4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
40 #include <lustre_dlm.h>
41 #include <linux/pagemap.h>
42 #include <linux/file.h>
43 #include <linux/sched.h>
44 #include <linux/mount.h>
45 #include <uapi/linux/lustre/lustre_fiemap.h>
46 #include <uapi/linux/lustre/lustre_ioctl.h>
47 #include <lustre_swab.h>
49 #include <cl_object.h>
50 #include "llite_internal.h"
53 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
55 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
58 static enum llioc_iter
59 ll_iocontrol_call(struct inode *inode, struct file *file,
60 unsigned int cmd, unsigned long arg, int *rcp);
62 static struct ll_file_data *ll_file_data_get(void)
64 struct ll_file_data *fd;
66 fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS);
69 fd->fd_write_failed = false;
73 static void ll_file_data_put(struct ll_file_data *fd)
76 kmem_cache_free(ll_file_data_slab, fd);
80 * Packs all the attributes into @op_data for the CLOSE rpc.
82 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
83 struct obd_client_handle *och)
85 struct ll_inode_info *lli = ll_i2info(inode);
87 ll_prep_md_op_data(op_data, inode, NULL, NULL,
88 0, 0, LUSTRE_OPC_ANY, NULL);
90 op_data->op_attr.ia_mode = inode->i_mode;
91 op_data->op_attr.ia_atime = inode->i_atime;
92 op_data->op_attr.ia_mtime = inode->i_mtime;
93 op_data->op_attr.ia_ctime = inode->i_ctime;
94 op_data->op_attr.ia_size = i_size_read(inode);
95 op_data->op_attr.ia_valid |= ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
96 ATTR_MTIME | ATTR_MTIME_SET |
97 ATTR_CTIME | ATTR_CTIME_SET;
98 op_data->op_attr_blocks = inode->i_blocks;
99 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
100 op_data->op_handle = och->och_fh;
103 * For HSM: if inode data has been modified, pack it so that
104 * MDT can set data dirty flag in the archive.
106 if (och->och_flags & FMODE_WRITE &&
107 test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags))
108 op_data->op_bias |= MDS_DATA_MODIFIED;
112 * Perform a close, possibly with a bias.
113 * The meaning of "data" depends on the value of "bias".
115 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
116 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
119 static int ll_close_inode_openhandle(struct inode *inode,
120 struct obd_client_handle *och,
121 enum mds_op_bias bias,
124 const struct ll_inode_info *lli = ll_i2info(inode);
125 struct obd_export *md_exp = ll_i2mdexp(inode);
126 struct md_op_data *op_data;
127 struct ptlrpc_request *req = NULL;
130 if (!class_exp2obd(md_exp)) {
131 CERROR("%s: invalid MDC connection handle closing " DFID "\n",
132 ll_get_fsname(inode->i_sb, NULL, 0),
133 PFID(&lli->lli_fid));
138 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
140 * We leak openhandle and request here on error, but not much to be
141 * done in OOM case since app won't retry close on error either.
