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 <uapi/linux/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;
1195 cl_io_fini(env, io);
1197 if ((rc == 0 || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
1199 "%s: restart %s from %lld, count:%zu, result: %zd\n",
1200 file_dentry(file)->d_name.name,
1201 iot == CIT_READ ? "read" : "write",
1202 *ppos, count, result);
1206 if (iot == CIT_READ) {
1208 ll_stats_ops_tally(ll_i2sbi(inode),
1209 LPROC_LL_READ_BYTES, result);
1210 } else if (iot == CIT_WRITE) {
1212 ll_stats_ops_tally(ll_i2sbi(inode),
1213 LPROC_LL_WRITE_BYTES, result);
1214 fd->fd_write_failed = false;
1215 } else if (result == 0 && rc == 0) {
1218 fd->fd_write_failed = true;
1220 fd->fd_write_failed = false;
1221 } else if (rc != -ERESTARTSYS) {
1222 fd->fd_write_failed = true;
1226 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1228 RETURN(result > 0 ? result : rc);
1232 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1233 * especially for small I/O.
1235 * To serve a read request, CLIO has to create and initialize a cl_io and
1236 * then request DLM lock. This has turned out to have siginificant overhead
1237 * and affects the performance of small I/O dramatically.
1239 * It's not necessary to create a cl_io for each I/O. Under the help of read
1240 * ahead, most of the pages being read are already in memory cache and we can
1241 * read those pages directly because if the pages exist, the corresponding DLM
1242 * lock must exist so that page content must be valid.
1244 * In fast read implementation, the llite speculatively finds and reads pages
1245 * in memory cache. There are three scenarios for fast read:
1246 * - If the page exists and is uptodate, kernel VM will provide the data and
1247 * CLIO won't be intervened;
1248 * - If the page was brought into memory by read ahead, it will be exported
1249 * and read ahead parameters will be updated;
1250 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1251 * it will go back and invoke normal read, i.e., a cl_io will be created
1252 * and DLM lock will be requested.
1254 * POSIX compliance: posix standard states that read is intended to be atomic.
1255 * Lustre read implementation is in line with Linux kernel read implementation
1256 * and neither of them complies with POSIX standard in this matter. Fast read
1257 * doesn't make the situation worse on single node but it may interleave write
1258 * results from multiple nodes due to short read handling in ll_file_aio_read().
1260 * \param env - lu_env
1261 * \param iocb - kiocb from kernel
1262 * \param iter - user space buffers where the data will be copied
1264 * \retval - number of bytes have been read, or error code if error occurred.
1267 ll_do_fast_read(const struct lu_env *env, struct kiocb *iocb,
1268 struct iov_iter *iter)
1272 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1275 /* NB: we can't do direct IO for fast read because it will need a lock
1276 * to make IO engine happy. */
1277 if (iocb->ki_filp->f_flags & O_DIRECT)
1280 ll_cl_add(iocb->ki_filp, env, NULL, LCC_RW);
1281 result = generic_file_read_iter(iocb, iter);
1282 ll_cl_remove(iocb->ki_filp, env);
1284 /* If the first page is not in cache, generic_file_aio_read() will be
1285 * returned with -ENODATA.
1286 * See corresponding code in ll_readpage(). */
1287 if (result == -ENODATA)
1291 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1292 LPROC_LL_READ_BYTES, result);
1298 * Read from a file (through the page cache).
1300 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1303 struct vvp_io_args *args;
1308 env = cl_env_get(&refcheck);
1310 return PTR_ERR(env);
1312 result = ll_do_fast_read(env, iocb, to);
1313 if (result < 0 || iov_iter_count(to) == 0)
1316 args = ll_env_args(env, IO_NORMAL);
1317 args->u.normal.via_iter = to;
1318 args->u.normal.via_iocb = iocb;
1320 rc2 = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1321 &iocb->ki_pos, iov_iter_count(to));
1324 else if (result == 0)
1328 cl_env_put(env, &refcheck);
1333 * Write to a file (through the page cache).
1335 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1337 struct vvp_io_args *args;
1342 env = cl_env_get(&refcheck);
1344 return PTR_ERR(env);
1346 args = ll_env_args(env, IO_NORMAL);
1347 args->u.normal.via_iter = from;
1348 args->u.normal.via_iocb = iocb;
1350 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1351 &iocb->ki_pos, iov_iter_count(from));
1352 cl_env_put(env, &refcheck);
1356 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1358 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1360 static int ll_file_get_iov_count(const struct iovec *iov,
1361 unsigned long *nr_segs, size_t *count)
1366 for (seg = 0; seg < *nr_segs; seg++) {
1367 const struct iovec *iv = &iov[seg];
1370 * If any segment has a negative length, or the cumulative
1371 * length ever wraps negative then return -EINVAL.
1374 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1376 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1381 cnt -= iv->iov_len; /* This segment is no good */
1388 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1389 unsigned long nr_segs, loff_t pos)
1396 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1400 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1401 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
1402 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1403 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
1404 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1406 result = ll_file_read_iter(iocb, &to);
1411 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
1415 struct iovec iov = { .iov_base = buf, .iov_len = count };
1416 struct kiocb *kiocb;
1421 env = cl_env_get(&refcheck);
1423 RETURN(PTR_ERR(env));
1425 kiocb = &ll_env_info(env)->lti_kiocb;
1426 init_sync_kiocb(kiocb, file);
1427 kiocb->ki_pos = *ppos;
1428 #ifdef HAVE_KIOCB_KI_LEFT
1429 kiocb->ki_left = count;
1430 #elif defined(HAVE_KI_NBYTES)
1431 kiocb->ki_nbytes = count;
1434 result = ll_file_aio_read(kiocb, &iov, 1, kiocb->ki_pos);
1435 *ppos = kiocb->ki_pos;
1437 cl_env_put(env, &refcheck);
1442 * Write to a file (through the page cache).
1445 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1446 unsigned long nr_segs, loff_t pos)
1448 struct iov_iter from;
1453 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1457 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1458 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
1459 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1460 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
1461 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1463 result = ll_file_write_iter(iocb, &from);
1468 static ssize_t ll_file_write(struct file *file, const char __user *buf,
1469 size_t count, loff_t *ppos)
1472 struct iovec iov = { .iov_base = (void __user *)buf,
1474 struct kiocb *kiocb;
1479 env = cl_env_get(&refcheck);
1481 RETURN(PTR_ERR(env));
1483 kiocb = &ll_env_info(env)->lti_kiocb;
1484 init_sync_kiocb(kiocb, file);
1485 kiocb->ki_pos = *ppos;
1486 #ifdef HAVE_KIOCB_KI_LEFT
1487 kiocb->ki_left = count;
1488 #elif defined(HAVE_KI_NBYTES)
1489 kiocb->ki_nbytes = count;
1492 result = ll_file_aio_write(kiocb, &iov, 1, kiocb->ki_pos);
1493 *ppos = kiocb->ki_pos;
1495 cl_env_put(env, &refcheck);
1498 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
1501 * Send file content (through pagecache) somewhere with helper
1503 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1504 struct pipe_inode_info *pipe, size_t count,
1508 struct vvp_io_args *args;
1513 env = cl_env_get(&refcheck);
1515 RETURN(PTR_ERR(env));
1517 args = ll_env_args(env, IO_SPLICE);
1518 args->u.splice.via_pipe = pipe;
1519 args->u.splice.via_flags = flags;
1521 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1522 cl_env_put(env, &refcheck);
1526 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1527 __u64 flags, struct lov_user_md *lum,
1530 struct lookup_intent oit = {
1532 .it_flags = flags | MDS_OPEN_BY_FID,
1537 ll_inode_size_lock(inode);
1538 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1540 GOTO(out_unlock, rc);
1542 ll_release_openhandle(file_dentry(file), &oit);
1545 ll_inode_size_unlock(inode);
1546 ll_intent_release(&oit);
1547 cl_lov_delay_create_clear(&file->f_flags);
1552 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1553 struct lov_mds_md **lmmp, int *lmm_size,
1554 struct ptlrpc_request **request)
1556 struct ll_sb_info *sbi = ll_i2sbi(inode);
1557 struct mdt_body *body;
1558 struct lov_mds_md *lmm = NULL;
1559 struct ptlrpc_request *req = NULL;
1560 struct md_op_data *op_data;
1563 rc = ll_get_default_mdsize(sbi, &lmmsize);
1567 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1568 strlen(filename), lmmsize,
1569 LUSTRE_OPC_ANY, NULL);
1570 if (IS_ERR(op_data))
1571 RETURN(PTR_ERR(op_data));
1573 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1574 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1575 ll_finish_md_op_data(op_data);
1577 CDEBUG(D_INFO, "md_getattr_name failed "
1578 "on %s: rc %d\n", filename, rc);
1582 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1583 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1585 lmmsize = body->mbo_eadatasize;
1587 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1589 GOTO(out, rc = -ENODATA);
1592 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1593 LASSERT(lmm != NULL);
1595 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
1596 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
1597 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1))
1598 GOTO(out, rc = -EPROTO);
1601 * This is coming from the MDS, so is probably in
1602 * little endian. We convert it to host endian before
1603 * passing it to userspace.
