 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
965 struct ldlm_lock *lock;
966 bool cancelled = true;
970 lock = ldlm_handle2lock(&och->och_lease_handle);
972 lock_res_and_lock(lock);
973 cancelled = ldlm_is_cancel(lock);
974 unlock_res_and_lock(lock);
978 CDEBUG(D_INODE, "lease for "DFID" broken? %d\n",
979 PFID(&ll_i2info(inode)->lli_fid), cancelled);
982 ldlm_cli_cancel(&och->och_lease_handle, 0);
984 if (lease_broken != NULL)
985 *lease_broken = cancelled;
987 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
991 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
993 struct ll_inode_info *lli = ll_i2info(inode);
994 struct cl_object *obj = lli->lli_clob;
995 struct cl_attr *attr = vvp_env_thread_attr(env);
1003 ll_inode_size_lock(inode);
1005 /* Merge timestamps the most recently obtained from MDS with
1006 * timestamps obtained from OSTs.
1008 * Do not overwrite atime of inode because it may be refreshed
1009 * by file_accessed() function. If the read was served by cache
1010 * data, there is no RPC to be sent so that atime may not be
1011 * transferred to OSTs at all. MDT only updates atime at close time
1012 * if it's at least 'mdd.*.atime_diff' older.
1013 * All in all, the atime in Lustre does not strictly comply with
1014 * POSIX. Solving this problem needs to send an RPC to MDT for each
1015 * read, this will hurt performance. */
1016 if (LTIME_S(inode->i_atime) < lli->lli_atime || lli->lli_update_atime) {
1017 LTIME_S(inode->i_atime) = lli->lli_atime;
1018 lli->lli_update_atime = 0;
1020 LTIME_S(inode->i_mtime) = lli->lli_mtime;
1021 LTIME_S(inode->i_ctime) = lli->lli_ctime;
1023 atime = LTIME_S(inode->i_atime);
1024 mtime = LTIME_S(inode->i_mtime);
1025 ctime = LTIME_S(inode->i_ctime);
1027 cl_object_attr_lock(obj);
1028 rc = cl_object_attr_get(env, obj, attr);
1029 cl_object_attr_unlock(obj);
1032 GOTO(out_size_unlock, rc);
1034 if (atime < attr->cat_atime)
1035 atime = attr->cat_atime;
1037 if (ctime < attr->cat_ctime)
1038 ctime = attr->cat_ctime;
1040 if (mtime < attr->cat_mtime)
1041 mtime = attr->cat_mtime;
1043 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1044 PFID(&lli->lli_fid), attr->cat_size);
1046 i_size_write(inode, attr->cat_size);
1047 inode->i_blocks = attr->cat_blocks;
1049 LTIME_S(inode->i_atime) = atime;
1050 LTIME_S(inode->i_mtime) = mtime;
1051 LTIME_S(inode->i_ctime) = ctime;
1054 ll_inode_size_unlock(inode);
1059 static bool file_is_noatime(const struct file *file)
1061 const struct vfsmount *mnt = file->f_path.mnt;
1062 const struct inode *inode = file_inode((struct file *)file);
1064 /* Adapted from file_accessed() and touch_atime().*/
1065 if (file->f_flags & O_NOATIME)
1068 if (inode->i_flags & S_NOATIME)
1071 if (IS_NOATIME(inode))
1074 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1077 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1080 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1086 static void ll_io_init(struct cl_io *io, const struct file *file, int write)
1088 struct inode *inode = file_inode((struct file *)file);
1090 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1092 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1093 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1094 file->f_flags & O_DIRECT ||
1097 io->ci_obj = ll_i2info(inode)->lli_clob;
1098 io->ci_lockreq = CILR_MAYBE;
1099 if (ll_file_nolock(file)) {
1100 io->ci_lockreq = CILR_NEVER;
1101 io->ci_no_srvlock = 1;
1102 } else if (file->f_flags & O_APPEND) {
1103 io->ci_lockreq = CILR_MANDATORY;
1106 io->ci_noatime = file_is_noatime(file);
1110 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1111 struct file *file, enum cl_io_type iot,
1112 loff_t *ppos, size_t count)
1114 struct vvp_io *vio = vvp_env_io(env);
1115 struct inode *inode = file_inode(file);
1116 struct ll_inode_info *lli = ll_i2info(inode);
1117 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1121 struct range_lock range;
1125 CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: %llu, count: %zu\n",
1126 file_dentry(file)->d_name.name, iot, *ppos, count);
1129 io = vvp_env_thread_io(env);
1130 ll_io_init(io, file, iot == CIT_WRITE);
1132 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1133 bool range_locked = false;
1135 if (file->f_flags & O_APPEND)
1136 range_lock_init(&range, 0, LUSTRE_EOF);
1138 range_lock_init(&range, *ppos, *ppos + count - 1);
1140 vio->vui_fd = LUSTRE_FPRIVATE(file);
1141 vio->vui_io_subtype = args->via_io_subtype;
1143 switch (vio->vui_io_subtype) {
1145 vio->vui_iter = args->u.normal.via_iter;
1146 vio->vui_iocb = args->u.normal.via_iocb;
1147 /* Direct IO reads must also take range lock,
1148 * or multiple reads will try to work on the same pages
1149 * See LU-6227 for details. */
1150 if (((iot == CIT_WRITE) ||
1151 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1152 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1153 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1155 rc = range_lock(&lli->lli_write_tree, &range);
1159 range_locked = true;
1163 vio->u.splice.vui_pipe = args->u.splice.via_pipe;
1164 vio->u.splice.vui_flags = args->u.splice.via_flags;
1167 CERROR("unknown IO subtype %u\n", vio->vui_io_subtype);
1171 ll_cl_add(file, env, io, LCC_RW);
1172 rc = cl_io_loop(env, io);
1173 ll_cl_remove(file, env);
1176 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1178 range_unlock(&lli->lli_write_tree, &range);
1181 /* cl_io_rw_init() handled IO */
1185 if (io->ci_nob > 0) {
1186 result += io->ci_nob;
1187 count -= io->ci_nob;
1188 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1190 /* prepare IO restart */
1191 if (count > 0 && args->via_io_subtype == IO_NORMAL)
1192 args->u.normal.via_iter = vio->vui_iter;
1196 cl_io_fini(env, io);
1198 if ((rc == 0 || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
1200 "%s: restart %s from %lld, count:%zu, result: %zd\n",
1201 file_dentry(file)->d_name.name,
1202 iot == CIT_READ ? "read" : "write",
1203 *ppos, count, result);
1207 if (iot == CIT_READ) {
1209 ll_stats_ops_tally(ll_i2sbi(inode),
1210 LPROC_LL_READ_BYTES, result);
1211 } else if (iot == CIT_WRITE) {
1213 ll_stats_ops_tally(ll_i2sbi(inode),
1214 LPROC_LL_WRITE_BYTES, result);
1215 fd->fd_write_failed = false;
1216 } else if (result == 0 && rc == 0) {
1219 fd->fd_write_failed = true;
1221 fd->fd_write_failed = false;
1222 } else if (rc != -ERESTARTSYS) {
1223 fd->fd_write_failed = true;
1227 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1229 return result > 0 ? result : rc;
1233 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1234 * especially for small I/O.
1236 * To serve a read request, CLIO has to create and initialize a cl_io and
1237 * then request DLM lock. This has turned out to have siginificant overhead
1238 * and affects the performance of small I/O dramatically.
1240 * It's not necessary to create a cl_io for each I/O. Under the help of read
1241 * ahead, most of the pages being read are already in memory cache and we can
1242 * read those pages directly because if the pages exist, the corresponding DLM
1243 * lock must exist so that page content must be valid.
1245 * In fast read implementation, the llite speculatively finds and reads pages
1246 * in memory cache. There are three scenarios for fast read:
1247 * - If the page exists and is uptodate, kernel VM will provide the data and
1248 * CLIO won't be intervened;
1249 * - If the page was brought into memory by read ahead, it will be exported
1250 * and read ahead parameters will be updated;
1251 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1252 * it will go back and invoke normal read, i.e., a cl_io will be created
1253 * and DLM lock will be requested.
1255 * POSIX compliance: posix standard states that read is intended to be atomic.
1256 * Lustre read implementation is in line with Linux kernel read implementation
1257 * and neither of them complies with POSIX standard in this matter. Fast read
1258 * doesn't make the situation worse on single node but it may interleave write
1259 * results from multiple nodes due to short read handling in ll_file_aio_read().
1261 * \param env - lu_env
1262 * \param iocb - kiocb from kernel
1263 * \param iter - user space buffers where the data will be copied
1265 * \retval - number of bytes have been read, or error code if error occurred.
1268 ll_do_fast_read(const struct lu_env *env, struct kiocb *iocb,
1269 struct iov_iter *iter)
1273 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1276 /* NB: we can't do direct IO for fast read because it will need a lock
1277 * to make IO engine happy. */
1278 if (iocb->ki_filp->f_flags & O_DIRECT)
1281 ll_cl_add(iocb->ki_filp, env, NULL, LCC_RW);
1282 result = generic_file_read_iter(iocb, iter);
1283 ll_cl_remove(iocb->ki_filp, env);
1285 /* If the first page is not in cache, generic_file_aio_read() will be
1286 * returned with -ENODATA.
1287 * See corresponding code in ll_readpage(). */
1288 if (result == -ENODATA)
1292 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1293 LPROC_LL_READ_BYTES, result);
1299 * Read from a file (through the page cache).
1301 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1304 struct vvp_io_args *args;
1309 env = cl_env_get(&refcheck);
1311 return PTR_ERR(env);
1313 result = ll_do_fast_read(env, iocb, to);
1314 if (result < 0 || iov_iter_count(to) == 0)
1317 args = ll_env_args(env, IO_NORMAL);
1318 args->u.normal.via_iter = to;
1319 args->u.normal.via_iocb = iocb;
1321 rc2 = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1322 &iocb->ki_pos, iov_iter_count(to));
1325 else if (result == 0)
1329 cl_env_put(env, &refcheck);
1334 * Write to a file (through the page cache).