148 ll_prepare_close(inode, op_data, och);
150 case MDS_CLOSE_LAYOUT_SWAP:
152 op_data->op_bias |= MDS_CLOSE_LAYOUT_SWAP;
153 op_data->op_data_version = 0;
154 op_data->op_lease_handle = och->och_lease_handle;
155 op_data->op_fid2 = *ll_inode2fid(data);
158 case MDS_HSM_RELEASE:
160 op_data->op_bias |= MDS_HSM_RELEASE;
161 op_data->op_data_version = *(__u64 *)data;
162 op_data->op_lease_handle = och->och_lease_handle;
163 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
171 rc = md_close(md_exp, op_data, och->och_mod, &req);
172 if (rc && rc != -EINTR) {
173 CERROR("%s: inode " DFID " mdc close failed: rc = %d\n",
174 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
177 if (op_data->op_bias & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP) &&
179 struct mdt_body *body;
181 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
182 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
186 ll_finish_md_op_data(op_data);
189 md_clear_open_replay_data(md_exp, och);
190 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
193 ptlrpc_req_finished(req);
197 int ll_md_real_close(struct inode *inode, fmode_t fmode)
199 struct ll_inode_info *lli = ll_i2info(inode);
200 struct obd_client_handle **och_p;
201 struct obd_client_handle *och;
205 if (fmode & FMODE_WRITE) {
206 och_p = &lli->lli_mds_write_och;
207 och_usecount = &lli->lli_open_fd_write_count;
208 } else if (fmode & FMODE_EXEC) {
209 och_p = &lli->lli_mds_exec_och;
210 och_usecount = &lli->lli_open_fd_exec_count;
212 LASSERT(fmode & FMODE_READ);
213 och_p = &lli->lli_mds_read_och;
214 och_usecount = &lli->lli_open_fd_read_count;
217 mutex_lock(&lli->lli_och_mutex);
218 if (*och_usecount > 0) {
219 /* There are still users of this handle, so skip
222 mutex_unlock(&lli->lli_och_mutex);
228 mutex_unlock(&lli->lli_och_mutex);
231 /* There might be a race and this handle may already
234 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
240 static int ll_md_close(struct inode *inode, struct file *file)
242 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
243 struct ll_inode_info *lli = ll_i2info(inode);
245 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
246 struct lustre_handle lockh;
247 union ldlm_policy_data policy = {
248 .l_inodebits = { MDS_INODELOCK_OPEN }
252 /* clear group lock, if present */
253 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
254 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
256 if (fd->fd_lease_och) {
259 /* Usually the lease is not released when the
260 * application crashed, we need to release here.
262 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
263 CDEBUG(rc ? D_ERROR : D_INODE,
264 "Clean up lease " DFID " %d/%d\n",
265 PFID(&lli->lli_fid), rc, lease_broken);
267 fd->fd_lease_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
280 mutex_lock(&lli->lli_och_mutex);
281 if (fd->fd_omode & FMODE_WRITE) {
283 LASSERT(lli->lli_open_fd_write_count);
284 lli->lli_open_fd_write_count--;
285 } else if (fd->fd_omode & FMODE_EXEC) {
287 LASSERT(lli->lli_open_fd_exec_count);
288 lli->lli_open_fd_exec_count--;
291 LASSERT(lli->lli_open_fd_read_count);
292 lli->lli_open_fd_read_count--;
294 mutex_unlock(&lli->lli_och_mutex);
296 if (!md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
297 LDLM_IBITS, &policy, lockmode, &lockh))
298 rc = ll_md_real_close(inode, fd->fd_omode);
301 LUSTRE_FPRIVATE(file) = NULL;
302 ll_file_data_put(fd);
307 /* While this returns an error code, fput() the caller does not, so we need
308 * to make every effort to clean up all of our state here. Also, applications
309 * rarely check close errors and even if an error is returned they will not
310 * re-try the close call.
312 int ll_file_release(struct inode *inode, struct file *file)
314 struct ll_file_data *fd;
315 struct ll_sb_info *sbi = ll_i2sbi(inode);
316 struct ll_inode_info *lli = ll_i2info(inode);
319 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p)\n",
320 PFID(ll_inode2fid(inode)), inode);
322 if (!is_root_inode(inode))
323 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
324 fd = LUSTRE_FPRIVATE(file);
327 /* The last ref on @file, maybe not be the owner pid of statahead,
328 * because parent and child process can share the same file handle.