1605 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1608 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
1609 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1610 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1611 if (le32_to_cpu(lmm->lmm_pattern) &
1612 LOV_PATTERN_F_RELEASED)
1616 /* if function called for directory - we should
1617 * avoid swab not existent lsm objects */
1618 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1619 lustre_swab_lov_user_md_v1(
1620 (struct lov_user_md_v1 *)lmm);
1621 if (S_ISREG(body->mbo_mode))
1622 lustre_swab_lov_user_md_objects(
1623 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1625 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1626 lustre_swab_lov_user_md_v3(
1627 (struct lov_user_md_v3 *)lmm);
1628 if (S_ISREG(body->mbo_mode))
1629 lustre_swab_lov_user_md_objects(
1630 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1632 } else if (lmm->lmm_magic ==
1633 cpu_to_le32(LOV_MAGIC_COMP_V1)) {
1634 lustre_swab_lov_comp_md_v1(
1635 (struct lov_comp_md_v1 *)lmm);
1641 *lmm_size = lmmsize;
1646 static int ll_lov_setea(struct inode *inode, struct file *file,
1649 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1650 struct lov_user_md *lump;
1651 int lum_size = sizeof(struct lov_user_md) +
1652 sizeof(struct lov_user_ost_data);
1656 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1659 OBD_ALLOC_LARGE(lump, lum_size);
1663 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size))
1664 GOTO(out_lump, rc = -EFAULT);
1666 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1669 OBD_FREE_LARGE(lump, lum_size);
1673 static int ll_file_getstripe(struct inode *inode,
1674 struct lov_user_md __user *lum)
1681 env = cl_env_get(&refcheck);
1683 RETURN(PTR_ERR(env));
1685 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum);
1686 cl_env_put(env, &refcheck);
1690 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1693 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
1694 struct lov_user_md *klum;
1696 __u64 flags = FMODE_WRITE;
1699 rc = ll_copy_user_md(lum, &klum);
1704 rc = ll_lov_setstripe_ea_info(inode, file, flags, klum, lum_size);
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 cl_object *obj = lli->lli_clob;
1715 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1716 struct ll_grouplock grouplock;
1721 CWARN("group id for group lock must not be 0\n");
1725 if (ll_file_nolock(file))
1726 RETURN(-EOPNOTSUPP);
1728 spin_lock(&lli->lli_lock);
1729 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1730 CWARN("group lock already existed with gid %lu\n",
1731 fd->fd_grouplock.lg_gid);
1732 spin_unlock(&lli->lli_lock);
1735 LASSERT(fd->fd_grouplock.lg_lock == NULL);
1736 spin_unlock(&lli->lli_lock);
1739 * XXX: group lock needs to protect all OST objects while PFL
1740 * can add new OST objects during the IO, so we'd instantiate
1741 * all OST objects before getting its group lock.
1746 struct cl_layout cl = {
1747 .cl_is_composite = false,
1750 env = cl_env_get(&refcheck);
1752 RETURN(PTR_ERR(env));
1754 rc = cl_object_layout_get(env, obj, &cl);
1755 if (!rc && cl.cl_is_composite)
1756 rc = ll_layout_write_intent(inode, 0, OBD_OBJECT_EOF);
1758 cl_env_put(env, &refcheck);
1763 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1764 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1768 spin_lock(&lli->lli_lock);
1769 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1770 spin_unlock(&lli->lli_lock);
1771 CERROR("another thread just won the race\n");
1772 cl_put_grouplock(&grouplock);
1776 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1777 fd->fd_grouplock = grouplock;
1778 spin_unlock(&lli->lli_lock);
1780 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1784 static int ll_put_grouplock(struct inode *inode, struct file *file,
1787 struct ll_inode_info *lli = ll_i2info(inode);
1788 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1789 struct ll_grouplock grouplock;
1792 spin_lock(&lli->lli_lock);
1793 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1794 spin_unlock(&lli->lli_lock);
1795 CWARN("no group lock held\n");
1799 LASSERT(fd->fd_grouplock.lg_lock != NULL);
1801 if (fd->fd_grouplock.lg_gid != arg) {
1802 CWARN("group lock %lu doesn't match current id %lu\n",
1803 arg, fd->fd_grouplock.lg_gid);
1804 spin_unlock(&lli->lli_lock);
1808 grouplock = fd->fd_grouplock;
1809 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1810 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1811 spin_unlock(&lli->lli_lock);
1813 cl_put_grouplock(&grouplock);
1814 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1819 * Close inode open handle
1821 * \param dentry [in] dentry which contains the inode
1822 * \param it [in,out] intent which contains open info and result
1825 * \retval <0 failure
1827 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1829 struct inode *inode = dentry->d_inode;
1830 struct obd_client_handle *och;
1836 /* Root ? Do nothing. */
1837 if (dentry->d_inode->i_sb->s_root == dentry)
1840 /* No open handle to close? Move away */
1841 if (!it_disposition(it, DISP_OPEN_OPEN))
1844 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1846 OBD_ALLOC(och, sizeof(*och));
1848 GOTO(out, rc = -ENOMEM);
1850 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1852 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
1854 /* this one is in place of ll_file_open */
1855 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1856 ptlrpc_req_finished(it->it_request);
1857 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1863 * Get size for inode for which FIEMAP mapping is requested.
1864 * Make the FIEMAP get_info call and returns the result.
1865 * \param fiemap kernel buffer to hold extens
1866 * \param num_bytes kernel buffer size
1868 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
1874 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
1877 /* Checks for fiemap flags */
1878 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1879 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1883 /* Check for FIEMAP_FLAG_SYNC */
1884 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1885 rc = filemap_fdatawrite(inode->i_mapping);
1890 env = cl_env_get(&refcheck);
1892 RETURN(PTR_ERR(env));
1894 if (i_size_read(inode) == 0) {
1895 rc = ll_glimpse_size(inode);
1900 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1901 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
1902 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
1904 /* If filesize is 0, then there would be no objects for mapping */
1905 if (fmkey.lfik_oa.o_size == 0) {
1906 fiemap->fm_mapped_extents = 0;
1910 fmkey.lfik_fiemap = *fiemap;
1912 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
1913 &fmkey, fiemap, &num_bytes);
1915 cl_env_put(env, &refcheck);
1919 int ll_fid2path(struct inode *inode, void __user *arg)
1921 struct obd_export *exp = ll_i2mdexp(inode);
1922 const struct getinfo_fid2path __user *gfin = arg;
1924 struct getinfo_fid2path *gfout;
1930 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1931 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1934 /* Only need to get the buflen */
1935 if (get_user(pathlen, &gfin->gf_pathlen))
1938 if (pathlen > PATH_MAX)
1941 outsize = sizeof(*gfout) + pathlen;
1942 OBD_ALLOC(gfout, outsize);
1946 if (copy_from_user(gfout, arg, sizeof(*gfout)))
1947 GOTO(gf_free, rc = -EFAULT);
1948 /* append root FID after gfout to let MDT know the root FID so that it
1949 * can lookup the correct path, this is mainly for fileset.
1950 * old server without fileset mount support will ignore this. */
1951 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
1953 /* Call mdc_iocontrol */
1954 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1958 if (copy_to_user(arg, gfout, outsize))
1962 OBD_FREE(gfout, outsize);
1967 * Read the data_version for inode.
1969 * This value is computed using stripe object version on OST.
1970 * Version is computed using server side locking.
1972 * @param flags if do sync on the OST side;
1974 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1975 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1977 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1979 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1987 /* If no file object initialized, we consider its version is 0. */
1993 env = cl_env_get(&refcheck);
1995 RETURN(PTR_ERR(env));
1997 io = vvp_env_thread_io(env);
1999 io->u.ci_data_version.dv_data_version = 0;
2000 io->u.ci_data_version.dv_flags = flags;
2003 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2004 result = cl_io_loop(env, io);
2006 result = io->ci_result;
2008 *data_version = io->u.ci_data_version.dv_data_version;
2010 cl_io_fini(env, io);
2012 if (unlikely(io->ci_need_restart))
2015 cl_env_put(env, &refcheck);
2021 * Trigger a HSM release request for the provided inode.
2023 int ll_hsm_release(struct inode *inode)
2026 struct obd_client_handle *och = NULL;
2027 __u64 data_version = 0;
2032 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2033 ll_get_fsname(inode->i_sb, NULL, 0),
2034 PFID(&ll_i2info(inode)->lli_fid));
2036 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2038 GOTO(out, rc = PTR_ERR(och));
2040 /* Grab latest data_version and [am]time values */
2041 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2045 env = cl_env_get(&refcheck);
2047 GOTO(out, rc = PTR_ERR(env));
2049 ll_merge_attr(env, inode);
2050 cl_env_put(env, &refcheck);
2052 /* Release the file.