1336 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1338 struct vvp_io_args *args;
1343 env = cl_env_get(&refcheck);
1345 return PTR_ERR(env);
1347 args = ll_env_args(env, IO_NORMAL);
1348 args->u.normal.via_iter = from;
1349 args->u.normal.via_iocb = iocb;
1351 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1352 &iocb->ki_pos, iov_iter_count(from));
1353 cl_env_put(env, &refcheck);
1357 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1359 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1361 static int ll_file_get_iov_count(const struct iovec *iov,
1362 unsigned long *nr_segs, size_t *count)
1367 for (seg = 0; seg < *nr_segs; seg++) {
1368 const struct iovec *iv = &iov[seg];
1371 * If any segment has a negative length, or the cumulative
1372 * length ever wraps negative then return -EINVAL.
1375 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1377 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1382 cnt -= iv->iov_len; /* This segment is no good */
1389 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1390 unsigned long nr_segs, loff_t pos)
1397 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1401 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1402 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
1403 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1404 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
1405 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1407 result = ll_file_read_iter(iocb, &to);
1412 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
1416 struct iovec iov = { .iov_base = buf, .iov_len = count };
1417 struct kiocb *kiocb;
1422 env = cl_env_get(&refcheck);
1424 RETURN(PTR_ERR(env));
1426 kiocb = &ll_env_info(env)->lti_kiocb;
1427 init_sync_kiocb(kiocb, file);
1428 kiocb->ki_pos = *ppos;
1429 #ifdef HAVE_KIOCB_KI_LEFT
1430 kiocb->ki_left = count;
1431 #elif defined(HAVE_KI_NBYTES)
1432 kiocb->ki_nbytes = count;
1435 result = ll_file_aio_read(kiocb, &iov, 1, kiocb->ki_pos);
1436 *ppos = kiocb->ki_pos;
1438 cl_env_put(env, &refcheck);
1443 * Write to a file (through the page cache).
1446 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1447 unsigned long nr_segs, loff_t pos)
1449 struct iov_iter from;
1454 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1458 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1459 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
1460 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1461 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
1462 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1464 result = ll_file_write_iter(iocb, &from);
1469 static ssize_t ll_file_write(struct file *file, const char __user *buf,
1470 size_t count, loff_t *ppos)
1473 struct iovec iov = { .iov_base = (void __user *)buf,
1475 struct kiocb *kiocb;
1480 env = cl_env_get(&refcheck);
1482 RETURN(PTR_ERR(env));
1484 kiocb = &ll_env_info(env)->lti_kiocb;
1485 init_sync_kiocb(kiocb, file);
1486 kiocb->ki_pos = *ppos;
1487 #ifdef HAVE_KIOCB_KI_LEFT
1488 kiocb->ki_left = count;
1489 #elif defined(HAVE_KI_NBYTES)
1490 kiocb->ki_nbytes = count;
1493 result = ll_file_aio_write(kiocb, &iov, 1, kiocb->ki_pos);
1494 *ppos = kiocb->ki_pos;
1496 cl_env_put(env, &refcheck);
1499 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
1502 * Send file content (through pagecache) somewhere with helper
1504 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1505 struct pipe_inode_info *pipe, size_t count,
1509 struct vvp_io_args *args;
1514 env = cl_env_get(&refcheck);
1516 RETURN(PTR_ERR(env));
1518 args = ll_env_args(env, IO_SPLICE);
1519 args->u.splice.via_pipe = pipe;
1520 args->u.splice.via_flags = flags;
1522 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1523 cl_env_put(env, &refcheck);
1527 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1528 __u64 flags, struct lov_user_md *lum,
1531 struct lookup_intent oit = {
1533 .it_flags = flags | MDS_OPEN_BY_FID,
1538 ll_inode_size_lock(inode);
1539 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1541 GOTO(out_unlock, rc);
1543 ll_release_openhandle(file_dentry(file), &oit);
1546 ll_inode_size_unlock(inode);
1547 ll_intent_release(&oit);
1548 cl_lov_delay_create_clear(&file->f_flags);
1553 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1554 struct lov_mds_md **lmmp, int *lmm_size,
1555 struct ptlrpc_request **request)
1557 struct ll_sb_info *sbi = ll_i2sbi(inode);
1558 struct mdt_body *body;
1559 struct lov_mds_md *lmm = NULL;
1560 struct ptlrpc_request *req = NULL;
1561 struct md_op_data *op_data;
1564 rc = ll_get_default_mdsize(sbi, &lmmsize);
1568 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1569 strlen(filename), lmmsize,
1570 LUSTRE_OPC_ANY, NULL);
1571 if (IS_ERR(op_data))
1572 RETURN(PTR_ERR(op_data));
1574 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1575 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1576 ll_finish_md_op_data(op_data);
1578 CDEBUG(D_INFO, "md_getattr_name failed "
1579 "on %s: rc %d\n", filename, rc);
1583 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1584 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1586 lmmsize = body->mbo_eadatasize;
1588 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1590 GOTO(out, rc = -ENODATA);
1593 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1594 LASSERT(lmm != NULL);
1596 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1597 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1598 GOTO(out, rc = -EPROTO);
1602 * This is coming from the MDS, so is probably in
1603 * little endian. We convert it to host endian before
1604 * passing it to userspace.
1606 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1609 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1610 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1613 /* if function called for directory - we should
1614 * avoid swab not existent lsm objects */
1615 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1616 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1617 if (S_ISREG(body->mbo_mode))
1618 lustre_swab_lov_user_md_objects(
1619 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1621 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1622 lustre_swab_lov_user_md_v3(
1623 (struct lov_user_md_v3 *)lmm);
1624 if (S_ISREG(body->mbo_mode))
1625 lustre_swab_lov_user_md_objects(
1626 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1633 *lmm_size = lmmsize;
1638 static int ll_lov_setea(struct inode *inode, struct file *file,
1641 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1642 struct lov_user_md *lump;
1643 int lum_size = sizeof(struct lov_user_md) +
1644 sizeof(struct lov_user_ost_data);
1648 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1651 OBD_ALLOC_LARGE(lump, lum_size);
1655 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size))
1656 GOTO(out_lump, rc = -EFAULT);
1658 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1661 OBD_FREE_LARGE(lump, lum_size);
1665 static int ll_file_getstripe(struct inode *inode,
1666 struct lov_user_md __user *lum)
1673 env = cl_env_get(&refcheck);
1675 RETURN(PTR_ERR(env));
1677 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum);
1678 cl_env_put(env, &refcheck);
1682 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1685 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
1686 struct lov_user_md *klum;
1688 __u64 flags = FMODE_WRITE;
1691 rc = ll_copy_user_md(lum, &klum);
1696 rc = ll_lov_setstripe_ea_info(inode, file, flags, klum, lum_size);
1700 put_user(0, &lum->lmm_stripe_count);
1702 ll_layout_refresh(inode, &gen);
1703 rc = ll_file_getstripe(inode, (struct lov_user_md __user *)arg);
1706 OBD_FREE(klum, lum_size);
1711 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1713 struct ll_inode_info *lli = ll_i2info(inode);
1714 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1715 struct ll_grouplock grouplock;
1720 CWARN("group id for group lock must not be 0\n");
1724 if (ll_file_nolock(file))
1725 RETURN(-EOPNOTSUPP);
1727 spin_lock(&lli->lli_lock);
1728 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1729 CWARN("group lock already existed with gid %lu\n",
1730 fd->fd_grouplock.lg_gid);
1731 spin_unlock(&lli->lli_lock);
1734 LASSERT(fd->fd_grouplock.lg_lock == NULL);
1735 spin_unlock(&lli->lli_lock);
1737 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1738 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1742 spin_lock(&lli->lli_lock);
1743 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1744 spin_unlock(&lli->lli_lock);
1745 CERROR("another thread just won the race\n");
1746 cl_put_grouplock(&grouplock);
1750 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1751 fd->fd_grouplock = grouplock;
1752 spin_unlock(&lli->lli_lock);
1754 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1758 static int ll_put_grouplock(struct inode *inode, struct file *file,
1761 struct ll_inode_info *lli = ll_i2info(inode);
1762 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1763 struct ll_grouplock grouplock;
1766 spin_lock(&lli->lli_lock);
1767 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1768 spin_unlock(&lli->lli_lock);
1769 CWARN("no group lock held\n");
1773 LASSERT(fd->fd_grouplock.lg_lock != NULL);
1775 if (fd->fd_grouplock.lg_gid != arg) {
1776 CWARN("group lock %lu doesn't match current id %lu\n",
1777 arg, fd->fd_grouplock.lg_gid);
1778 spin_unlock(&lli->lli_lock);
1782 grouplock = fd->fd_grouplock;
1783 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1784 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1785 spin_unlock(&lli->lli_lock);
1787 cl_put_grouplock(&grouplock);
1788 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1793 * Close inode open handle
1795 * \param dentry [in] dentry which contains the inode
1796 * \param it [in,out] intent which contains open info and result
1799 * \retval <0 failure
1801 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1803 struct inode *inode = dentry->d_inode;
1804 struct obd_client_handle *och;
1810 /* Root ? Do nothing. */
1811 if (dentry->d_inode->i_sb->s_root == dentry)
1814 /* No open handle to close? Move away */
1815 if (!it_disposition(it, DISP_OPEN_OPEN))
1818 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1820 OBD_ALLOC(och, sizeof(*och));
1822 GOTO(out, rc = -ENOMEM);
1824 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1826 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
1828 /* this one is in place of ll_file_open */
1829 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1830 ptlrpc_req_finished(it->it_request);
1831 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1837 * Get size for inode for which FIEMAP mapping is requested.