330 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
331 ll_deauthorize_statahead(inode, fd);
333 if (is_root_inode(inode)) {
334 LUSTRE_FPRIVATE(file) = NULL;
335 ll_file_data_put(fd);
339 if (!S_ISDIR(inode->i_mode)) {
341 lov_read_and_clear_async_rc(lli->lli_clob);
342 lli->lli_async_rc = 0;
345 rc = ll_md_close(inode, file);
347 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
348 libcfs_debug_dumplog();
353 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
354 struct lookup_intent *itp)
356 struct inode *inode = d_inode(de);
357 struct ll_sb_info *sbi = ll_i2sbi(inode);
358 struct dentry *parent = de->d_parent;
359 const char *name = NULL;
360 struct md_op_data *op_data;
361 struct ptlrpc_request *req = NULL;
365 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
368 * if server supports open-by-fid, or file name is invalid, don't pack
369 * name in open request
371 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
372 lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
373 name = de->d_name.name;
374 len = de->d_name.len;
377 op_data = ll_prep_md_op_data(NULL, d_inode(parent), inode, name, len,
378 O_RDWR, LUSTRE_OPC_ANY, NULL);
380 return PTR_ERR(op_data);
381 op_data->op_data = lmm;
382 op_data->op_data_size = lmmsize;
384 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
385 &ll_md_blocking_ast, 0);
386 ll_finish_md_op_data(op_data);
388 /* reason for keep own exit path - don`t flood log
389 * with messages with -ESTALE errors.
391 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
392 it_open_error(DISP_OPEN_OPEN, itp))
394 ll_release_openhandle(inode, itp);
398 if (it_disposition(itp, DISP_LOOKUP_NEG)) {
403 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
404 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
405 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
409 rc = ll_prep_inode(&inode, req, NULL, itp);
410 if (!rc && itp->it_lock_mode)
411 ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL);
414 ptlrpc_req_finished(req);
415 ll_intent_drop_lock(itp);
418 * We did open by fid, but by the time we got to the server,
419 * the object disappeared. If this is a create, we cannot really
420 * tell the userspace that the file it was trying to create
421 * does not exist. Instead let's return -ESTALE, and the VFS will
422 * retry the create with LOOKUP_REVAL that we are going to catch
423 * in ll_revalidate_dentry() and use lookup then.
425 if (rc == -ENOENT && itp->it_op & IT_CREAT)
431 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
432 struct obd_client_handle *och)
434 struct mdt_body *body;
436 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
437 och->och_fh = body->mbo_handle;
438 och->och_fid = body->mbo_fid1;
439 och->och_lease_handle.cookie = it->it_lock_handle;
440 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
441 och->och_flags = it->it_flags;
443 return md_set_open_replay_data(md_exp, och, it);
446 static int ll_local_open(struct file *file, struct lookup_intent *it,
447 struct ll_file_data *fd, struct obd_client_handle *och)
449 struct inode *inode = file_inode(file);
451 LASSERT(!LUSTRE_FPRIVATE(file));
458 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
463 LUSTRE_FPRIVATE(file) = fd;
464 ll_readahead_init(inode, &fd->fd_ras);
465 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
467 /* ll_cl_context initialize */
468 rwlock_init(&fd->fd_lock);
469 INIT_LIST_HEAD(&fd->fd_lccs);
474 /* Open a file, and (for the very first open) create objects on the OSTs at
475 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
476 * creation or open until ll_lov_setstripe() ioctl is called.
478 * If we already have the stripe MD locally then we don't request it in
479 * md_open(), by passing a lmm_size = 0.
481 * It is up to the application to ensure no other processes open this file
482 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
483 * used. We might be able to avoid races of that sort by getting lli_open_sem
484 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
485 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
487 int ll_file_open(struct inode *inode, struct file *file)
489 struct ll_inode_info *lli = ll_i2info(inode);
490 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
491 .it_flags = file->f_flags };
492 struct obd_client_handle **och_p = NULL;
493 __u64 *och_usecount = NULL;
494 struct ll_file_data *fd;
497 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p), flags %o\n",
498 PFID(ll_inode2fid(inode)), inode, file->f_flags);
500 it = file->private_data; /* XXX: compat macro */
501 file->private_data = NULL; /* prevent ll_local_open assertion */
503 fd = ll_file_data_get();
510 if (S_ISDIR(inode->i_mode))
511 ll_authorize_statahead(inode, fd);
513 if (is_root_inode(inode)) {
514 LUSTRE_FPRIVATE(file) = fd;
518 if (!it || !it->it_disposition) {
519 /* Convert f_flags into access mode. We cannot use file->f_mode,
520 * because everything but O_ACCMODE mask was stripped from
523 if ((oit.it_flags + 1) & O_ACCMODE)
525 if (file->f_flags & O_TRUNC)
526 oit.it_flags |= FMODE_WRITE;
528 /* kernel only call f_op->open in dentry_open. filp_open calls
529 * dentry_open after call to open_namei that checks permissions.