2053 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2054 * we still need it to pack l_remote_handle to MDT. */
2055 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2061 if (och != NULL && !IS_ERR(och)) /* close the file */
2062 ll_lease_close(och, inode, NULL);
2067 struct ll_swap_stack {
2070 struct inode *inode1;
2071 struct inode *inode2;
2076 static int ll_swap_layouts(struct file *file1, struct file *file2,
2077 struct lustre_swap_layouts *lsl)
2079 struct mdc_swap_layouts msl;
2080 struct md_op_data *op_data;
2083 struct ll_swap_stack *llss = NULL;
2086 OBD_ALLOC_PTR(llss);
2090 llss->inode1 = file_inode(file1);
2091 llss->inode2 = file_inode(file2);
2093 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2097 /* we use 2 bool because it is easier to swap than 2 bits */
2098 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2099 llss->check_dv1 = true;
2101 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2102 llss->check_dv2 = true;
2104 /* we cannot use lsl->sl_dvX directly because we may swap them */
2105 llss->dv1 = lsl->sl_dv1;
2106 llss->dv2 = lsl->sl_dv2;
2108 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2109 if (rc == 0) /* same file, done! */
2112 if (rc < 0) { /* sequentialize it */
2113 swap(llss->inode1, llss->inode2);
2115 swap(llss->dv1, llss->dv2);
2116 swap(llss->check_dv1, llss->check_dv2);
2120 if (gid != 0) { /* application asks to flush dirty cache */
2121 rc = ll_get_grouplock(llss->inode1, file1, gid);
2125 rc = ll_get_grouplock(llss->inode2, file2, gid);
2127 ll_put_grouplock(llss->inode1, file1, gid);
2132 /* ultimate check, before swaping the layouts we check if
2133 * dataversion has changed (if requested) */
2134 if (llss->check_dv1) {
2135 rc = ll_data_version(llss->inode1, &dv, 0);
2138 if (dv != llss->dv1)
2139 GOTO(putgl, rc = -EAGAIN);
2142 if (llss->check_dv2) {
2143 rc = ll_data_version(llss->inode2, &dv, 0);
2146 if (dv != llss->dv2)
2147 GOTO(putgl, rc = -EAGAIN);
2150 /* struct md_op_data is used to send the swap args to the mdt
2151 * only flags is missing, so we use struct mdc_swap_layouts
2152 * through the md_op_data->op_data */
2153 /* flags from user space have to be converted before they are send to
2154 * server, no flag is sent today, they are only used on the client */
2157 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2158 0, LUSTRE_OPC_ANY, &msl);
2159 if (IS_ERR(op_data))
2160 GOTO(free, rc = PTR_ERR(op_data));
2162 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2163 sizeof(*op_data), op_data, NULL);
2164 ll_finish_md_op_data(op_data);
2171 ll_put_grouplock(llss->inode2, file2, gid);
2172 ll_put_grouplock(llss->inode1, file1, gid);
2182 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2184 struct md_op_data *op_data;
2188 /* Detect out-of range masks */
2189 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2192 /* Non-root users are forbidden to set or clear flags which are
2193 * NOT defined in HSM_USER_MASK. */
2194 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2195 !cfs_capable(CFS_CAP_SYS_ADMIN))
2198 /* Detect out-of range archive id */
2199 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2200 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
2203 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2204 LUSTRE_OPC_ANY, hss);
2205 if (IS_ERR(op_data))
2206 RETURN(PTR_ERR(op_data));
2208 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2209 sizeof(*op_data), op_data, NULL);
2211 ll_finish_md_op_data(op_data);
2216 static int ll_hsm_import(struct inode *inode, struct file *file,
2217 struct hsm_user_import *hui)
2219 struct hsm_state_set *hss = NULL;
2220 struct iattr *attr = NULL;
2224 if (!S_ISREG(inode->i_mode))
2230 GOTO(out, rc = -ENOMEM);
2232 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2233 hss->hss_archive_id = hui->hui_archive_id;
2234 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2235 rc = ll_hsm_state_set(inode, hss);
2239 OBD_ALLOC_PTR(attr);
2241 GOTO(out, rc = -ENOMEM);
2243 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2244 attr->ia_mode |= S_IFREG;
2245 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2246 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2247 attr->ia_size = hui->hui_size;
2248 attr->ia_mtime.tv_sec = hui->hui_mtime;
2249 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2250 attr->ia_atime.tv_sec = hui->hui_atime;
2251 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2253 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2254 ATTR_UID | ATTR_GID |
2255 ATTR_MTIME | ATTR_MTIME_SET |
2256 ATTR_ATIME | ATTR_ATIME_SET;
2260 rc = ll_setattr_raw(file_dentry(file), attr, true);
2264 inode_unlock(inode);
2276 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2278 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2279 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2282 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
2284 struct inode *inode = file_inode(file);
2286 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
2287 ATTR_MTIME | ATTR_MTIME_SET |
2288 ATTR_CTIME | ATTR_CTIME_SET,
2290 .tv_sec = lfu->lfu_atime_sec,
2291 .tv_nsec = lfu->lfu_atime_nsec,
2294 .tv_sec = lfu->lfu_mtime_sec,
2295 .tv_nsec = lfu->lfu_mtime_nsec,
2298 .tv_sec = lfu->lfu_ctime_sec,
2299 .tv_nsec = lfu->lfu_ctime_nsec,
2305 if (!capable(CAP_SYS_ADMIN))
2308 if (!S_ISREG(inode->i_mode))
2312 rc = ll_setattr_raw(file_dentry(file), &ia, false);
2313 inode_unlock(inode);
2319 * Give file access advices
2321 * The ladvise interface is similar to Linux fadvise() system call, except it
2322 * forwards the advices directly from Lustre client to server. The server side
2323 * codes will apply appropriate read-ahead and caching techniques for the
2324 * corresponding files.
2326 * A typical workload for ladvise is e.g. a bunch of different clients are
2327 * doing small random reads of a file, so prefetching pages into OSS cache
2328 * with big linear reads before the random IO is a net benefit. Fetching
2329 * all that data into each client cache with fadvise() may not be, due to
2330 * much more data being sent to the client.
2332 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
2333 struct llapi_lu_ladvise *ladvise)
2337 struct cl_ladvise_io *lio;
2342 env = cl_env_get(&refcheck);
2344 RETURN(PTR_ERR(env));
2346 io = vvp_env_thread_io(env);
2347 io->ci_obj = ll_i2info(inode)->lli_clob;
2349 /* initialize parameters for ladvise */
2350 lio = &io->u.ci_ladvise;
2351 lio->li_start = ladvise->lla_start;
2352 lio->li_end = ladvise->lla_end;
2353 lio->li_fid = ll_inode2fid(inode);
2354 lio->li_advice = ladvise->lla_advice;
2355 lio->li_flags = flags;
2357 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
2358 rc = cl_io_loop(env, io);
2362 cl_io_fini(env, io);
2363 cl_env_put(env, &refcheck);
2367 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
2370 struct fsxattr fsxattr;
2372 if (copy_from_user(&fsxattr,
2373 (const struct fsxattr __user *)arg,
2377 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
2378 if (copy_to_user((struct fsxattr __user *)arg,
2379 &fsxattr, sizeof(fsxattr)))
2385 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
2389 struct md_op_data *op_data;
2390 struct ptlrpc_request *req = NULL;
2392 struct fsxattr fsxattr;
2394 /* only root could change project ID */
2395 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2398 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2399 LUSTRE_OPC_ANY, NULL);
2400 if (IS_ERR(op_data))
2401 RETURN(PTR_ERR(op_data));
2403 if (copy_from_user(&fsxattr,
2404 (const struct fsxattr __user *)arg,
2406 GOTO(out_fsxattr1, rc = -EFAULT);
2408 op_data->op_projid = fsxattr.fsx_projid;
2409 op_data->op_attr.ia_valid |= MDS_ATTR_PROJID;
2410 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL,
2412 ptlrpc_req_finished(req);
2415 ll_finish_md_op_data(op_data);
2422 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2424 struct inode *inode = file_inode(file);
2425 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2429 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
2430 PFID(ll_inode2fid(inode)), inode, cmd);
2431 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2433 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2434 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2438 case LL_IOC_GETFLAGS:
2439 /* Get the current value of the file flags */
2440 return put_user(fd->fd_flags, (int __user *)arg);
2441 case LL_IOC_SETFLAGS:
2442 case LL_IOC_CLRFLAGS:
2443 /* Set or clear specific file flags */
2444 /* XXX This probably needs checks to ensure the flags are
2445 * not abused, and to handle any flag side effects.
2447 if (get_user(flags, (int __user *) arg))
2450 if (cmd == LL_IOC_SETFLAGS) {
2451 if ((flags & LL_FILE_IGNORE_LOCK) &&
2452 !(file->f_flags & O_DIRECT)) {
2453 CERROR("%s: unable to disable locking on "
2454 "non-O_DIRECT file\n", current->comm);
2458 fd->fd_flags |= flags;
2460 fd->fd_flags &= ~flags;
2463 case LL_IOC_LOV_SETSTRIPE:
2464 RETURN(ll_lov_setstripe(inode, file, arg));
2465 case LL_IOC_LOV_SETEA:
2466 RETURN(ll_lov_setea(inode, file, arg));
2467 case LL_IOC_LOV_SWAP_LAYOUTS: {
2469 struct lustre_swap_layouts lsl;
2471 if (copy_from_user(&lsl, (char __user *)arg,
2472 sizeof(struct lustre_swap_layouts)))
2475 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
2478 file2 = fget(lsl.sl_fd);
2482 /* O_WRONLY or O_RDWR */
2483 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
2484 GOTO(out, rc = -EPERM);
2486 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
2487 struct inode *inode2;
2488 struct ll_inode_info *lli;
2489 struct obd_client_handle *och = NULL;
2491 if (lsl.sl_flags != SWAP_LAYOUTS_CLOSE)
2492 GOTO(out, rc = -EINVAL);
2494 lli = ll_i2info(inode);
2495 mutex_lock(&lli->lli_och_mutex);
2496 if (fd->fd_lease_och != NULL) {
2497 och = fd->fd_lease_och;
2498 fd->fd_lease_och = NULL;
2500 mutex_unlock(&lli->lli_och_mutex);
2502 GOTO(out, rc = -ENOLCK);
2503 inode2 = file_inode(file2);
2504 rc = ll_swap_layouts_close(och, inode, inode2);
2506 rc = ll_swap_layouts(file, file2, &lsl);
2512 case LL_IOC_LOV_GETSTRIPE:
2513 RETURN(ll_file_getstripe(inode,
2514 (struct lov_user_md __user *)arg));
2515 case FSFILT_IOC_GETFLAGS:
2516 case FSFILT_IOC_SETFLAGS:
2517 RETURN(ll_iocontrol(inode, file, cmd, arg));
2518 case FSFILT_IOC_GETVERSION_OLD:
2519 case FSFILT_IOC_GETVERSION:
2520 RETURN(put_user(inode->i_generation, (int __user *)arg));
2521 case LL_IOC_GROUP_LOCK:
2522 RETURN(ll_get_grouplock(inode, file, arg));
2523 case LL_IOC_GROUP_UNLOCK:
2524 RETURN(ll_put_grouplock(inode, file, arg));
2525 case IOC_OBD_STATFS:
2526 RETURN(ll_obd_statfs(inode, (void __user *)arg));
2528 /* We need to special case any other ioctls we want to handle,
2529 * to send them to the MDS/OST as appropriate and to properly
2530 * network encode the arg field.