1838 * Make the FIEMAP get_info call and returns the result.
1839 * \param fiemap kernel buffer to hold extens
1840 * \param num_bytes kernel buffer size
1842 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
1848 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
1851 /* Checks for fiemap flags */
1852 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1853 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1857 /* Check for FIEMAP_FLAG_SYNC */
1858 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1859 rc = filemap_fdatawrite(inode->i_mapping);
1864 env = cl_env_get(&refcheck);
1866 RETURN(PTR_ERR(env));
1868 if (i_size_read(inode) == 0) {
1869 rc = ll_glimpse_size(inode);
1874 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1875 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
1876 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
1878 /* If filesize is 0, then there would be no objects for mapping */
1879 if (fmkey.lfik_oa.o_size == 0) {
1880 fiemap->fm_mapped_extents = 0;
1884 fmkey.lfik_fiemap = *fiemap;
1886 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
1887 &fmkey, fiemap, &num_bytes);
1889 cl_env_put(env, &refcheck);
1893 int ll_fid2path(struct inode *inode, void __user *arg)
1895 struct obd_export *exp = ll_i2mdexp(inode);
1896 const struct getinfo_fid2path __user *gfin = arg;
1898 struct getinfo_fid2path *gfout;
1904 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1905 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1908 /* Only need to get the buflen */
1909 if (get_user(pathlen, &gfin->gf_pathlen))
1912 if (pathlen > PATH_MAX)
1915 outsize = sizeof(*gfout) + pathlen;
1916 OBD_ALLOC(gfout, outsize);
1920 if (copy_from_user(gfout, arg, sizeof(*gfout)))
1921 GOTO(gf_free, rc = -EFAULT);
1922 /* append root FID after gfout to let MDT know the root FID so that it
1923 * can lookup the correct path, this is mainly for fileset.
1924 * old server without fileset mount support will ignore this. */
1925 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
1927 /* Call mdc_iocontrol */
1928 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1932 if (copy_to_user(arg, gfout, outsize))
1936 OBD_FREE(gfout, outsize);
1941 * Read the data_version for inode.
1943 * This value is computed using stripe object version on OST.
1944 * Version is computed using server side locking.
1946 * @param flags if do sync on the OST side;
1948 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1949 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1951 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1953 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1961 /* If no file object initialized, we consider its version is 0. */
1967 env = cl_env_get(&refcheck);
1969 RETURN(PTR_ERR(env));
1971 io = vvp_env_thread_io(env);
1973 io->u.ci_data_version.dv_data_version = 0;
1974 io->u.ci_data_version.dv_flags = flags;
1977 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
1978 result = cl_io_loop(env, io);
1980 result = io->ci_result;
1982 *data_version = io->u.ci_data_version.dv_data_version;
1984 cl_io_fini(env, io);
1986 if (unlikely(io->ci_need_restart))
1989 cl_env_put(env, &refcheck);
1995 * Trigger a HSM release request for the provided inode.
1997 int ll_hsm_release(struct inode *inode)
2000 struct obd_client_handle *och = NULL;
2001 __u64 data_version = 0;
2006 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2007 ll_get_fsname(inode->i_sb, NULL, 0),
2008 PFID(&ll_i2info(inode)->lli_fid));
2010 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2012 GOTO(out, rc = PTR_ERR(och));
2014 /* Grab latest data_version and [am]time values */
2015 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2019 env = cl_env_get(&refcheck);
2021 GOTO(out, rc = PTR_ERR(env));
2023 ll_merge_attr(env, inode);
2024 cl_env_put(env, &refcheck);
2026 /* Release the file.
2027 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2028 * we still need it to pack l_remote_handle to MDT. */
2029 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2035 if (och != NULL && !IS_ERR(och)) /* close the file */
2036 ll_lease_close(och, inode, NULL);
2041 struct ll_swap_stack {
2044 struct inode *inode1;
2045 struct inode *inode2;
2050 static int ll_swap_layouts(struct file *file1, struct file *file2,
2051 struct lustre_swap_layouts *lsl)
2053 struct mdc_swap_layouts msl;
2054 struct md_op_data *op_data;
2057 struct ll_swap_stack *llss = NULL;
2060 OBD_ALLOC_PTR(llss);
2064 llss->inode1 = file_inode(file1);
2065 llss->inode2 = file_inode(file2);
2067 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2071 /* we use 2 bool because it is easier to swap than 2 bits */
2072 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2073 llss->check_dv1 = true;
2075 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2076 llss->check_dv2 = true;
2078 /* we cannot use lsl->sl_dvX directly because we may swap them */
2079 llss->dv1 = lsl->sl_dv1;
2080 llss->dv2 = lsl->sl_dv2;
2082 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2083 if (rc == 0) /* same file, done! */
2086 if (rc < 0) { /* sequentialize it */
2087 swap(llss->inode1, llss->inode2);
2089 swap(llss->dv1, llss->dv2);
2090 swap(llss->check_dv1, llss->check_dv2);
2094 if (gid != 0) { /* application asks to flush dirty cache */
2095 rc = ll_get_grouplock(llss->inode1, file1, gid);
2099 rc = ll_get_grouplock(llss->inode2, file2, gid);
2101 ll_put_grouplock(llss->inode1, file1, gid);
2106 /* ultimate check, before swaping the layouts we check if
2107 * dataversion has changed (if requested) */
2108 if (llss->check_dv1) {
2109 rc = ll_data_version(llss->inode1, &dv, 0);
2112 if (dv != llss->dv1)
2113 GOTO(putgl, rc = -EAGAIN);
2116 if (llss->check_dv2) {
2117 rc = ll_data_version(llss->inode2, &dv, 0);
2120 if (dv != llss->dv2)
2121 GOTO(putgl, rc = -EAGAIN);
2124 /* struct md_op_data is used to send the swap args to the mdt
2125 * only flags is missing, so we use struct mdc_swap_layouts
2126 * through the md_op_data->op_data */
2127 /* flags from user space have to be converted before they are send to
2128 * server, no flag is sent today, they are only used on the client */
2131 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2132 0, LUSTRE_OPC_ANY, &msl);
2133 if (IS_ERR(op_data))
2134 GOTO(free, rc = PTR_ERR(op_data));
2136 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2137 sizeof(*op_data), op_data, NULL);
2138 ll_finish_md_op_data(op_data);
2145 ll_put_grouplock(llss->inode2, file2, gid);
2146 ll_put_grouplock(llss->inode1, file1, gid);
2156 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2158 struct md_op_data *op_data;
2162 /* Detect out-of range masks */
2163 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2166 /* Non-root users are forbidden to set or clear flags which are
2167 * NOT defined in HSM_USER_MASK. */
2168 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2169 !cfs_capable(CFS_CAP_SYS_ADMIN))
2172 /* Detect out-of range archive id */
2173 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2174 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
2177 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2178 LUSTRE_OPC_ANY, hss);
2179 if (IS_ERR(op_data))
2180 RETURN(PTR_ERR(op_data));
2182 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2183 sizeof(*op_data), op_data, NULL);
2185 ll_finish_md_op_data(op_data);
2190 static int ll_hsm_import(struct inode *inode, struct file *file,
2191 struct hsm_user_import *hui)
2193 struct hsm_state_set *hss = NULL;
2194 struct iattr *attr = NULL;
2198 if (!S_ISREG(inode->i_mode))
2204 GOTO(out, rc = -ENOMEM);
2206 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2207 hss->hss_archive_id = hui->hui_archive_id;
2208 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2209 rc = ll_hsm_state_set(inode, hss);
2213 OBD_ALLOC_PTR(attr);
2215 GOTO(out, rc = -ENOMEM);
2217 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2218 attr->ia_mode |= S_IFREG;
2219 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2220 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2221 attr->ia_size = hui->hui_size;
2222 attr->ia_mtime.tv_sec = hui->hui_mtime;
2223 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2224 attr->ia_atime.tv_sec = hui->hui_atime;
2225 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2227 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2228 ATTR_UID | ATTR_GID |
2229 ATTR_MTIME | ATTR_MTIME_SET |
2230 ATTR_ATIME | ATTR_ATIME_SET;
2234 rc = ll_setattr_raw(file_dentry(file), attr, true);
2238 inode_unlock(inode);
2250 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2252 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2253 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2256 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
2258 struct inode *inode = file_inode(file);
2260 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
2261 ATTR_MTIME | ATTR_MTIME_SET |
2262 ATTR_CTIME | ATTR_CTIME_SET,
2264 .tv_sec = lfu->lfu_atime_sec,
2265 .tv_nsec = lfu->lfu_atime_nsec,
2268 .tv_sec = lfu->lfu_mtime_sec,
2269 .tv_nsec = lfu->lfu_mtime_nsec,
2272 .tv_sec = lfu->lfu_ctime_sec,
2273 .tv_nsec = lfu->lfu_ctime_nsec,
2279 if (!capable(CAP_SYS_ADMIN))
2282 if (!S_ISREG(inode->i_mode))
2286 rc = ll_setattr_raw(file_dentry(file), &ia, false);
2287 inode_unlock(inode);
2293 * Give file access advices
2295 * The ladvise interface is similar to Linux fadvise() system call, except it
2296 * forwards the advices directly from Lustre client to server. The server side
2297 * codes will apply appropriate read-ahead and caching techniques for the
2298 * corresponding files.
2300 * A typical workload for ladvise is e.g. a bunch of different clients are
2301 * doing small random reads of a file, so prefetching pages into OSS cache
2302 * with big linear reads before the random IO is a net benefit. Fetching
2303 * all that data into each client cache with fadvise() may not be, due to
2304 * much more data being sent to the client.