530 * Only nfsd_open call dentry_open directly without checking
531 * permissions and because of that this code below is safe.
533 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
534 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
536 /* We do not want O_EXCL here, presumably we opened the file
537 * already? XXX - NFS implications?
539 oit.it_flags &= ~O_EXCL;
541 /* bug20584, if "it_flags" contains O_CREAT, the file will be
542 * created if necessary, then "IT_CREAT" should be set to keep
545 if (oit.it_flags & O_CREAT)
546 oit.it_op |= IT_CREAT;
552 /* Let's see if we have file open on MDS already. */
553 if (it->it_flags & FMODE_WRITE) {
554 och_p = &lli->lli_mds_write_och;
555 och_usecount = &lli->lli_open_fd_write_count;
556 } else if (it->it_flags & FMODE_EXEC) {
557 och_p = &lli->lli_mds_exec_och;
558 och_usecount = &lli->lli_open_fd_exec_count;
560 och_p = &lli->lli_mds_read_och;
561 och_usecount = &lli->lli_open_fd_read_count;
564 mutex_lock(&lli->lli_och_mutex);
565 if (*och_p) { /* Open handle is present */
566 if (it_disposition(it, DISP_OPEN_OPEN)) {
567 /* Well, there's extra open request that we do not need,
568 * let's close it somehow. This will decref request.
570 rc = it_open_error(DISP_OPEN_OPEN, it);
572 mutex_unlock(&lli->lli_och_mutex);
576 ll_release_openhandle(inode, it);
580 rc = ll_local_open(file, it, fd, NULL);
583 mutex_unlock(&lli->lli_och_mutex);
587 LASSERT(*och_usecount == 0);
588 if (!it->it_disposition) {
589 /* We cannot just request lock handle now, new ELC code
590 * means that one of other OPEN locks for this file
591 * could be cancelled, and since blocking ast handler
592 * would attempt to grab och_mutex as well, that would
593 * result in a deadlock
595 mutex_unlock(&lli->lli_och_mutex);
597 * Normally called under two situations:
599 * 2. revalidate with IT_OPEN (revalidate doesn't
600 * execute this intent any more).
602 * Always fetch MDS_OPEN_LOCK if this is not setstripe.
604 * Always specify MDS_OPEN_BY_FID because we don't want
605 * to get file with different fid.
607 it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
608 rc = ll_intent_file_open(file->f_path.dentry, NULL, 0, it);
614 *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS);
622 /* md_intent_lock() didn't get a request ref if there was an
623 * open error, so don't do cleanup on the request here
626 /* XXX (green): Should not we bail out on any error here, not
629 rc = it_open_error(DISP_OPEN_OPEN, it);
633 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
634 "inode %p: disposition %x, status %d\n", inode,
635 it_disposition(it, ~0), it->it_status);
637 rc = ll_local_open(file, it, fd, *och_p);
641 mutex_unlock(&lli->lli_och_mutex);
644 /* Must do this outside lli_och_mutex lock to prevent deadlock where
645 * different kind of OPEN lock for this same inode gets cancelled
648 if (!S_ISREG(inode->i_mode))
651 cl_lov_delay_create_clear(&file->f_flags);
656 if (och_p && *och_p) {
661 mutex_unlock(&lli->lli_och_mutex);
664 if (lli->lli_opendir_key == fd)
665 ll_deauthorize_statahead(inode, fd);
667 ll_file_data_put(fd);
669 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
672 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
673 ptlrpc_req_finished(it->it_request);
674 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
680 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
681 struct ldlm_lock_desc *desc,
682 void *data, int flag)
685 struct lustre_handle lockh;
688 case LDLM_CB_BLOCKING:
689 ldlm_lock2handle(lock, &lockh);
690 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
692 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
696 case LDLM_CB_CANCELING:
704 * Acquire a lease and open the file.