2531 case FSFILT_IOC_SETVERSION_OLD:
2532 case FSFILT_IOC_SETVERSION:
2534 case LL_IOC_FLUSHCTX:
2535 RETURN(ll_flush_ctx(inode));
2536 case LL_IOC_PATH2FID: {
2537 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2538 sizeof(struct lu_fid)))
2543 case LL_IOC_GETPARENT:
2544 RETURN(ll_getparent(file, (struct getparent __user *)arg));
2546 case OBD_IOC_FID2PATH:
2547 RETURN(ll_fid2path(inode, (void __user *)arg));
2548 case LL_IOC_DATA_VERSION: {
2549 struct ioc_data_version idv;
2552 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2555 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2556 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2559 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
2565 case LL_IOC_GET_MDTIDX: {
2568 mdtidx = ll_get_mdt_idx(inode);
2572 if (put_user((int)mdtidx, (int __user *)arg))
2577 case OBD_IOC_GETDTNAME:
2578 case OBD_IOC_GETMDNAME:
2579 RETURN(ll_get_obd_name(inode, cmd, arg));
2580 case LL_IOC_HSM_STATE_GET: {
2581 struct md_op_data *op_data;
2582 struct hsm_user_state *hus;
2589 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2590 LUSTRE_OPC_ANY, hus);
2591 if (IS_ERR(op_data)) {
2593 RETURN(PTR_ERR(op_data));
2596 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2599 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2602 ll_finish_md_op_data(op_data);
2606 case LL_IOC_HSM_STATE_SET: {
2607 struct hsm_state_set *hss;
2614 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
2619 rc = ll_hsm_state_set(inode, hss);
2624 case LL_IOC_HSM_ACTION: {
2625 struct md_op_data *op_data;
2626 struct hsm_current_action *hca;
2633 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2634 LUSTRE_OPC_ANY, hca);
2635 if (IS_ERR(op_data)) {
2637 RETURN(PTR_ERR(op_data));
2640 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2643 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2646 ll_finish_md_op_data(op_data);
2650 case LL_IOC_SET_LEASE: {
2651 struct ll_inode_info *lli = ll_i2info(inode);
2652 struct obd_client_handle *och = NULL;
2657 case LL_LEASE_WRLCK:
2658 if (!(file->f_mode & FMODE_WRITE))
2660 fmode = FMODE_WRITE;
2662 case LL_LEASE_RDLCK:
2663 if (!(file->f_mode & FMODE_READ))
2667 case LL_LEASE_UNLCK:
2668 mutex_lock(&lli->lli_och_mutex);
2669 if (fd->fd_lease_och != NULL) {
2670 och = fd->fd_lease_och;
2671 fd->fd_lease_och = NULL;
2673 mutex_unlock(&lli->lli_och_mutex);
2678 fmode = och->och_flags;
2679 rc = ll_lease_close(och, inode, &lease_broken);
2683 rc = ll_lease_och_release(inode, file);
2690 RETURN(ll_lease_type_from_fmode(fmode));
2695 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
2697 /* apply for lease */
2698 och = ll_lease_open(inode, file, fmode, 0);
2700 RETURN(PTR_ERR(och));
2703 mutex_lock(&lli->lli_och_mutex);
2704 if (fd->fd_lease_och == NULL) {
2705 fd->fd_lease_och = och;
2708 mutex_unlock(&lli->lli_och_mutex);
2710 /* impossible now that only excl is supported for now */
2711 ll_lease_close(och, inode, &lease_broken);
2716 case LL_IOC_GET_LEASE: {
2717 struct ll_inode_info *lli = ll_i2info(inode);
2718 struct ldlm_lock *lock = NULL;
2721 mutex_lock(&lli->lli_och_mutex);
2722 if (fd->fd_lease_och != NULL) {
2723 struct obd_client_handle *och = fd->fd_lease_och;
2725 lock = ldlm_handle2lock(&och->och_lease_handle);
2727 lock_res_and_lock(lock);
2728 if (!ldlm_is_cancel(lock))
2729 fmode = och->och_flags;
2731 unlock_res_and_lock(lock);
2732 LDLM_LOCK_PUT(lock);
2735 mutex_unlock(&lli->lli_och_mutex);
2737 RETURN(ll_lease_type_from_fmode(fmode));
2739 case LL_IOC_HSM_IMPORT: {
2740 struct hsm_user_import *hui;
2746 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
2751 rc = ll_hsm_import(inode, file, hui);
2756 case LL_IOC_FUTIMES_3: {
2757 struct ll_futimes_3 lfu;
2759 if (copy_from_user(&lfu,
2760 (const struct ll_futimes_3 __user *)arg,
2764 RETURN(ll_file_futimes_3(file, &lfu));
2766 case LL_IOC_LADVISE: {
2767 struct llapi_ladvise_hdr *ladvise_hdr;
2770 int alloc_size = sizeof(*ladvise_hdr);
2773 OBD_ALLOC_PTR(ladvise_hdr);
2774 if (ladvise_hdr == NULL)
2777 if (copy_from_user(ladvise_hdr,
2778 (const struct llapi_ladvise_hdr __user *)arg,
2780 GOTO(out_ladvise, rc = -EFAULT);
2782 if (ladvise_hdr->lah_magic != LADVISE_MAGIC ||
2783 ladvise_hdr->lah_count < 1)
2784 GOTO(out_ladvise, rc = -EINVAL);
2786 num_advise = ladvise_hdr->lah_count;
2787 if (num_advise >= LAH_COUNT_MAX)
2788 GOTO(out_ladvise, rc = -EFBIG);
2790 OBD_FREE_PTR(ladvise_hdr);
2791 alloc_size = offsetof(typeof(*ladvise_hdr),
2792 lah_advise[num_advise]);
2793 OBD_ALLOC(ladvise_hdr, alloc_size);
2794 if (ladvise_hdr == NULL)
2798 * TODO: submit multiple advices to one server in a single RPC
2800 if (copy_from_user(ladvise_hdr,
2801 (const struct llapi_ladvise_hdr __user *)arg,
2803 GOTO(out_ladvise, rc = -EFAULT);
2805 for (i = 0; i < num_advise; i++) {
2806 rc = ll_ladvise(inode, file, ladvise_hdr->lah_flags,
2807 &ladvise_hdr->lah_advise[i]);
2813 OBD_FREE(ladvise_hdr, alloc_size);
2816 case LL_IOC_FSGETXATTR:
2817 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2818 case LL_IOC_FSSETXATTR:
2819 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2824 ll_iocontrol_call(inode, file, cmd, arg, &err))
2827 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2828 (void __user *)arg));
2833 #ifndef HAVE_FILE_LLSEEK_SIZE
2834 static inline loff_t
2835 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2837 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2839 if (offset > maxsize)
2842 if (offset != file->f_pos) {
2843 file->f_pos = offset;
2844 file->f_version = 0;
2850 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2851 loff_t maxsize, loff_t eof)
2853 struct inode *inode = file_inode(file);
2861 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2862 * position-querying operation. Avoid rewriting the "same"
2863 * f_pos value back to the file because a concurrent read(),
2864 * write() or lseek() might have altered it
2869 * f_lock protects against read/modify/write race with other
2870 * SEEK_CURs. Note that parallel writes and reads behave
2874 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2875 inode_unlock(inode);
2879 * In the generic case the entire file is data, so as long as
2880 * offset isn't at the end of the file then the offset is data.
2887 * There is a virtual hole at the end of the file, so as long as
2888 * offset isn't i_size or larger, return i_size.