2306 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
2307 struct llapi_lu_ladvise *ladvise)
2311 struct cl_ladvise_io *lio;
2316 env = cl_env_get(&refcheck);
2318 RETURN(PTR_ERR(env));
2320 io = vvp_env_thread_io(env);
2321 io->ci_obj = ll_i2info(inode)->lli_clob;
2323 /* initialize parameters for ladvise */
2324 lio = &io->u.ci_ladvise;
2325 lio->li_start = ladvise->lla_start;
2326 lio->li_end = ladvise->lla_end;
2327 lio->li_fid = ll_inode2fid(inode);
2328 lio->li_advice = ladvise->lla_advice;
2329 lio->li_flags = flags;
2331 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
2332 rc = cl_io_loop(env, io);
2336 cl_io_fini(env, io);
2337 cl_env_put(env, &refcheck);
2342 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2344 struct inode *inode = file_inode(file);
2345 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2349 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
2350 PFID(ll_inode2fid(inode)), inode, cmd);
2351 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2353 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2354 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2358 case LL_IOC_GETFLAGS:
2359 /* Get the current value of the file flags */
2360 return put_user(fd->fd_flags, (int __user *)arg);
2361 case LL_IOC_SETFLAGS:
2362 case LL_IOC_CLRFLAGS:
2363 /* Set or clear specific file flags */
2364 /* XXX This probably needs checks to ensure the flags are
2365 * not abused, and to handle any flag side effects.
2367 if (get_user(flags, (int __user *) arg))
2370 if (cmd == LL_IOC_SETFLAGS) {
2371 if ((flags & LL_FILE_IGNORE_LOCK) &&
2372 !(file->f_flags & O_DIRECT)) {
2373 CERROR("%s: unable to disable locking on "
2374 "non-O_DIRECT file\n", current->comm);
2378 fd->fd_flags |= flags;
2380 fd->fd_flags &= ~flags;
2383 case LL_IOC_LOV_SETSTRIPE:
2384 RETURN(ll_lov_setstripe(inode, file, arg));
2385 case LL_IOC_LOV_SETEA:
2386 RETURN(ll_lov_setea(inode, file, arg));
2387 case LL_IOC_LOV_SWAP_LAYOUTS: {
2389 struct lustre_swap_layouts lsl;
2391 if (copy_from_user(&lsl, (char __user *)arg,
2392 sizeof(struct lustre_swap_layouts)))
2395 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
2398 file2 = fget(lsl.sl_fd);
2402 /* O_WRONLY or O_RDWR */
2403 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
2404 GOTO(out, rc = -EPERM);
2406 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
2407 struct inode *inode2;
2408 struct ll_inode_info *lli;
2409 struct obd_client_handle *och = NULL;
2411 if (lsl.sl_flags != SWAP_LAYOUTS_CLOSE)
2412 GOTO(out, rc = -EINVAL);
2414 lli = ll_i2info(inode);
2415 mutex_lock(&lli->lli_och_mutex);
2416 if (fd->fd_lease_och != NULL) {
2417 och = fd->fd_lease_och;
2418 fd->fd_lease_och = NULL;
2420 mutex_unlock(&lli->lli_och_mutex);
2422 GOTO(out, rc = -ENOLCK);
2423 inode2 = file_inode(file2);
2424 rc = ll_swap_layouts_close(och, inode, inode2);
2426 rc = ll_swap_layouts(file, file2, &lsl);
2432 case LL_IOC_LOV_GETSTRIPE:
2433 RETURN(ll_file_getstripe(inode,
2434 (struct lov_user_md __user *)arg));
2435 case FSFILT_IOC_GETFLAGS:
2436 case FSFILT_IOC_SETFLAGS:
2437 RETURN(ll_iocontrol(inode, file, cmd, arg));
2438 case FSFILT_IOC_GETVERSION_OLD:
2439 case FSFILT_IOC_GETVERSION:
2440 RETURN(put_user(inode->i_generation, (int __user *)arg));
2441 case LL_IOC_GROUP_LOCK:
2442 RETURN(ll_get_grouplock(inode, file, arg));
2443 case LL_IOC_GROUP_UNLOCK:
2444 RETURN(ll_put_grouplock(inode, file, arg));
2445 case IOC_OBD_STATFS:
2446 RETURN(ll_obd_statfs(inode, (void __user *)arg));
2448 /* We need to special case any other ioctls we want to handle,
2449 * to send them to the MDS/OST as appropriate and to properly
2450 * network encode the arg field.
2451 case FSFILT_IOC_SETVERSION_OLD:
2452 case FSFILT_IOC_SETVERSION:
2454 case LL_IOC_FLUSHCTX:
2455 RETURN(ll_flush_ctx(inode));
2456 case LL_IOC_PATH2FID: {
2457 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2458 sizeof(struct lu_fid)))
2463 case LL_IOC_GETPARENT:
2464 RETURN(ll_getparent(file, (struct getparent __user *)arg));
2466 case OBD_IOC_FID2PATH:
2467 RETURN(ll_fid2path(inode, (void __user *)arg));
2468 case LL_IOC_DATA_VERSION: {
2469 struct ioc_data_version idv;
2472 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2475 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2476 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2479 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
2485 case LL_IOC_GET_MDTIDX: {
2488 mdtidx = ll_get_mdt_idx(inode);
2492 if (put_user((int)mdtidx, (int __user *)arg))
2497 case OBD_IOC_GETDTNAME:
2498 case OBD_IOC_GETMDNAME:
2499 RETURN(ll_get_obd_name(inode, cmd, arg));
2500 case LL_IOC_HSM_STATE_GET: {
2501 struct md_op_data *op_data;
2502 struct hsm_user_state *hus;
2509 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2510 LUSTRE_OPC_ANY, hus);
2511 if (IS_ERR(op_data)) {
2513 RETURN(PTR_ERR(op_data));
2516 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2519 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2522 ll_finish_md_op_data(op_data);
2526 case LL_IOC_HSM_STATE_SET: {
2527 struct hsm_state_set *hss;
2534 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
2539 rc = ll_hsm_state_set(inode, hss);
2544 case LL_IOC_HSM_ACTION: {
2545 struct md_op_data *op_data;
2546 struct hsm_current_action *hca;
2553 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2554 LUSTRE_OPC_ANY, hca);
2555 if (IS_ERR(op_data)) {
2557 RETURN(PTR_ERR(op_data));
2560 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2563 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2566 ll_finish_md_op_data(op_data);
2570 case LL_IOC_SET_LEASE: {
2571 struct ll_inode_info *lli = ll_i2info(inode);
2572 struct obd_client_handle *och = NULL;
2577 case LL_LEASE_WRLCK:
2578 if (!(file->f_mode & FMODE_WRITE))
2580 fmode = FMODE_WRITE;
2582 case LL_LEASE_RDLCK:
2583 if (!(file->f_mode & FMODE_READ))
2587 case LL_LEASE_UNLCK:
2588 mutex_lock(&lli->lli_och_mutex);
2589 if (fd->fd_lease_och != NULL) {
2590 och = fd->fd_lease_och;
2591 fd->fd_lease_och = NULL;
2593 mutex_unlock(&lli->lli_och_mutex);
2598 fmode = och->och_flags;
2599 rc = ll_lease_close(och, inode, &lease_broken);
2603 rc = ll_lease_och_release(inode, file);
2610 RETURN(ll_lease_type_from_fmode(fmode));
2615 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
2617 /* apply for lease */
2618 och = ll_lease_open(inode, file, fmode, 0);
2620 RETURN(PTR_ERR(och));
2623 mutex_lock(&lli->lli_och_mutex);
2624 if (fd->fd_lease_och == NULL) {
2625 fd->fd_lease_och = och;
2628 mutex_unlock(&lli->lli_och_mutex);
2630 /* impossible now that only excl is supported for now */
2631 ll_lease_close(och, inode, &lease_broken);
2636 case LL_IOC_GET_LEASE: {
2637 struct ll_inode_info *lli = ll_i2info(inode);
2638 struct ldlm_lock *lock = NULL;
2641 mutex_lock(&lli->lli_och_mutex);
2642 if (fd->fd_lease_och != NULL) {
2643 struct obd_client_handle *och = fd->fd_lease_och;
2645 lock = ldlm_handle2lock(&och->och_lease_handle);
2647 lock_res_and_lock(lock);
2648 if (!ldlm_is_cancel(lock))
2649 fmode = och->och_flags;
2651 unlock_res_and_lock(lock);
2652 LDLM_LOCK_PUT(lock);
2655 mutex_unlock(&lli->lli_och_mutex);
2657 RETURN(ll_lease_type_from_fmode(fmode));
2659 case LL_IOC_HSM_IMPORT: {
2660 struct hsm_user_import *hui;
2666 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
2671 rc = ll_hsm_import(inode, file, hui);
2676 case LL_IOC_FUTIMES_3: {
2677 struct ll_futimes_3 lfu;
2679 if (copy_from_user(&lfu,
2680 (const struct ll_futimes_3 __user *)arg,
2684 RETURN(ll_file_futimes_3(file, &lfu));
2686 case LL_IOC_LADVISE: {
2687 struct llapi_ladvise_hdr *ladvise_hdr;
2690 int alloc_size = sizeof(*ladvise_hdr);
2693 OBD_ALLOC_PTR(ladvise_hdr);
2694 if (ladvise_hdr == NULL)
2697 if (copy_from_user(ladvise_hdr,
2698 (const struct llapi_ladvise_hdr __user *)arg,
2700 GOTO(out_ladvise, rc = -EFAULT);
2702 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2703 ladvise_hdr->lah_count < 1)
2704 GOTO(out_ladvise, rc = -EINVAL);
2706 num_advise = ladvise_hdr->lah_count;
2707 if (num_advise >= LAH_COUNT_MAX)
2708 GOTO(out_ladvise, rc = -EFBIG);
2710 OBD_FREE_PTR(ladvise_hdr);
2711 alloc_size = offsetof(typeof(*ladvise_hdr),
2712 lah_advise[num_advise]);
2713 OBD_ALLOC(ladvise_hdr, alloc_size);
2714 if (ladvise_hdr == NULL)
2718 * TODO: submit multiple advices to one server in a single RPC
2720 if (copy_from_user(ladvise_hdr,
2721 (const struct llapi_ladvise_hdr __user *)arg,
2723 GOTO(out_ladvise, rc = -EFAULT);
2725 for (i = 0; i < num_advise; i++) {
2726 rc = ll_ladvise(inode, file, ladvise_hdr->lah_flags,
2727 &ladvise_hdr->lah_advise[i]);
2733 OBD_FREE(ladvise_hdr, alloc_size);
2740 ll_iocontrol_call(inode, file, cmd, arg, &err))
2743 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2744 (void __user *)arg));
2749 #ifndef HAVE_FILE_LLSEEK_SIZE
2750 static inline loff_t
2751 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2753 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2755 if (offset > maxsize)
2758 if (offset != file->f_pos) {
2759 file->f_pos = offset;
2760 file->f_version = 0;
2766 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2767 loff_t maxsize, loff_t eof)
2769 struct inode *inode = file_inode(file);
2777 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2778 * position-querying operation. Avoid rewriting the "same"
2779 * f_pos value back to the file because a concurrent read(),
2780 * write() or lseek() might have altered it
2785 * f_lock protects against read/modify/write race with other
2786 * SEEK_CURs. Note that parallel writes and reads behave
2790 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2791 inode_unlock(inode);
2795 * In the generic case the entire file is data, so as long as
2796 * offset isn't at the end of the file then the offset is data.