706 static struct obd_client_handle *
707 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
710 struct lookup_intent it = { .it_op = IT_OPEN };
711 struct ll_sb_info *sbi = ll_i2sbi(inode);
712 struct md_op_data *op_data;
713 struct ptlrpc_request *req = NULL;
714 struct lustre_handle old_handle = { 0 };
715 struct obd_client_handle *och = NULL;
719 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
720 return ERR_PTR(-EINVAL);
723 struct ll_inode_info *lli = ll_i2info(inode);
724 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
725 struct obd_client_handle **och_p;
728 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
729 return ERR_PTR(-EPERM);
731 /* Get the openhandle of the file */
733 mutex_lock(&lli->lli_och_mutex);
734 if (fd->fd_lease_och) {
735 mutex_unlock(&lli->lli_och_mutex);
740 if (file->f_mode & FMODE_WRITE) {
741 LASSERT(lli->lli_mds_write_och);
742 och_p = &lli->lli_mds_write_och;
743 och_usecount = &lli->lli_open_fd_write_count;
745 LASSERT(lli->lli_mds_read_och);
746 och_p = &lli->lli_mds_read_och;
747 och_usecount = &lli->lli_open_fd_read_count;
749 if (*och_usecount == 1) {
756 mutex_unlock(&lli->lli_och_mutex);
757 if (rc < 0) /* more than 1 opener */
761 old_handle = fd->fd_och->och_fh;
764 och = kzalloc(sizeof(*och), GFP_NOFS);
766 return ERR_PTR(-ENOMEM);
768 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
769 LUSTRE_OPC_ANY, NULL);
770 if (IS_ERR(op_data)) {
771 rc = PTR_ERR(op_data);
775 /* To tell the MDT this openhandle is from the same owner */
776 op_data->op_handle = old_handle;
778 it.it_flags = fmode | open_flags;
779 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
780 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
781 &ll_md_blocking_lease_ast,
782 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
783 * it can be cancelled which may mislead applications that the lease is
785 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
786 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
787 * doesn't deal with openhandle, so normal openhandle will be leaked.
789 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
790 ll_finish_md_op_data(op_data);
791 ptlrpc_req_finished(req);
795 if (it_disposition(&it, DISP_LOOKUP_NEG)) {
800 rc = it_open_error(DISP_OPEN_OPEN, &it);
804 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
805 ll_och_fill(sbi->ll_md_exp, &it, och);
807 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ {
812 /* already get lease, handle lease lock */
813 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
814 if (it.it_lock_mode == 0 ||
815 it.it_lock_bits != MDS_INODELOCK_OPEN) {
816 /* open lock must return for lease */
817 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
818 PFID(ll_inode2fid(inode)), it.it_lock_mode,
824 ll_intent_release(&it);
828 /* Cancel open lock */
829 if (it.it_lock_mode != 0) {
830 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
833 och->och_lease_handle.cookie = 0ULL;
835 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
837 CERROR("%s: error closing file " DFID ": %d\n",
838 ll_get_fsname(inode->i_sb, NULL, 0),
839 PFID(&ll_i2info(inode)->lli_fid), rc2);
840 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
842 ll_intent_release(&it);
849 * Check whether a layout swap can be done between two inodes.