2896 return llseek_execute(file, offset, maxsize);
2900 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2902 struct inode *inode = file_inode(file);
2903 loff_t retval, eof = 0;
2906 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2907 (origin == SEEK_CUR) ? file->f_pos : 0);
2908 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2909 PFID(ll_inode2fid(inode)), inode, retval, retval,
2911 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2913 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2914 retval = ll_glimpse_size(inode);
2917 eof = i_size_read(inode);
2920 retval = ll_generic_file_llseek_size(file, offset, origin,
2921 ll_file_maxbytes(inode), eof);
2925 static int ll_flush(struct file *file, fl_owner_t id)
2927 struct inode *inode = file_inode(file);
2928 struct ll_inode_info *lli = ll_i2info(inode);
2929 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2932 LASSERT(!S_ISDIR(inode->i_mode));
2934 /* catch async errors that were recorded back when async writeback
2935 * failed for pages in this mapping. */
2936 rc = lli->lli_async_rc;
2937 lli->lli_async_rc = 0;
2938 if (lli->lli_clob != NULL) {
2939 err = lov_read_and_clear_async_rc(lli->lli_clob);
2944 /* The application has been told write failure already.
2945 * Do not report failure again. */
2946 if (fd->fd_write_failed)
2948 return rc ? -EIO : 0;
2952 * Called to make sure a portion of file has been written out.
2953 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2955 * Return how many pages have been written.
2957 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2958 enum cl_fsync_mode mode, int ignore_layout)
2962 struct cl_fsync_io *fio;
2967 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2968 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2971 env = cl_env_get(&refcheck);
2973 RETURN(PTR_ERR(env));
2975 io = vvp_env_thread_io(env);
2976 io->ci_obj = ll_i2info(inode)->lli_clob;
2977 io->ci_ignore_layout = ignore_layout;
2979 /* initialize parameters for sync */
2980 fio = &io->u.ci_fsync;
2981 fio->fi_start = start;
2983 fio->fi_fid = ll_inode2fid(inode);
2984 fio->fi_mode = mode;
2985 fio->fi_nr_written = 0;
2987 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2988 result = cl_io_loop(env, io);
2990 result = io->ci_result;
2992 result = fio->fi_nr_written;
2993 cl_io_fini(env, io);
2994 cl_env_put(env, &refcheck);
3000 * When dentry is provided (the 'else' case), file_dentry() may be
3001 * null and dentry must be used directly rather than pulled from
3002 * file_dentry() as is done otherwise.
3005 #ifdef HAVE_FILE_FSYNC_4ARGS
3006 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
3008 struct dentry *dentry = file_dentry(file);
3009 #elif defined(HAVE_FILE_FSYNC_2ARGS)
3010 int ll_fsync(struct file *file, int datasync)
3012 struct dentry *dentry = file_dentry(file);
3014 loff_t end = LLONG_MAX;
3016 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
3019 loff_t end = LLONG_MAX;
3021 struct inode *inode = dentry->d_inode;
3022 struct ll_inode_info *lli = ll_i2info(inode);
3023 struct ptlrpc_request *req;
3027 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
3028 PFID(ll_inode2fid(inode)), inode);
3029 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
3031 #ifdef HAVE_FILE_FSYNC_4ARGS
3032 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
3035 /* fsync's caller has already called _fdata{sync,write}, we want
3036 * that IO to finish before calling the osc and mdc sync methods */
3037 rc = filemap_fdatawait(inode->i_mapping);
3040 /* catch async errors that were recorded back when async writeback
3041 * failed for pages in this mapping. */
3042 if (!S_ISDIR(inode->i_mode)) {
3043 err = lli->lli_async_rc;
3044 lli->lli_async_rc = 0;
3047 if (lli->lli_clob != NULL) {
3048 err = lov_read_and_clear_async_rc(lli->lli_clob);
3054 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
3058 ptlrpc_req_finished(req);
3060 if (S_ISREG(inode->i_mode)) {
3061 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3063 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
3064 if (rc == 0 && err < 0)
3067 fd->fd_write_failed = true;
3069 fd->fd_write_failed = false;
3072 #ifdef HAVE_FILE_FSYNC_4ARGS
3073 inode_unlock(inode);
3079 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
3081 struct inode *inode = file_inode(file);
3082 struct ll_sb_info *sbi = ll_i2sbi(inode);
3083 struct ldlm_enqueue_info einfo = {
3084 .ei_type = LDLM_FLOCK,
3085 .ei_cb_cp = ldlm_flock_completion_ast,
3086 .ei_cbdata = file_lock,
3088 struct md_op_data *op_data;
3089 struct lustre_handle lockh = { 0 };
3090 union ldlm_policy_data flock = { { 0 } };
3091 int fl_type = file_lock->fl_type;
3097 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
3098 PFID(ll_inode2fid(inode)), file_lock);
3100 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
3102 if (file_lock->fl_flags & FL_FLOCK) {
3103 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
3104 /* flocks are whole-file locks */
3105 flock.l_flock.end = OFFSET_MAX;
3106 /* For flocks owner is determined by the local file desctiptor*/
3107 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
3108 } else if (file_lock->fl_flags & FL_POSIX) {
3109 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
3110 flock.l_flock.start = file_lock->fl_start;
3111 flock.l_flock.end = file_lock->fl_end;
3115 flock.l_flock.pid = file_lock->fl_pid;
3117 /* Somewhat ugly workaround for svc lockd.
3118 * lockd installs custom fl_lmops->lm_compare_owner that checks
3119 * for the fl_owner to be the same (which it always is on local node
3120 * I guess between lockd processes) and then compares pid.
3121 * As such we assign pid to the owner field to make it all work,
3122 * conflict with normal locks is unlikely since pid space and
3123 * pointer space for current->files are not intersecting */
3124 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
3125 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
3129 einfo.ei_mode = LCK_PR;
3132 /* An unlock request may or may not have any relation to
3133 * existing locks so we may not be able to pass a lock handle
3134 * via a normal ldlm_lock_cancel() request. The request may even
3135 * unlock a byte range in the middle of an existing lock. In
3136 * order to process an unlock request we need all of the same
3137 * information that is given with a normal read or write record
3138 * lock request. To avoid creating another ldlm unlock (cancel)
3139 * message we'll treat a LCK_NL flock request as an unlock. */
3140 einfo.ei_mode = LCK_NL;
3143 einfo.ei_mode = LCK_PW;
3146 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
3161 flags = LDLM_FL_BLOCK_NOWAIT;
3167 flags = LDLM_FL_TEST_LOCK;
3170 CERROR("unknown fcntl lock command: %d\n", cmd);
3174 /* Save the old mode so that if the mode in the lock changes we
3175 * can decrement the appropriate reader or writer refcount. */
3176 file_lock->fl_type = einfo.ei_mode;
3178 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3179 LUSTRE_OPC_ANY, NULL);
3180 if (IS_ERR(op_data))
3181 RETURN(PTR_ERR(op_data));
3183 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
3184 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
3185 flock.l_flock.pid, flags, einfo.ei_mode,
3186 flock.l_flock.start, flock.l_flock.end);
3188 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
3191 /* Restore the file lock type if not TEST lock. */
3192 if (!(flags & LDLM_FL_TEST_LOCK))
3193 file_lock->fl_type = fl_type;
3195 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
3196 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
3197 !(flags & LDLM_FL_TEST_LOCK))
3198 rc2 = locks_lock_file_wait(file, file_lock);
3200 if ((file_lock->fl_flags & FL_FLOCK) &&
3201 (rc == 0 || file_lock->fl_type == F_UNLCK))
3202 rc2 = flock_lock_file_wait(file, file_lock);
3203 if ((file_lock->fl_flags & FL_POSIX) &&
3204 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
3205 !(flags & LDLM_FL_TEST_LOCK))
3206 rc2 = posix_lock_file_wait(file, file_lock);
3207 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
3209 if (rc2 && file_lock->fl_type != F_UNLCK) {
3210 einfo.ei_mode = LCK_NL;
3211 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
3216 ll_finish_md_op_data(op_data);
3221 int ll_get_fid_by_name(struct inode *parent, const char *name,
3222 int namelen, struct lu_fid *fid,
3223 struct inode **inode)
3225 struct md_op_data *op_data = NULL;
3226 struct mdt_body *body;
3227 struct ptlrpc_request *req;
3231 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
3232 LUSTRE_OPC_ANY, NULL);
3233 if (IS_ERR(op_data))
3234 RETURN(PTR_ERR(op_data));
3236 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
3237 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
3238 ll_finish_md_op_data(op_data);
3242 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3244 GOTO(out_req, rc = -EFAULT);
3246 *fid = body->mbo_fid1;
3249 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
3251 ptlrpc_req_finished(req);
3255 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
3256 const char *name, int namelen)