2803 * There is a virtual hole at the end of the file, so as long as
2804 * offset isn't i_size or larger, return i_size.
2812 return llseek_execute(file, offset, maxsize);
2816 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2818 struct inode *inode = file_inode(file);
2819 loff_t retval, eof = 0;
2822 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2823 (origin == SEEK_CUR) ? file->f_pos : 0);
2824 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2825 PFID(ll_inode2fid(inode)), inode, retval, retval,
2827 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2829 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2830 retval = ll_glimpse_size(inode);
2833 eof = i_size_read(inode);
2836 retval = ll_generic_file_llseek_size(file, offset, origin,
2837 ll_file_maxbytes(inode), eof);
2841 static int ll_flush(struct file *file, fl_owner_t id)
2843 struct inode *inode = file_inode(file);
2844 struct ll_inode_info *lli = ll_i2info(inode);
2845 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2848 LASSERT(!S_ISDIR(inode->i_mode));
2850 /* catch async errors that were recorded back when async writeback
2851 * failed for pages in this mapping. */
2852 rc = lli->lli_async_rc;
2853 lli->lli_async_rc = 0;
2854 if (lli->lli_clob != NULL) {
2855 err = lov_read_and_clear_async_rc(lli->lli_clob);
2860 /* The application has been told write failure already.
2861 * Do not report failure again. */
2862 if (fd->fd_write_failed)
2864 return rc ? -EIO : 0;
2868 * Called to make sure a portion of file has been written out.
2869 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2871 * Return how many pages have been written.
2873 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2874 enum cl_fsync_mode mode, int ignore_layout)
2878 struct cl_fsync_io *fio;
2883 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2884 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2887 env = cl_env_get(&refcheck);
2889 RETURN(PTR_ERR(env));
2891 io = vvp_env_thread_io(env);
2892 io->ci_obj = ll_i2info(inode)->lli_clob;
2893 io->ci_ignore_layout = ignore_layout;
2895 /* initialize parameters for sync */
2896 fio = &io->u.ci_fsync;
2897 fio->fi_start = start;
2899 fio->fi_fid = ll_inode2fid(inode);
2900 fio->fi_mode = mode;
2901 fio->fi_nr_written = 0;
2903 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2904 result = cl_io_loop(env, io);
2906 result = io->ci_result;
2908 result = fio->fi_nr_written;
2909 cl_io_fini(env, io);
2910 cl_env_put(env, &refcheck);
2916 * When dentry is provided (the 'else' case), file_dentry() may be
2917 * null and dentry must be used directly rather than pulled from
2918 * file_dentry() as is done otherwise.
2921 #ifdef HAVE_FILE_FSYNC_4ARGS
2922 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2924 struct dentry *dentry = file_dentry(file);
2925 #elif defined(HAVE_FILE_FSYNC_2ARGS)
2926 int ll_fsync(struct file *file, int datasync)
2928 struct dentry *dentry = file_dentry(file);
2930 loff_t end = LLONG_MAX;
2932 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
2935 loff_t end = LLONG_MAX;
2937 struct inode *inode = dentry->d_inode;
2938 struct ll_inode_info *lli = ll_i2info(inode);
2939 struct ptlrpc_request *req;
2943 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2944 PFID(ll_inode2fid(inode)), inode);
2945 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2947 #ifdef HAVE_FILE_FSYNC_4ARGS
2948 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2951 /* fsync's caller has already called _fdata{sync,write}, we want
2952 * that IO to finish before calling the osc and mdc sync methods */
2953 rc = filemap_fdatawait(inode->i_mapping);
2956 /* catch async errors that were recorded back when async writeback
2957 * failed for pages in this mapping. */
2958 if (!S_ISDIR(inode->i_mode)) {
2959 err = lli->lli_async_rc;
2960 lli->lli_async_rc = 0;
2963 if (lli->lli_clob != NULL) {
2964 err = lov_read_and_clear_async_rc(lli->lli_clob);
2970 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
2974 ptlrpc_req_finished(req);
2976 if (S_ISREG(inode->i_mode)) {
2977 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2979 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2980 if (rc == 0 && err < 0)
2983 fd->fd_write_failed = true;
2985 fd->fd_write_failed = false;
2988 #ifdef HAVE_FILE_FSYNC_4ARGS
2989 inode_unlock(inode);
2995 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2997 struct inode *inode = file_inode(file);
2998 struct ll_sb_info *sbi = ll_i2sbi(inode);
2999 struct ldlm_enqueue_info einfo = {
3000 .ei_type = LDLM_FLOCK,
3001 .ei_cb_cp = ldlm_flock_completion_ast,
3002 .ei_cbdata = file_lock,
3004 struct md_op_data *op_data;
3005 struct lustre_handle lockh = { 0 };
3006 union ldlm_policy_data flock = { { 0 } };
3007 int fl_type = file_lock->fl_type;
3013 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
3014 PFID(ll_inode2fid(inode)), file_lock);
3016 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
3018 if (file_lock->fl_flags & FL_FLOCK) {
3019 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
3020 /* flocks are whole-file locks */
3021 flock.l_flock.end = OFFSET_MAX;
3022 /* For flocks owner is determined by the local file desctiptor*/
3023 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
3024 } else if (file_lock->fl_flags & FL_POSIX) {
3025 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
3026 flock.l_flock.start = file_lock->fl_start;
3027 flock.l_flock.end = file_lock->fl_end;
3031 flock.l_flock.pid = file_lock->fl_pid;
3033 /* Somewhat ugly workaround for svc lockd.
3034 * lockd installs custom fl_lmops->lm_compare_owner that checks
3035 * for the fl_owner to be the same (which it always is on local node
3036 * I guess between lockd processes) and then compares pid.