851 * \param[in] inode1 First inode to check
852 * \param[in] inode2 Second inode to check
854 * \retval 0 on success, layout swap can be performed between both inodes
855 * \retval negative error code if requirements are not met
857 static int ll_check_swap_layouts_validity(struct inode *inode1,
858 struct inode *inode2)
860 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
863 if (inode_permission(inode1, MAY_WRITE) ||
864 inode_permission(inode2, MAY_WRITE))
867 if (inode1->i_sb != inode2->i_sb)
873 static int ll_swap_layouts_close(struct obd_client_handle *och,
874 struct inode *inode, struct inode *inode2)
876 const struct lu_fid *fid1 = ll_inode2fid(inode);
877 const struct lu_fid *fid2;
880 CDEBUG(D_INODE, "%s: biased close of file " DFID "\n",
881 ll_get_fsname(inode->i_sb, NULL, 0), PFID(fid1));
883 rc = ll_check_swap_layouts_validity(inode, inode2);
887 /* We now know that inode2 is a lustre inode */
888 fid2 = ll_inode2fid(inode2);
890 rc = lu_fid_cmp(fid1, fid2);
897 * Close the file and swap layouts between inode & inode2.
898 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
899 * because we still need it to pack l_remote_handle to MDT.
901 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
904 och = NULL; /* freed in ll_close_inode_openhandle() */
912 * Release lease and close the file.
913 * It will check if the lease has ever broken.
915 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
918 struct ldlm_lock *lock;
919 bool cancelled = true;
921 lock = ldlm_handle2lock(&och->och_lease_handle);
923 lock_res_and_lock(lock);
924 cancelled = ldlm_is_cancel(lock);
925 unlock_res_and_lock(lock);
929 CDEBUG(D_INODE, "lease for " DFID " broken? %d\n",
930 PFID(&ll_i2info(inode)->lli_fid), cancelled);
933 ldlm_cli_cancel(&och->och_lease_handle, 0);
935 *lease_broken = cancelled;
937 return ll_close_inode_openhandle(inode, och, 0, NULL);
940 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
942 struct ll_inode_info *lli = ll_i2info(inode);
943 struct cl_object *obj = lli->lli_clob;
944 struct cl_attr *attr = vvp_env_thread_attr(env);
950 ll_inode_size_lock(inode);
952 /* merge timestamps the most recently obtained from mds with
953 * timestamps obtained from osts
955 LTIME_S(inode->i_atime) = lli->lli_atime;
956 LTIME_S(inode->i_mtime) = lli->lli_mtime;
957 LTIME_S(inode->i_ctime) = lli->lli_ctime;
959 mtime = LTIME_S(inode->i_mtime);
960 atime = LTIME_S(inode->i_atime);
961 ctime = LTIME_S(inode->i_ctime);
963 cl_object_attr_lock(obj);
964 rc = cl_object_attr_get(env, obj, attr);
965 cl_object_attr_unlock(obj);
968 goto out_size_unlock;
970 if (atime < attr->cat_atime)
971 atime = attr->cat_atime;
973 if (ctime < attr->cat_ctime)
974 ctime = attr->cat_ctime;
976 if (mtime < attr->cat_mtime)
977 mtime = attr->cat_mtime;
979 CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n",
980 PFID(&lli->lli_fid), attr->cat_size);
982 i_size_write(inode, attr->cat_size);
984 inode->i_blocks = attr->cat_blocks;
986 LTIME_S(inode->i_mtime) = mtime;
987 LTIME_S(inode->i_atime) = atime;
988 LTIME_S(inode->i_ctime) = ctime;
991 ll_inode_size_unlock(inode);
996 static bool file_is_noatime(const struct file *file)
998 const struct vfsmount *mnt = file->f_path.mnt;
999 const struct inode *inode = file_inode(file);
1001 /* Adapted from file_accessed() and touch_atime().*/
1002 if (file->f_flags & O_NOATIME)
1005 if (inode->i_flags & S_NOATIME)
1008 if (IS_NOATIME(inode))
1011 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1014 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1017 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1023 static void ll_io_init(struct cl_io *io, const struct file *file, int write)