3258 struct dentry *dchild = NULL;
3259 struct inode *child_inode = NULL;
3260 struct md_op_data *op_data;
3261 struct ptlrpc_request *request = NULL;
3262 struct obd_client_handle *och = NULL;
3264 struct mdt_body *body;
3266 __u64 data_version = 0;
3269 CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%04x\n",
3270 name, PFID(ll_inode2fid(parent)), mdtidx);
3272 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
3273 0, LUSTRE_OPC_ANY, NULL);
3274 if (IS_ERR(op_data))
3275 RETURN(PTR_ERR(op_data));
3277 /* Get child FID first */
3278 qstr.hash = full_name_hash(name, namelen);
3281 dchild = d_lookup(file_dentry(file), &qstr);
3282 if (dchild != NULL) {
3283 if (dchild->d_inode != NULL)
3284 child_inode = igrab(dchild->d_inode);
3288 if (child_inode == NULL) {
3289 rc = ll_get_fid_by_name(parent, name, namelen,
3290 &op_data->op_fid3, &child_inode);
3295 if (child_inode == NULL)
3296 GOTO(out_free, rc = -EINVAL);
3299 * lfs migrate command needs to be blocked on the client
3300 * by checking the migrate FID against the FID of the
3303 if (child_inode == parent->i_sb->s_root->d_inode)
3304 GOTO(out_iput, rc = -EINVAL);
3306 inode_lock(child_inode);
3307 op_data->op_fid3 = *ll_inode2fid(child_inode);
3308 if (!fid_is_sane(&op_data->op_fid3)) {
3309 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
3310 ll_get_fsname(parent->i_sb, NULL, 0), name,
3311 PFID(&op_data->op_fid3));
3312 GOTO(out_unlock, rc = -EINVAL);
3315 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
3317 GOTO(out_unlock, rc);
3320 CDEBUG(D_INFO, "%s: "DFID" is already on MDT%04x\n", name,
3321 PFID(&op_data->op_fid3), mdtidx);
3322 GOTO(out_unlock, rc = 0);
3325 if (S_ISREG(child_inode->i_mode)) {
3326 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
3330 GOTO(out_unlock, rc);
3333 rc = ll_data_version(child_inode, &data_version,
3336 GOTO(out_close, rc);
3338 op_data->op_handle = och->och_fh;
3339 op_data->op_data = och->och_mod;
3340 op_data->op_data_version = data_version;
3341 op_data->op_lease_handle = och->och_lease_handle;
3342 op_data->op_bias |= MDS_RENAME_MIGRATE;
3345 op_data->op_mds = mdtidx;
3346 op_data->op_cli_flags = CLI_MIGRATE;
3347 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
3348 namelen, name, namelen, &request);
3350 LASSERT(request != NULL);
3351 ll_update_times(request, parent);
3353 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
3354 LASSERT(body != NULL);
3356 /* If the server does release layout lock, then we cleanup
3357 * the client och here, otherwise release it in out_close: */
3359 body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
3360 obd_mod_put(och->och_mod);
3361 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
3363 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
3369 if (request != NULL) {
3370 ptlrpc_req_finished(request);
3374 /* Try again if the file layout has changed. */
3375 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
3379 if (och != NULL) /* close the file */
3380 ll_lease_close(och, child_inode, NULL);
3382 clear_nlink(child_inode);
3384 inode_unlock(child_inode);
3388 ll_finish_md_op_data(op_data);
3393 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
3401 * test if some locks matching bits and l_req_mode are acquired
3402 * - bits can be in different locks
3403 * - if found clear the common lock bits in *bits
3404 * - the bits not found, are kept in *bits
3406 * \param bits [IN] searched lock bits [IN]
3407 * \param l_req_mode [IN] searched lock mode
3408 * \retval boolean, true iff all bits are found
3410 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
3412 struct lustre_handle lockh;
3413 union ldlm_policy_data policy;
3414 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
3415 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
3424 fid = &ll_i2info(inode)->lli_fid;
3425 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
3426 ldlm_lockname[mode]);
3428 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
3429 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
3430 policy.l_inodebits.bits = *bits & (1 << i);
3431 if (policy.l_inodebits.bits == 0)
3434 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
3435 &policy, mode, &lockh)) {
3436 struct ldlm_lock *lock;
3438 lock = ldlm_handle2lock(&lockh);
3441 ~(lock->l_policy_data.l_inodebits.bits);
3442 LDLM_LOCK_PUT(lock);
3444 *bits &= ~policy.l_inodebits.bits;
3451 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
3452 struct lustre_handle *lockh, __u64 flags,
3453 enum ldlm_mode mode)
3455 union ldlm_policy_data policy = { .l_inodebits = { bits } };
3460 fid = &ll_i2info(inode)->lli_fid;
3461 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
3463 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
3464 fid, LDLM_IBITS, &policy, mode, lockh);
3469 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
3471 /* Already unlinked. Just update nlink and return success */
3472 if (rc == -ENOENT) {
3474 /* If it is striped directory, and there is bad stripe
3475 * Let's revalidate the dentry again, instead of returning
3477 if (S_ISDIR(inode->i_mode) &&
3478 ll_i2info(inode)->lli_lsm_md != NULL)
3481 /* This path cannot be hit for regular files unless in
3482 * case of obscure races, so no need to to validate
3484 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
3486 } else if (rc != 0) {
3487 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
3488 "%s: revalidate FID "DFID" error: rc = %d\n",
3489 ll_get_fsname(inode->i_sb, NULL, 0),
3490 PFID(ll_inode2fid(inode)), rc);
3496 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3498 struct inode *inode = dentry->d_inode;
3499 struct ptlrpc_request *req = NULL;
3500 struct obd_export *exp;
3504 LASSERT(inode != NULL);
3506 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
3507 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
3509 exp = ll_i2mdexp(inode);
3511 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
3512 * But under CMD case, it caused some lock issues, should be fixed
3513 * with new CMD ibits lock. See bug 12718 */
3514 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
3515 struct lookup_intent oit = { .it_op = IT_GETATTR };
3516 struct md_op_data *op_data;
3518 if (ibits == MDS_INODELOCK_LOOKUP)
3519 oit.it_op = IT_LOOKUP;
3521 /* Call getattr by fid, so do not provide name at all. */
3522 op_data = ll_prep_md_op_data(NULL, dentry->d_inode,
3523 dentry->d_inode, NULL, 0, 0,
3524 LUSTRE_OPC_ANY, NULL);
3525 if (IS_ERR(op_data))
3526 RETURN(PTR_ERR(op_data));
3528 rc = md_intent_lock(exp, op_data, &oit, &req,
3529 &ll_md_blocking_ast, 0);
3530 ll_finish_md_op_data(op_data);
3532 rc = ll_inode_revalidate_fini(inode, rc);
3536 rc = ll_revalidate_it_finish(req, &oit, dentry);
3538 ll_intent_release(&oit);
3542 /* Unlinked? Unhash dentry, so it is not picked up later by
3543 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
3544 here to preserve get_cwd functionality on 2.6.
3546 if (!dentry->d_inode->i_nlink) {
3547 ll_lock_dcache(inode);
3548 d_lustre_invalidate(dentry, 0);
3549 ll_unlock_dcache(inode);
3552 ll_lookup_finish_locks(&oit, dentry);
3553 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
3554 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
3555 u64 valid = OBD_MD_FLGETATTR;
3556 struct md_op_data *op_data;
3559 if (S_ISREG(inode->i_mode)) {
3560 rc = ll_get_default_mdsize(sbi, &ealen);
3563 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
3566 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
3567 0, ealen, LUSTRE_OPC_ANY,
3569 if (IS_ERR(op_data))
3570 RETURN(PTR_ERR(op_data));
3572 op_data->op_valid = valid;
3573 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3574 ll_finish_md_op_data(op_data);
3576 rc = ll_inode_revalidate_fini(inode, rc);
3580 rc = ll_prep_inode(&inode, req, NULL, NULL);
3583 ptlrpc_req_finished(req);
3587 static int ll_merge_md_attr(struct inode *inode)
3589 struct cl_attr attr = { 0 };
3592 LASSERT(ll_i2info(inode)->lli_lsm_md != NULL);
3593 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
3594 &attr, ll_md_blocking_ast);
3598 set_nlink(inode, attr.cat_nlink);
3599 inode->i_blocks = attr.cat_blocks;
3600 i_size_write(inode, attr.cat_size);
3602 ll_i2info(inode)->lli_atime = attr.cat_atime;
3603 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
3604 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
3610 ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3612 struct inode *inode = dentry->d_inode;
3616 rc = __ll_inode_revalidate(dentry, ibits);
3620 /* if object isn't regular file, don't validate size */
3621 if (!S_ISREG(inode->i_mode)) {
3622 if (S_ISDIR(inode->i_mode) &&
3623 ll_i2info(inode)->lli_lsm_md != NULL) {
3624 rc = ll_merge_md_attr(inode);
3629 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
3630 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
3631 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
3633 /* In case of restore, the MDT has the right size and has
3634 * already send it back without granting the layout lock,
3635 * inode is up-to-date so glimpse is useless.