3037 * As such we assign pid to the owner field to make it all work,
3038 * conflict with normal locks is unlikely since pid space and
3039 * pointer space for current->files are not intersecting */
3040 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
3041 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
3045 einfo.ei_mode = LCK_PR;
3048 /* An unlock request may or may not have any relation to
3049 * existing locks so we may not be able to pass a lock handle
3050 * via a normal ldlm_lock_cancel() request. The request may even
3051 * unlock a byte range in the middle of an existing lock. In
3052 * order to process an unlock request we need all of the same
3053 * information that is given with a normal read or write record
3054 * lock request. To avoid creating another ldlm unlock (cancel)
3055 * message we'll treat a LCK_NL flock request as an unlock. */
3056 einfo.ei_mode = LCK_NL;
3059 einfo.ei_mode = LCK_PW;
3062 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
3077 flags = LDLM_FL_BLOCK_NOWAIT;
3083 flags = LDLM_FL_TEST_LOCK;
3086 CERROR("unknown fcntl lock command: %d\n", cmd);
3090 /* Save the old mode so that if the mode in the lock changes we
3091 * can decrement the appropriate reader or writer refcount. */
3092 file_lock->fl_type = einfo.ei_mode;
3094 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3095 LUSTRE_OPC_ANY, NULL);
3096 if (IS_ERR(op_data))
3097 RETURN(PTR_ERR(op_data));
3099 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
3100 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
3101 flock.l_flock.pid, flags, einfo.ei_mode,
3102 flock.l_flock.start, flock.l_flock.end);
3104 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
3107 /* Restore the file lock type if not TEST lock. */
3108 if (!(flags & LDLM_FL_TEST_LOCK))
3109 file_lock->fl_type = fl_type;
3111 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
3112 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
3113 !(flags & LDLM_FL_TEST_LOCK))
3114 rc2 = locks_lock_file_wait(file, file_lock);
3116 if ((file_lock->fl_flags & FL_FLOCK) &&
3117 (rc == 0 || file_lock->fl_type == F_UNLCK))
3118 rc2 = flock_lock_file_wait(file, file_lock);
3119 if ((file_lock->fl_flags & FL_POSIX) &&
3120 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
3121 !(flags & LDLM_FL_TEST_LOCK))
3122 rc2 = posix_lock_file_wait(file, file_lock);
3123 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
3125 if (rc2 && file_lock->fl_type != F_UNLCK) {
3126 einfo.ei_mode = LCK_NL;
3127 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
3132 ll_finish_md_op_data(op_data);
3137 int ll_get_fid_by_name(struct inode *parent, const char *name,
3138 int namelen, struct lu_fid *fid,
3139 struct inode **inode)
3141 struct md_op_data *op_data = NULL;
3142 struct mdt_body *body;
3143 struct ptlrpc_request *req;
3147 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
3148 LUSTRE_OPC_ANY, NULL);
3149 if (IS_ERR(op_data))
3150 RETURN(PTR_ERR(op_data));
3152 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
3153 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
3154 ll_finish_md_op_data(op_data);
3158 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3160 GOTO(out_req, rc = -EFAULT);
3162 *fid = body->mbo_fid1;
3165 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
3167 ptlrpc_req_finished(req);
3171 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
3172 const char *name, int namelen)
3174 struct dentry *dchild = NULL;
3175 struct inode *child_inode = NULL;
3176 struct md_op_data *op_data;
3177 struct ptlrpc_request *request = NULL;
3178 struct obd_client_handle *och = NULL;
3180 struct mdt_body *body;
3182 __u64 data_version = 0;
3185 CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%04x\n",
3186 name, PFID(ll_inode2fid(parent)), mdtidx);
3188 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
3189 0, LUSTRE_OPC_ANY, NULL);
3190 if (IS_ERR(op_data))
3191 RETURN(PTR_ERR(op_data));
3193 /* Get child FID first */
3194 qstr.hash = full_name_hash(name, namelen);
3197 dchild = d_lookup(file_dentry(file), &qstr);
3198 if (dchild != NULL) {
3199 if (dchild->d_inode != NULL)
3200 child_inode = igrab(dchild->d_inode);
3204 if (child_inode == NULL) {
3205 rc = ll_get_fid_by_name(parent, name, namelen,
3206 &op_data->op_fid3, &child_inode);
3211 if (child_inode == NULL)
3212 GOTO(out_free, rc = -EINVAL);
3215 * lfs migrate command needs to be blocked on the client
3216 * by checking the migrate FID against the FID of the
3219 if (child_inode == parent->i_sb->s_root->d_inode)
3220 GOTO(out_iput, rc = -EINVAL);
3222 inode_lock(child_inode);
3223 op_data->op_fid3 = *ll_inode2fid(child_inode);
3224 if (!fid_is_sane(&op_data->op_fid3)) {
3225 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
3226 ll_get_fsname(parent->i_sb, NULL, 0), name,
3227 PFID(&op_data->op_fid3));
3228 GOTO(out_unlock, rc = -EINVAL);
3231 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
3233 GOTO(out_unlock, rc);
3236 CDEBUG(D_INFO, "%s: "DFID" is already on MDT%04x\n", name,
3237 PFID(&op_data->op_fid3), mdtidx);
3238 GOTO(out_unlock, rc = 0);
3241 if (S_ISREG(child_inode->i_mode)) {
3242 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
3246 GOTO(out_unlock, rc);
3249 rc = ll_data_version(child_inode, &data_version,
3252 GOTO(out_close, rc);
3254 op_data->op_handle = och->och_fh;
3255 op_data->op_data = och->och_mod;
3256 op_data->op_data_version = data_version;
3257 op_data->op_lease_handle = och->och_lease_handle;
3258 op_data->op_bias |= MDS_RENAME_MIGRATE;
3261 op_data->op_mds = mdtidx;
3262 op_data->op_cli_flags = CLI_MIGRATE;
3263 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
3264 namelen, name, namelen, &request);
3266 LASSERT(request != NULL);
3267 ll_update_times(request, parent);
3269 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
3270 LASSERT(body != NULL);
3272 /* If the server does release layout lock, then we cleanup
3273 * the client och here, otherwise release it in out_close: */
3275 body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
3276 obd_mod_put(och->och_mod);
3277 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
3279 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
3285 if (request != NULL) {
3286 ptlrpc_req_finished(request);
3290 /* Try again if the file layout has changed. */
3291 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
3295 if (och != NULL) /* close the file */
3296 ll_lease_close(och, child_inode, NULL);
3298 clear_nlink(child_inode);
3300 inode_unlock(child_inode);
3304 ll_finish_md_op_data(op_data);
3309 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
3317 * test if some locks matching bits and l_req_mode are acquired
3318 * - bits can be in different locks
3319 * - if found clear the common lock bits in *bits
3320 * - the bits not found, are kept in *bits
3322 * \param bits [IN] searched lock bits [IN]
3323 * \param l_req_mode [IN] searched lock mode
3324 * \retval boolean, true iff all bits are found
3326 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
3328 struct lustre_handle lockh;
3329 union ldlm_policy_data policy;
3330 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
3331 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
3340 fid = &ll_i2info(inode)->lli_fid;
3341 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
3342 ldlm_lockname[mode]);
3344 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
3345 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
3346 policy.l_inodebits.bits = *bits & (1 << i);
3347 if (policy.l_inodebits.bits == 0)
3350 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
3351 &policy, mode, &lockh)) {
3352 struct ldlm_lock *lock;
3354 lock = ldlm_handle2lock(&lockh);
3357 ~(lock->l_policy_data.l_inodebits.bits);
3358 LDLM_LOCK_PUT(lock);
3360 *bits &= ~policy.l_inodebits.bits;
3367 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
3368 struct lustre_handle *lockh, __u64 flags,
3369 enum ldlm_mode mode)
3371 union ldlm_policy_data policy = { .l_inodebits = { bits } };
3376 fid = &ll_i2info(inode)->lli_fid;
3377 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
3379 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
3380 fid, LDLM_IBITS, &policy, mode, lockh);
3385 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
3387 /* Already unlinked. Just update nlink and return success */
3388 if (rc == -ENOENT) {
3390 /* If it is striped directory, and there is bad stripe
3391 * Let's revalidate the dentry again, instead of returning
3393 if (S_ISDIR(inode->i_mode) &&
3394 ll_i2info(inode)->lli_lsm_md != NULL)
3397 /* This path cannot be hit for regular files unless in
3398 * case of obscure races, so no need to to validate
3400 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
3402 } else if (rc != 0) {
3403 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
3404 "%s: revalidate FID "DFID" error: rc = %d\n",
3405 ll_get_fsname(inode->i_sb, NULL, 0),
3406 PFID(ll_inode2fid(inode)), rc);
3412 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3414 struct inode *inode = dentry->d_inode;
3415 struct ptlrpc_request *req = NULL;
3416 struct obd_export *exp;
3420 LASSERT(inode != NULL);
3422 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
3423 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
3425 exp = ll_i2mdexp(inode);
3427 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
3428 * But under CMD case, it caused some lock issues, should be fixed
3429 * with new CMD ibits lock. See bug 12718 */
3430 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
3431 struct lookup_intent oit = { .it_op = IT_GETATTR };
3432 struct md_op_data *op_data;
3434 if (ibits == MDS_INODELOCK_LOOKUP)
3435 oit.it_op = IT_LOOKUP;
3437 /* Call getattr by fid, so do not provide name at all. */
3438 op_data = ll_prep_md_op_data(NULL, dentry->d_inode,
3439 dentry->d_inode, NULL, 0, 0,
3440 LUSTRE_OPC_ANY, NULL);
3441 if (IS_ERR(op_data))
3442 RETURN(PTR_ERR(op_data));
3444 rc = md_intent_lock(exp, op_data, &oit, &req,
3445 &ll_md_blocking_ast, 0);
3446 ll_finish_md_op_data(op_data);
3448 rc = ll_inode_revalidate_fini(inode, rc);
3452 rc = ll_revalidate_it_finish(req, &oit, dentry);
3454 ll_intent_release(&oit);
3458 /* Unlinked? Unhash dentry, so it is not picked up later by
3459 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
3460 here to preserve get_cwd functionality on 2.6.
3462 if (!dentry->d_inode->i_nlink) {
3463 ll_lock_dcache(inode);
3464 d_lustre_invalidate(dentry, 0);
3465 ll_unlock_dcache(inode);
3468 ll_lookup_finish_locks(&oit, dentry);
3469 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
3470 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
3471 u64 valid = OBD_MD_FLGETATTR;
3472 struct md_op_data *op_data;
3475 if (S_ISREG(inode->i_mode)) {
3476 rc = ll_get_default_mdsize(sbi, &ealen);
3479 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
3482 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
3483 0, ealen, LUSTRE_OPC_ANY,
3485 if (IS_ERR(op_data))
3486 RETURN(PTR_ERR(op_data));
3488 op_data->op_valid = valid;
3489 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3490 ll_finish_md_op_data(op_data);
3492 rc = ll_inode_revalidate_fini(inode, rc);
3496 rc = ll_prep_inode(&inode, req, NULL, NULL);
3499 ptlrpc_req_finished(req);
3503 static int ll_merge_md_attr(struct inode *inode)
3505 struct cl_attr attr = { 0 };
3508 LASSERT(ll_i2info(inode)->lli_lsm_md != NULL);
3509 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
3510 &attr, ll_md_blocking_ast);
3514 set_nlink(inode, attr.cat_nlink);
3515 inode->i_blocks = attr.cat_blocks;
3516 i_size_write(inode, attr.cat_size);
3518 ll_i2info(inode)->lli_atime = attr.cat_atime;
3519 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
3520 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
3526 ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3528 struct inode *inode = dentry->d_inode;
3532 rc = __ll_inode_revalidate(dentry, ibits);
3536 /* if object isn't regular file, don't validate size */
3537 if (!S_ISREG(inode->i_mode)) {
3538 if (S_ISDIR(inode->i_mode) &&
3539 ll_i2info(inode)->lli_lsm_md != NULL) {
3540 rc = ll_merge_md_attr(inode);
3545 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
3546 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
3547 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
3549 /* In case of restore, the MDT has the right size and has
3550 * already send it back without granting the layout lock,
3551 * inode is up-to-date so glimpse is useless.