1025 struct inode *inode = file_inode(file);
1027 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1029 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1030 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1031 file->f_flags & O_DIRECT ||
1034 io->ci_obj = ll_i2info(inode)->lli_clob;
1035 io->ci_lockreq = CILR_MAYBE;
1036 if (ll_file_nolock(file)) {
1037 io->ci_lockreq = CILR_NEVER;
1038 io->ci_no_srvlock = 1;
1039 } else if (file->f_flags & O_APPEND) {
1040 io->ci_lockreq = CILR_MANDATORY;
1043 io->ci_noatime = file_is_noatime(file);
1047 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1048 struct file *file, enum cl_io_type iot,
1049 loff_t *ppos, size_t count)
1051 struct ll_inode_info *lli = ll_i2info(file_inode(file));
1052 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1053 struct vvp_io *vio = vvp_env_io(env);
1054 struct range_lock range;
1059 CDEBUG(D_VFSTRACE, "file: %pD, type: %d ppos: %llu, count: %zu\n",
1060 file, iot, *ppos, count);
1063 io = vvp_env_thread_io(env);
1064 ll_io_init(io, file, iot == CIT_WRITE);
1066 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1067 struct vvp_io *vio = vvp_env_io(env);
1068 bool range_locked = false;
1070 if (file->f_flags & O_APPEND)
1071 range_lock_init(&range, 0, LUSTRE_EOF);
1073 range_lock_init(&range, *ppos, *ppos + count - 1);
1075 vio->vui_fd = LUSTRE_FPRIVATE(file);
1076 vio->vui_iter = args->u.normal.via_iter;
1077 vio->vui_iocb = args->u.normal.via_iocb;
1079 * Direct IO reads must also take range lock,
1080 * or multiple reads will try to work on the same pages
1081 * See LU-6227 for details.
1083 if (((iot == CIT_WRITE) ||
1084 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1085 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1086 CDEBUG(D_VFSTRACE, "Range lock [%llu, %llu]\n",
1087 range.rl_node.in_extent.start,
1088 range.rl_node.in_extent.end);
1089 rc = range_lock(&lli->lli_write_tree, &range);
1093 range_locked = true;
1095 ll_cl_add(file, env, io);
1096 rc = cl_io_loop(env, io);
1097 ll_cl_remove(file, env);
1099 CDEBUG(D_VFSTRACE, "Range unlock [%llu, %llu]\n",
1100 range.rl_node.in_extent.start,
1101 range.rl_node.in_extent.end);
1102 range_unlock(&lli->lli_write_tree, &range);
1105 /* cl_io_rw_init() handled IO */
1109 if (io->ci_nob > 0) {
1110 result = io->ci_nob;
1111 count -= io->ci_nob;
1112 *ppos = io->u.ci_wr.wr.crw_pos;
1114 /* prepare IO restart */
1116 args->u.normal.via_iter = vio->vui_iter;
1119 cl_io_fini(env, io);
1121 if ((!rc || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
1122 CDEBUG(D_VFSTRACE, "%s: restart %s from %lld, count:%zu, result: %zd\n",
1123 file_dentry(file)->d_name.name,
1124 iot == CIT_READ ? "read" : "write",
1125 *ppos, count, result);
1129 if (iot == CIT_READ) {
1131 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1132 LPROC_LL_READ_BYTES, result);
1133 } else if (iot == CIT_WRITE) {
1135 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1136 LPROC_LL_WRITE_BYTES, result);
1137 fd->fd_write_failed = false;
1138 } else if (!result && !rc) {
1141 fd->fd_write_failed = true;
1143 fd->fd_write_failed = false;
1144 } else if (rc != -ERESTARTSYS) {
1145 fd->fd_write_failed = true;
1148 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1150 return result > 0 ? result : rc;
1153 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1156 struct vvp_io_args *args;
1160 env = cl_env_get(&refcheck);
1162 return PTR_ERR(env);
1164 args = ll_env_args(env);
1165 args->u.normal.via_iter = to;
1166 args->u.normal.via_iocb = iocb;
1168 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1169 &iocb->ki_pos, iov_iter_count(to));
1170 cl_env_put(env, &refcheck);
1175 * Write to a file (through the page cache).