3636 * Also to glimpse we need the layout, in case of a running
3637 * restore the MDT holds the layout lock so the glimpse will
3638 * block up to the end of restore (getattr will block)
3640 if (!ll_file_test_flag(ll_i2info(inode), LLIF_FILE_RESTORING))
3641 rc = ll_glimpse_size(inode);
3646 static inline dev_t ll_compat_encode_dev(dev_t dev)
3648 /* The compat_sys_*stat*() syscalls will fail unless the
3649 * device majors and minors are both less than 256. Note that
3650 * the value returned here will be passed through
3651 * old_encode_dev() in cp_compat_stat(). And so we are not
3652 * trying to return a valid compat (u16) device number, just
3653 * one that will pass the old_valid_dev() check. */
3655 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
3658 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3660 struct inode *inode = de->d_inode;
3661 struct ll_sb_info *sbi = ll_i2sbi(inode);
3662 struct ll_inode_info *lli = ll_i2info(inode);
3665 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3666 MDS_INODELOCK_LOOKUP);
3667 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3672 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
3674 if (ll_need_32bit_api(sbi)) {
3675 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3676 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
3677 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
3679 stat->ino = inode->i_ino;
3680 stat->dev = inode->i_sb->s_dev;
3681 stat->rdev = inode->i_rdev;
3684 stat->mode = inode->i_mode;
3685 stat->uid = inode->i_uid;
3686 stat->gid = inode->i_gid;
3687 stat->atime = inode->i_atime;
3688 stat->mtime = inode->i_mtime;
3689 stat->ctime = inode->i_ctime;
3690 stat->blksize = 1 << inode->i_blkbits;
3692 stat->nlink = inode->i_nlink;
3693 stat->size = i_size_read(inode);
3694 stat->blocks = inode->i_blocks;
3699 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3700 __u64 start, __u64 len)
3704 struct fiemap *fiemap;
3705 unsigned int extent_count = fieinfo->fi_extents_max;
3707 num_bytes = sizeof(*fiemap) + (extent_count *
3708 sizeof(struct fiemap_extent));
3709 OBD_ALLOC_LARGE(fiemap, num_bytes);
3714 fiemap->fm_flags = fieinfo->fi_flags;
3715 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3716 fiemap->fm_start = start;
3717 fiemap->fm_length = len;
3718 if (extent_count > 0 &&
3719 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3720 sizeof(struct fiemap_extent)) != 0)
3721 GOTO(out, rc = -EFAULT);
3723 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3725 fieinfo->fi_flags = fiemap->fm_flags;
3726 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3727 if (extent_count > 0 &&
3728 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3729 fiemap->fm_mapped_extents *
3730 sizeof(struct fiemap_extent)) != 0)
3731 GOTO(out, rc = -EFAULT);
3733 OBD_FREE_LARGE(fiemap, num_bytes);
3737 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3739 struct ll_inode_info *lli = ll_i2info(inode);
3740 struct posix_acl *acl = NULL;
3743 spin_lock(&lli->lli_lock);
3744 /* VFS' acl_permission_check->check_acl will release the refcount */
3745 acl = posix_acl_dup(lli->lli_posix_acl);
3746 spin_unlock(&lli->lli_lock);
3751 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
3753 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3754 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
3756 ll_check_acl(struct inode *inode, int mask)
3759 # ifdef CONFIG_FS_POSIX_ACL
3760 struct posix_acl *acl;
3764 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3765 if (flags & IPERM_FLAG_RCU)
3768 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
3773 rc = posix_acl_permission(inode, acl, mask);
3774 posix_acl_release(acl);
3777 # else /* !CONFIG_FS_POSIX_ACL */
3779 # endif /* CONFIG_FS_POSIX_ACL */
3781 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
3783 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
3784 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
3786 # ifdef HAVE_INODE_PERMISION_2ARGS
3787 int ll_inode_permission(struct inode *inode, int mask)
3789 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
3794 struct ll_sb_info *sbi;
3795 struct root_squash_info *squash;
3796 struct cred *cred = NULL;
3797 const struct cred *old_cred = NULL;
3799 bool squash_id = false;
3802 #ifdef MAY_NOT_BLOCK
3803 if (mask & MAY_NOT_BLOCK)
3805 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
3806 if (flags & IPERM_FLAG_RCU)
3810 /* as root inode are NOT getting validated in lookup operation,
3811 * need to do it before permission check. */
3813 if (inode == inode->i_sb->s_root->d_inode) {
3814 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3815 MDS_INODELOCK_LOOKUP);
3820 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3821 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3823 /* squash fsuid/fsgid if needed */
3824 sbi = ll_i2sbi(inode);
3825 squash = &sbi->ll_squash;
3826 if (unlikely(squash->rsi_uid != 0 &&
3827 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3828 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3832 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3833 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3834 squash->rsi_uid, squash->rsi_gid);
3836 /* update current process's credentials
3837 * and FS capability */
3838 cred = prepare_creds();
3842 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3843 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3844 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3845 if ((1 << cap) & CFS_CAP_FS_MASK)
3846 cap_lower(cred->cap_effective, cap);
3848 old_cred = override_creds(cred);
3851 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
3852 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
3853 /* restore current process's credentials and FS capability */
3855 revert_creds(old_cred);
3862 /* -o localflock - only provides locally consistent flock locks */
3863 struct file_operations ll_file_operations = {
3864 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
3865 # ifdef HAVE_SYNC_READ_WRITE
3866 .read = new_sync_read,
3867 .write = new_sync_write,
3869 .read_iter = ll_file_read_iter,
3870 .write_iter = ll_file_write_iter,
3871 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3872 .read = ll_file_read,
3873 .aio_read = ll_file_aio_read,
3874 .write = ll_file_write,
3875 .aio_write = ll_file_aio_write,
3876 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3877 .unlocked_ioctl = ll_file_ioctl,
3878 .open = ll_file_open,
3879 .release = ll_file_release,
3880 .mmap = ll_file_mmap,
3881 .llseek = ll_file_seek,
3882 .splice_read = ll_file_splice_read,
3887 struct file_operations ll_file_operations_flock = {
3888 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
3889 # ifdef HAVE_SYNC_READ_WRITE
3890 .read = new_sync_read,
3891 .write = new_sync_write,
3892 # endif /* HAVE_SYNC_READ_WRITE */
3893 .read_iter = ll_file_read_iter,
3894 .write_iter = ll_file_write_iter,
3895 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3896 .read = ll_file_read,
3897 .aio_read = ll_file_aio_read,
3898 .write = ll_file_write,
3899 .aio_write = ll_file_aio_write,
3900 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3901 .unlocked_ioctl = ll_file_ioctl,
3902 .open = ll_file_open,
3903 .release = ll_file_release,
3904 .mmap = ll_file_mmap,
3905 .llseek = ll_file_seek,
3906 .splice_read = ll_file_splice_read,
3909 .flock = ll_file_flock,
3910 .lock = ll_file_flock
3913 /* These are for -o noflock - to return ENOSYS on flock calls */
3914 struct file_operations ll_file_operations_noflock = {
3915 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
3916 # ifdef HAVE_SYNC_READ_WRITE
3917 .read = new_sync_read,
3918 .write = new_sync_write,
3919 # endif /* HAVE_SYNC_READ_WRITE */
3920 .read_iter = ll_file_read_iter,
3921 .write_iter = ll_file_write_iter,
3922 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3923 .read = ll_file_read,
3924 .aio_read = ll_file_aio_read,
3925 .write = ll_file_write,
3926 .aio_write = ll_file_aio_write,
3927 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
3928 .unlocked_ioctl = ll_file_ioctl,
3929 .open = ll_file_open,
3930 .release = ll_file_release,
3931 .mmap = ll_file_mmap,
3932 .llseek = ll_file_seek,
3933 .splice_read = ll_file_splice_read,
3936 .flock = ll_file_noflock,
3937 .lock = ll_file_noflock
3940 struct inode_operations ll_file_inode_operations = {
3941 .setattr = ll_setattr,
3942 .getattr = ll_getattr,
3943 .permission = ll_inode_permission,
3944 .setxattr = ll_setxattr,
3945 .getxattr = ll_getxattr,
3946 .listxattr = ll_listxattr,
3947 .removexattr = ll_removexattr,
3948 .fiemap = ll_fiemap,
3949 #ifdef HAVE_IOP_GET_ACL
3950 .get_acl = ll_get_acl,
3954 /* dynamic ioctl number support routins */
3955 static struct llioc_ctl_data {
3956 struct rw_semaphore ioc_sem;
3957 struct list_head ioc_head;
3959 __RWSEM_INITIALIZER(llioc.ioc_sem),
3960 LIST_HEAD_INIT(llioc.ioc_head)
3965 struct list_head iocd_list;
3966 unsigned int iocd_size;
3967 llioc_callback_t iocd_cb;
3968 unsigned int iocd_count;
3969 unsigned int iocd_cmd[0];
3972 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3975 struct llioc_data *in_data = NULL;
3978 if (cb == NULL || cmd == NULL ||
3979 count > LLIOC_MAX_CMD || count < 0)
3982 size = sizeof(*in_data) + count * sizeof(unsigned int);
3983 OBD_ALLOC(in_data, size);
3984 if (in_data == NULL)
3987 memset(in_data, 0, sizeof(*in_data));
3988 in_data->iocd_size = size;
3989 in_data->iocd_cb = cb;
3990 in_data->iocd_count = count;
3991 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3993 down_write(&llioc.ioc_sem);
3994 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3995 up_write(&llioc.ioc_sem);
4000 void ll_iocontrol_unregister(void *magic)
4002 struct llioc_data *tmp;