3552 * Also to glimpse we need the layout, in case of a running
3553 * restore the MDT holds the layout lock so the glimpse will
3554 * block up to the end of restore (getattr will block)
3556 if (!ll_file_test_flag(ll_i2info(inode), LLIF_FILE_RESTORING))
3557 rc = ll_glimpse_size(inode);
3562 static inline dev_t ll_compat_encode_dev(dev_t dev)
3564 /* The compat_sys_*stat*() syscalls will fail unless the
3565 * device majors and minors are both less than 256. Note that
3566 * the value returned here will be passed through
3567 * old_encode_dev() in cp_compat_stat(). And so we are not
3568 * trying to return a valid compat (u16) device number, just
3569 * one that will pass the old_valid_dev() check. */
3571 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
3574 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3576 struct inode *inode = de->d_inode;
3577 struct ll_sb_info *sbi = ll_i2sbi(inode);
3578 struct ll_inode_info *lli = ll_i2info(inode);
3581 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3582 MDS_INODELOCK_LOOKUP);
3583 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3588 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
3590 if (ll_need_32bit_api(sbi)) {
3591 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3592 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
3593 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
3595 stat->ino = inode->i_ino;
3596 stat->dev = inode->i_sb->s_dev;
3597 stat->rdev = inode->i_rdev;
3600 stat->mode = inode->i_mode;
3601 stat->uid = inode->i_uid;
3602 stat->gid = inode->i_gid;
3603 stat->atime = inode->i_atime;
3604 stat->mtime = inode->i_mtime;
3605 stat->ctime = inode->i_ctime;
3606 stat->blksize = 1 << inode->i_blkbits;
3608 stat->nlink = inode->i_nlink;
3609 stat->size = i_size_read(inode);
3610 stat->blocks = inode->i_blocks;
3615 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3616 __u64 start, __u64 len)
3620 struct fiemap *fiemap;
3621 unsigned int extent_count = fieinfo->fi_extents_max;
3623 num_bytes = sizeof(*fiemap) + (extent_count *
3624 sizeof(struct fiemap_extent));
3625 OBD_ALLOC_LARGE(fiemap, num_bytes);
3630 fiemap->fm_flags = fieinfo->fi_flags;
3631 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3632 fiemap->fm_start = start;
3633 fiemap->fm_length = len;
3634 if (extent_count > 0 &&
3635 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3636 sizeof(struct fiemap_extent)) != 0)
3637 GOTO(out, rc = -EFAULT);
3639 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3641 fieinfo->fi_flags = fiemap->fm_flags;
3642 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3643 if (extent_count > 0 &&
3644 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3645 fiemap->fm_mapped_extents *
3646 sizeof(struct fiemap_extent)) != 0)
3647 GOTO(out, rc = -EFAULT);
3649 OBD_FREE_LARGE(fiemap, num_bytes);
3653 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3655 struct ll_inode_info *lli = ll_i2info(inode);
3656 struct posix_acl *acl = NULL;
3659 spin_lock(&lli->lli_lock);
3660 /* VFS' acl_permission_check->check_acl will release the refcount */
3661 acl = posix_acl_dup(lli->lli_posix_acl);
3662 spin_unlock(&lli->lli_lock);
3667 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
3669 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3670 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
3672 ll_check_acl(struct inode *inode, int mask)
3675 # ifdef CONFIG_FS_POSIX_ACL
3676 struct posix_acl *acl;
3680 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3681 if (flags & IPERM_FLAG_RCU)
3684 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
3689 rc = posix_acl_permission(inode, acl, mask);
3690 posix_acl_release(acl);
3693 # else /* !CONFIG_FS_POSIX_ACL */
3695 # endif /* CONFIG_FS_POSIX_ACL */
3697 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
3699 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
3700 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
3702 # ifdef HAVE_INODE_PERMISION_2ARGS
3703 int ll_inode_permission(struct inode *inode, int mask)
3705 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
3710 struct ll_sb_info *sbi;
3711 struct root_squash_info *squash;
3712 struct cred *cred = NULL;
3713 const struct cred *old_cred = NULL;
3715 bool squash_id = false;
3718 #ifdef MAY_NOT_BLOCK
3719 if (mask & MAY_NOT_BLOCK)
3721 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
3722 if (flags & IPERM_FLAG_RCU)
3726 /* as root inode are NOT getting validated in lookup operation,
3727 * need to do it before permission check. */
3729 if (inode == inode->i_sb->s_root->d_inode) {
3730 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3731 MDS_INODELOCK_LOOKUP);
3736 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3737 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3739 /* squash fsuid/fsgid if needed */
3740 sbi = ll_i2sbi(inode);
3741 squash = &sbi->ll_squash;
3742 if (unlikely(squash->rsi_uid != 0 &&
3743 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3744 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3748 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3749 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3750 squash->rsi_uid, squash->rsi_gid);
3752 /* update current process's credentials
3753 * and FS capability */
3754 cred = prepare_creds();
3758 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3759 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3760 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3761 if ((1 << cap) & CFS_CAP_FS_MASK)
3762 cap_lower(cred->cap_effective, cap);
3764 old_cred = override_creds(cred);
3767 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
3768 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
3769 /* restore current process's credentials and FS capability */
3771 revert_creds(old_cred);
3778 /* -o localflock - only provides locally consistent flock locks */
3779 struct file_operations ll_file_operations = {
3780 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
3781 # ifdef HAVE_SYNC_READ_WRITE
3782 .read = new_sync_read,
3783 .write = new_sync_write,
3785 .read_iter = ll_file_read_iter,
3786 .write_iter = ll_file_write_iter,
3787 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3788 .read = ll_file_read,
3789 .aio_read = ll_file_aio_read,
3790 .write = ll_file_write,
3791 .aio_write = ll_file_aio_write,
3792 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3793 .unlocked_ioctl = ll_file_ioctl,
3794 .open = ll_file_open,
3795 .release = ll_file_release,
3796 .mmap = ll_file_mmap,
3797 .llseek = ll_file_seek,
3798 .splice_read = ll_file_splice_read,
3803 struct file_operations ll_file_operations_flock = {
3804 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
3805 # ifdef HAVE_SYNC_READ_WRITE
3806 .read = new_sync_read,
3807 .write = new_sync_write,
3808 # endif /* HAVE_SYNC_READ_WRITE */
3809 .read_iter = ll_file_read_iter,
3810 .write_iter = ll_file_write_iter,
3811 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3812 .read = ll_file_read,
3813 .aio_read = ll_file_aio_read,
3814 .write = ll_file_write,
3815 .aio_write = ll_file_aio_write,
3816 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3817 .unlocked_ioctl = ll_file_ioctl,
3818 .open = ll_file_open,
3819 .release = ll_file_release,
3820 .mmap = ll_file_mmap,
3821 .llseek = ll_file_seek,
3822 .splice_read = ll_file_splice_read,
3825 .flock = ll_file_flock,
3826 .lock = ll_file_flock
3829 /* These are for -o noflock - to return ENOSYS on flock calls */
3830 struct file_operations ll_file_operations_noflock = {
3831 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
3832 # ifdef HAVE_SYNC_READ_WRITE
3833 .read = new_sync_read,
3834 .write = new_sync_write,
3835 # endif /* HAVE_SYNC_READ_WRITE */
3836 .read_iter = ll_file_read_iter,
3837 .write_iter = ll_file_write_iter,
3838 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3839 .read = ll_file_read,
3840 .aio_read = ll_file_aio_read,
3841 .write = ll_file_write,
3842 .aio_write = ll_file_aio_write,
3843 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3844 .unlocked_ioctl = ll_file_ioctl,
3845 .open = ll_file_open,
3846 .release = ll_file_release,
3847 .mmap = ll_file_mmap,
3848 .llseek = ll_file_seek,
3849 .splice_read = ll_file_splice_read,
3852 .flock = ll_file_noflock,
3853 .lock = ll_file_noflock
3856 struct inode_operations ll_file_inode_operations = {
3857 .setattr = ll_setattr,
3858 .getattr = ll_getattr,
3859 .permission = ll_inode_permission,
3860 .setxattr = ll_setxattr,
3861 .getxattr = ll_getxattr,
3862 .listxattr = ll_listxattr,
3863 .removexattr = ll_removexattr,
3864 .fiemap = ll_fiemap,
3865 #ifdef HAVE_IOP_GET_ACL
3866 .get_acl = ll_get_acl,
3870 /* dynamic ioctl number support routins */
3871 static struct llioc_ctl_data {
3872 struct rw_semaphore ioc_sem;
3873 struct list_head ioc_head;
3875 __RWSEM_INITIALIZER(llioc.ioc_sem),
3876 LIST_HEAD_INIT(llioc.ioc_head)
3881 struct list_head iocd_list;
3882 unsigned int iocd_size;
3883 llioc_callback_t iocd_cb;
3884 unsigned int iocd_count;
3885 unsigned int iocd_cmd[0];
3888 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3891 struct llioc_data *in_data = NULL;
3894 if (cb == NULL || cmd == NULL ||
3895 count > LLIOC_MAX_CMD || count < 0)
3898 size = sizeof(*in_data) + count * sizeof(unsigned int);
3899 OBD_ALLOC(in_data, size);
3900 if (in_data == NULL)
3903 memset(in_data, 0, sizeof(*in_data));
3904 in_data->iocd_size = size;
3905 in_data->iocd_cb = cb;
3906 in_data->iocd_count = count;
3907 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3909 down_write(&llioc.ioc_sem);
3910 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3911 up_write(&llioc.ioc_sem);
3916 void ll_iocontrol_unregister(void *magic)
3918 struct llioc_data *tmp;
3923 down_write(&llioc.ioc_sem);
3924 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3926 unsigned int size = tmp->iocd_size;
3928 list_del(&tmp->iocd_list);
3929 up_write(&llioc.ioc_sem);
3931 OBD_FREE(tmp, size);
3935 up_write(&llioc.ioc_sem);
3937 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3940 EXPORT_SYMBOL(ll_iocontrol_register);
3941 EXPORT_SYMBOL(ll_iocontrol_unregister);
3943 static enum llioc_iter
3944 ll_iocontrol_call(struct inode *inode, struct file *file,
3945 unsigned int cmd, unsigned long arg, int *rcp)
3947 enum llioc_iter ret = LLIOC_CONT;
3948 struct llioc_data *data;
3949 int rc = -EINVAL, i;
3951 down_read(&llioc.ioc_sem);
3952 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3953 for (i = 0; i < data->iocd_count; i++) {
3954 if (cmd != data->iocd_cmd[i])
3957 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3961 if (ret == LLIOC_STOP)
3964 up_read(&llioc.ioc_sem);
3971 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3973 struct ll_inode_info *lli = ll_i2info(inode);
3974 struct cl_object *obj = lli->lli_clob;
3983 env = cl_env_get(&refcheck);
3985 RETURN(PTR_ERR(env));
3987 rc = cl_conf_set(env, lli->lli_clob, conf);
3991 if (conf->coc_opc == OBJECT_CONF_SET) {
3992 struct ldlm_lock *lock = conf->coc_lock;
3993 struct cl_layout cl = {
3997 LASSERT(lock != NULL);
3998 LASSERT(ldlm_has_layout(lock));
4000 /* it can only be allowed to match after layout is
4001 * applied to inode otherwise false layout would be
4002 * seen. Applying layout shoud happen before dropping
4003 * the intent lock. */
4004 ldlm_lock_allow_match(lock);
4006 rc = cl_object_layout_get(env, obj, &cl);
4011 DFID": layout version change: %u -> %u\n",
4012 PFID(&lli->lli_fid), ll_layout_version_get(lli),
4014 ll_layout_version_set(lli, cl.cl_layout_gen);
4018 cl_env_put(env, &refcheck);
4023 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
4024 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
4027 struct ll_sb_info *sbi = ll_i2sbi(inode);
4028 struct ptlrpc_request *req;
4029 struct mdt_body *body;
4036 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
4037 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
4038 lock->l_lvb_data, lock->l_lvb_len);
4040 if (lock->l_lvb_data != NULL)
4043 /* if layout lock was granted right away, the layout is returned
4044 * within DLM_LVB of dlm reply; otherwise if the lock was ever
4045 * blocked and then granted via completion ast, we have to fetch
4046 * layout here. Please note that we can't use the LVB buffer in
4047 * completion AST because it doesn't have a large enough buffer */
4048 rc = ll_get_default_mdsize(sbi, &lmmsize);
4050 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
4051 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
4056 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4058 GOTO(out, rc = -EPROTO);
4060 lmmsize = body->mbo_eadatasize;
4061 if (lmmsize == 0) /* empty layout */
4064 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
4066 GOTO(out, rc = -EFAULT);
4068 OBD_ALLOC_LARGE(lvbdata, lmmsize);
4069 if (lvbdata == NULL)
4070 GOTO(out, rc = -ENOMEM);
4072 memcpy(lvbdata, lmm, lmmsize);
4073 lock_res_and_lock(lock);
4074 if (unlikely(lock->l_lvb_data == NULL)) {
4075 lock->l_lvb_type = LVB_T_LAYOUT;
4076 lock->l_lvb_data = lvbdata;
4077 lock->l_lvb_len = lmmsize;
4080 unlock_res_and_lock(lock);
4083 OBD_FREE_LARGE(lvbdata, lmmsize);
4088 ptlrpc_req_finished(req);
4093 * Apply the layout to the inode. Layout lock is held and will be released
4096 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
4097 struct inode *inode)
4099 struct ll_inode_info *lli = ll_i2info(inode);
4100 struct ll_sb_info *sbi = ll_i2sbi(inode);
4101 struct ldlm_lock *lock;
4102 struct cl_object_conf conf;
4105 bool wait_layout = false;
4108 LASSERT(lustre_handle_is_used(lockh));
4110 lock = ldlm_handle2lock(lockh);
4111 LASSERT(lock != NULL);
4112 LASSERT(ldlm_has_layout(lock));
4114 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
4115 PFID(&lli->lli_fid), inode);
4117 /* in case this is a caching lock and reinstate with new inode */
4118 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
4120 lock_res_and_lock(lock);
4121 lvb_ready = ldlm_is_lvb_ready(lock);
4122 unlock_res_and_lock(lock);
4123 /* checking lvb_ready is racy but this is okay. The worst case is
4124 * that multi processes may configure the file on the same time. */
4129 rc = ll_layout_fetch(inode, lock);
4133 /* for layout lock, lmm is stored in lock's lvb.
4134 * lvb_data is immutable if the lock is held so it's safe to access it
4137 * set layout to file. Unlikely this will fail as old layout was
4138 * surely eliminated */
4139 memset(&conf, 0, sizeof conf);
4140 conf.coc_opc = OBJECT_CONF_SET;
4141 conf.coc_inode = inode;
4142 conf.coc_lock = lock;
4143 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
4144 conf.u.coc_layout.lb_len = lock->l_lvb_len;
4145 rc = ll_layout_conf(inode, &conf);
4147 /* refresh layout failed, need to wait */
4148 wait_layout = rc == -EBUSY;
4152 LDLM_LOCK_PUT(lock);
4153 ldlm_lock_decref(lockh, mode);
4155 /* wait for IO to complete if it's still being used. */
4157 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
4158 ll_get_fsname(inode->i_sb, NULL, 0),
4159 PFID(&lli->lli_fid), inode);
4161 memset(&conf, 0, sizeof conf);
4162 conf.coc_opc = OBJECT_CONF_WAIT;
4163 conf.coc_inode = inode;
4164 rc = ll_layout_conf(inode, &conf);
4168 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
4169 ll_get_fsname(inode->i_sb, NULL, 0),
4170 PFID(&lli->lli_fid), rc);
4175 static int ll_layout_refresh_locked(struct inode *inode)
4177 struct ll_inode_info *lli = ll_i2info(inode);
4178 struct ll_sb_info *sbi = ll_i2sbi(inode);
4179 struct md_op_data *op_data;
4180 struct lookup_intent it;
4181 struct lustre_handle lockh;
4182 enum ldlm_mode mode;
4183 struct ptlrpc_request *req;
4188 /* mostly layout lock is caching on the local side, so try to match
4189 * it before grabbing layout lock mutex. */
4190 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
4191 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
4192 if (mode != 0) { /* hit cached lock */
4193 rc = ll_layout_lock_set(&lockh, mode, inode);
4200 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
4201 0, 0, LUSTRE_OPC_ANY, NULL);
4202 if (IS_ERR(op_data))
4203 RETURN(PTR_ERR(op_data));
4205 /* have to enqueue one */
4206 memset(&it, 0, sizeof(it));
4207 it.it_op = IT_LAYOUT;
4209 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
4210 ll_get_fsname(inode->i_sb, NULL, 0),
4211 PFID(&lli->lli_fid), inode);
4213 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
4214 &ll_md_blocking_ast, 0);
4215 if (it.it_request != NULL)
4216 ptlrpc_req_finished(it.it_request);
4217 it.it_request = NULL;
4219 ll_finish_md_op_data(op_data);
4221 mode = it.it_lock_mode;
4222 it.it_lock_mode = 0;
4223 ll_intent_drop_lock(&it);
4226 /* set lock data in case this is a new lock */
4227 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
4228 lockh.cookie = it.it_lock_handle;
4229 rc = ll_layout_lock_set(&lockh, mode, inode);
4238 * This function checks if there exists a LAYOUT lock on the client side,
4239 * or enqueues it if it doesn't have one in cache.
4241 * This function will not hold layout lock so it may be revoked any time after
4242 * this function returns. Any operations depend on layout should be redone
4245 * This function should be called before lov_io_init() to get an uptodate
4246 * layout version, the caller should save the version number and after IO
4247 * is finished, this function should be called again to verify that layout
4248 * is not changed during IO time.
4250 int ll_layout_refresh(struct inode *inode, __u32 *gen)
4252 struct ll_inode_info *lli = ll_i2info(inode);
4253 struct ll_sb_info *sbi = ll_i2sbi(inode);
4257 *gen = ll_layout_version_get(lli);
4258 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
4262 LASSERT(fid_is_sane(ll_inode2fid(inode)));
4263 LASSERT(S_ISREG(inode->i_mode));
4265 /* take layout lock mutex to enqueue layout lock exclusively. */
4266 mutex_lock(&lli->lli_layout_mutex);
4268 rc = ll_layout_refresh_locked(inode);
4272 *gen = ll_layout_version_get(lli);
4274 mutex_unlock(&lli->lli_layout_mutex);
4280 * This function send a restore request to the MDT
4282 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
4284 struct hsm_user_request *hur;
4288 len = sizeof(struct hsm_user_request) +
4289 sizeof(struct hsm_user_item);
4290 OBD_ALLOC(hur, len);
4294 hur->hur_request.hr_action = HUA_RESTORE;
4295 hur->hur_request.hr_archive_id = 0;
4296 hur->hur_request.hr_flags = 0;
4297 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
4298 sizeof(hur->hur_user_item[0].hui_fid));
4299 hur->hur_user_item[0].hui_extent.offset = offset;
4300 hur->hur_user_item[0].hui_extent.length = length;
4301 hur->hur_request.hr_itemcount = 1;
4302 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,