4007 down_write(&llioc.ioc_sem);
4008 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
4010 unsigned int size = tmp->iocd_size;
4012 list_del(&tmp->iocd_list);
4013 up_write(&llioc.ioc_sem);
4015 OBD_FREE(tmp, size);
4019 up_write(&llioc.ioc_sem);
4021 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
4024 EXPORT_SYMBOL(ll_iocontrol_register);
4025 EXPORT_SYMBOL(ll_iocontrol_unregister);
4027 static enum llioc_iter
4028 ll_iocontrol_call(struct inode *inode, struct file *file,
4029 unsigned int cmd, unsigned long arg, int *rcp)
4031 enum llioc_iter ret = LLIOC_CONT;
4032 struct llioc_data *data;
4033 int rc = -EINVAL, i;
4035 down_read(&llioc.ioc_sem);
4036 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
4037 for (i = 0; i < data->iocd_count; i++) {
4038 if (cmd != data->iocd_cmd[i])
4041 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
4045 if (ret == LLIOC_STOP)
4048 up_read(&llioc.ioc_sem);
4055 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
4057 struct ll_inode_info *lli = ll_i2info(inode);
4058 struct cl_object *obj = lli->lli_clob;
4067 env = cl_env_get(&refcheck);
4069 RETURN(PTR_ERR(env));
4071 rc = cl_conf_set(env, lli->lli_clob, conf);
4075 if (conf->coc_opc == OBJECT_CONF_SET) {
4076 struct ldlm_lock *lock = conf->coc_lock;
4077 struct cl_layout cl = {
4081 LASSERT(lock != NULL);
4082 LASSERT(ldlm_has_layout(lock));
4084 /* it can only be allowed to match after layout is
4085 * applied to inode otherwise false layout would be
4086 * seen. Applying layout shoud happen before dropping
4087 * the intent lock. */
4088 ldlm_lock_allow_match(lock);
4090 rc = cl_object_layout_get(env, obj, &cl);
4095 DFID": layout version change: %u -> %u\n",
4096 PFID(&lli->lli_fid), ll_layout_version_get(lli),
4098 ll_layout_version_set(lli, cl.cl_layout_gen);
4102 cl_env_put(env, &refcheck);
4107 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
4108 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
4111 struct ll_sb_info *sbi = ll_i2sbi(inode);
4112 struct ptlrpc_request *req;
4113 struct mdt_body *body;
4120 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
4121 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
4122 lock->l_lvb_data, lock->l_lvb_len);
4124 if (lock->l_lvb_data != NULL)
4127 /* if layout lock was granted right away, the layout is returned
4128 * within DLM_LVB of dlm reply; otherwise if the lock was ever
4129 * blocked and then granted via completion ast, we have to fetch
4130 * layout here. Please note that we can't use the LVB buffer in
4131 * completion AST because it doesn't have a large enough buffer */
4132 rc = ll_get_default_mdsize(sbi, &lmmsize);
4134 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
4135 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
4140 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4142 GOTO(out, rc = -EPROTO);
4144 lmmsize = body->mbo_eadatasize;
4145 if (lmmsize == 0) /* empty layout */
4148 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
4150 GOTO(out, rc = -EFAULT);
4152 OBD_ALLOC_LARGE(lvbdata, lmmsize);
4153 if (lvbdata == NULL)
4154 GOTO(out, rc = -ENOMEM);
4156 memcpy(lvbdata, lmm, lmmsize);
4157 lock_res_and_lock(lock);
4158 if (unlikely(lock->l_lvb_data == NULL)) {
4159 lock->l_lvb_type = LVB_T_LAYOUT;
4160 lock->l_lvb_data = lvbdata;
4161 lock->l_lvb_len = lmmsize;
4164 unlock_res_and_lock(lock);
4167 OBD_FREE_LARGE(lvbdata, lmmsize);
4172 ptlrpc_req_finished(req);
4177 * Apply the layout to the inode. Layout lock is held and will be released
4180 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
4181 struct inode *inode)
4183 struct ll_inode_info *lli = ll_i2info(inode);
4184 struct ll_sb_info *sbi = ll_i2sbi(inode);
4185 struct ldlm_lock *lock;
4186 struct cl_object_conf conf;
4189 bool wait_layout = false;
4192 LASSERT(lustre_handle_is_used(lockh));
4194 lock = ldlm_handle2lock(lockh);
4195 LASSERT(lock != NULL);
4196 LASSERT(ldlm_has_layout(lock));
4198 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
4199 PFID(&lli->lli_fid), inode);
4201 /* in case this is a caching lock and reinstate with new inode */
4202 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
4204 lock_res_and_lock(lock);
4205 lvb_ready = ldlm_is_lvb_ready(lock);
4206 unlock_res_and_lock(lock);
4208 /* checking lvb_ready is racy but this is okay. The worst case is
4209 * that multi processes may configure the file on the same time. */
4213 rc = ll_layout_fetch(inode, lock);
4217 /* for layout lock, lmm is stored in lock's lvb.
4218 * lvb_data is immutable if the lock is held so it's safe to access it
4221 * set layout to file. Unlikely this will fail as old layout was
4222 * surely eliminated */
4223 memset(&conf, 0, sizeof conf);
4224 conf.coc_opc = OBJECT_CONF_SET;
4225 conf.coc_inode = inode;
4226 conf.coc_lock = lock;
4227 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
4228 conf.u.coc_layout.lb_len = lock->l_lvb_len;
4229 rc = ll_layout_conf(inode, &conf);
4231 /* refresh layout failed, need to wait */
4232 wait_layout = rc == -EBUSY;
4235 LDLM_LOCK_PUT(lock);
4236 ldlm_lock_decref(lockh, mode);
4238 /* wait for IO to complete if it's still being used. */
4240 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
4241 ll_get_fsname(inode->i_sb, NULL, 0),
4242 PFID(&lli->lli_fid), inode);
4244 memset(&conf, 0, sizeof conf);
4245 conf.coc_opc = OBJECT_CONF_WAIT;
4246 conf.coc_inode = inode;
4247 rc = ll_layout_conf(inode, &conf);
4251 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
4252 ll_get_fsname(inode->i_sb, NULL, 0),
4253 PFID(&lli->lli_fid), rc);
4259 * Issue layout intent RPC to MDS.
4260 * \param inode [in] file inode
4261 * \param intent [in] layout intent
4263 * \retval 0 on success
4264 * \retval < 0 error code
4266 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
4268 struct ll_inode_info *lli = ll_i2info(inode);
4269 struct ll_sb_info *sbi = ll_i2sbi(inode);
4270 struct md_op_data *op_data;
4271 struct lookup_intent it;
4272 struct ptlrpc_request *req;
4276 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
4277 0, 0, LUSTRE_OPC_ANY, NULL);
4278 if (IS_ERR(op_data))
4279 RETURN(PTR_ERR(op_data));
4281 op_data->op_data = intent;
4282 op_data->op_data_size = sizeof(*intent);
4284 memset(&it, 0, sizeof(it));
4285 it.it_op = IT_LAYOUT;
4286 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
4287 intent->li_opc == LAYOUT_INTENT_TRUNC)
4288 it.it_flags = FMODE_WRITE;
4290 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
4291 ll_get_fsname(inode->i_sb, NULL, 0),
4292 PFID(&lli->lli_fid), inode);
4294 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
4295 &ll_md_blocking_ast, 0);
4296 if (it.it_request != NULL)
4297 ptlrpc_req_finished(it.it_request);
4298 it.it_request = NULL;
4300 ll_finish_md_op_data(op_data);
4302 /* set lock data in case this is a new lock */
4304 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
4306 ll_intent_drop_lock(&it);
4312 * This function checks if there exists a LAYOUT lock on the client side,
4313 * or enqueues it if it doesn't have one in cache.
4315 * This function will not hold layout lock so it may be revoked any time after
4316 * this function returns. Any operations depend on layout should be redone
4319 * This function should be called before lov_io_init() to get an uptodate
4320 * layout version, the caller should save the version number and after IO
4321 * is finished, this function should be called again to verify that layout
4322 * is not changed during IO time.
4324 int ll_layout_refresh(struct inode *inode, __u32 *gen)
4326 struct ll_inode_info *lli = ll_i2info(inode);
4327 struct ll_sb_info *sbi = ll_i2sbi(inode);
4328 struct lustre_handle lockh;
4329 struct layout_intent intent = {
4330 .li_opc = LAYOUT_INTENT_ACCESS,
4332 enum ldlm_mode mode;
4336 *gen = ll_layout_version_get(lli);
4337 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
4341 LASSERT(fid_is_sane(ll_inode2fid(inode)));
4342 LASSERT(S_ISREG(inode->i_mode));
4344 /* take layout lock mutex to enqueue layout lock exclusively. */
4345 mutex_lock(&lli->lli_layout_mutex);
4348 /* mostly layout lock is caching on the local side, so try to
4349 * match it before grabbing layout lock mutex. */
4350 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
4351 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
4352 if (mode != 0) { /* hit cached lock */
4353 rc = ll_layout_lock_set(&lockh, mode, inode);
4359 rc = ll_layout_intent(inode, &intent);
4365 *gen = ll_layout_version_get(lli);
4366 mutex_unlock(&lli->lli_layout_mutex);
4372 * Issue layout intent RPC indicating where in a file an IO is about to write.
4374 * \param[in] inode file inode.
4375 * \param[in] start start offset of fille in bytes where an IO is about to
4377 * \param[in] end exclusive end offset in bytes of the write range.
4379 * \retval 0 on success
4380 * \retval < 0 error code
4382 int ll_layout_write_intent(struct inode *inode, __u64 start, __u64 end)
4384 struct layout_intent intent = {
4385 .li_opc = LAYOUT_INTENT_WRITE,
4392 rc = ll_layout_intent(inode, &intent);
4398 * This function send a restore request to the MDT
4400 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
4402 struct hsm_user_request *hur;
4406 len = sizeof(struct hsm_user_request) +
4407 sizeof(struct hsm_user_item);
4408 OBD_ALLOC(hur, len);
4412 hur->hur_request.hr_action = HUA_RESTORE;
4413 hur->hur_request.hr_archive_id = 0;
4414 hur->hur_request.hr_flags = 0;
4415 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
4416 sizeof(hur->hur_user_item[0].hui_fid));
4417 hur->hur_user_item[0].hui_extent.offset = offset;
4418 hur->hur_user_item[0].hui_extent.length = length;
4419 hur->hur_request.hr_itemcount = 1;
4420 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,