1177 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1180 struct vvp_io_args *args;
1184 env = cl_env_get(&refcheck);
1186 return PTR_ERR(env);
1188 args = ll_env_args(env);
1189 args->u.normal.via_iter = from;
1190 args->u.normal.via_iocb = iocb;
1192 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1193 &iocb->ki_pos, iov_iter_count(from));
1194 cl_env_put(env, &refcheck);
1198 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1199 __u64 flags, struct lov_user_md *lum,
1202 struct lookup_intent oit = {
1204 .it_flags = flags | MDS_OPEN_BY_FID,
1208 ll_inode_size_lock(inode);
1209 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
1213 ll_release_openhandle(inode, &oit);
1216 ll_inode_size_unlock(inode);
1217 ll_intent_release(&oit);
1221 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1222 struct lov_mds_md **lmmp, int *lmm_size,
1223 struct ptlrpc_request **request)
1225 struct ll_sb_info *sbi = ll_i2sbi(inode);
1226 struct mdt_body *body;
1227 struct lov_mds_md *lmm = NULL;
1228 struct ptlrpc_request *req = NULL;
1229 struct md_op_data *op_data;
1232 rc = ll_get_default_mdsize(sbi, &lmmsize);
1236 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1237 strlen(filename), lmmsize,
1238 LUSTRE_OPC_ANY, NULL);
1239 if (IS_ERR(op_data))
1240 return PTR_ERR(op_data);
1242 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1243 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1244 ll_finish_md_op_data(op_data);
1246 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
1251 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1253 lmmsize = body->mbo_eadatasize;
1255 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1261 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1263 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1264 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1270 * This is coming from the MDS, so is probably in
1271 * little endian. We convert it to host endian before
1272 * passing it to userspace.
1274 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
1277 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1278 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1281 /* if function called for directory - we should
1282 * avoid swab not existent lsm objects
1284 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1285 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1286 if (S_ISREG(body->mbo_mode))
1287 lustre_swab_lov_user_md_objects(
1288 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1290 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1291 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1292 if (S_ISREG(body->mbo_mode))
1293 lustre_swab_lov_user_md_objects(
1294 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1301 *lmm_size = lmmsize;
1306 static int ll_lov_setea(struct inode *inode, struct file *file,
1309 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1310 struct lov_user_md *lump;
1311 int lum_size = sizeof(struct lov_user_md) +
1312 sizeof(struct lov_user_ost_data);
1315 if (!capable(CFS_CAP_SYS_ADMIN))
1318 lump = libcfs_kvzalloc(lum_size, GFP_NOFS);
1322 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) {
1327 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump,
1329 cl_lov_delay_create_clear(&file->f_flags);
1335 static int ll_file_getstripe(struct inode *inode,
1336 struct lov_user_md __user *lum)
1342 env = cl_env_get(&refcheck);
1344 return PTR_ERR(env);
1346 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum);
1347 cl_env_put(env, &refcheck);
1351 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1354 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
1355 struct lov_user_md *klum;
1357 __u64 flags = FMODE_WRITE;
1359 rc = ll_copy_user_md(lum, &klum);
1364 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, klum,
1366 cl_lov_delay_create_clear(&file->f_flags);
1370 put_user(0, &lum->lmm_stripe_count);
1372 ll_layout_refresh(inode, &gen);
1373 rc = ll_file_getstripe(inode, (struct lov_user_md __user *)arg);
1381 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1383 struct ll_inode_info *lli = ll_i2info(inode);
1384 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1385 struct ll_grouplock grouplock;
1389 CWARN("group id for group lock must not be 0\n");
1393 if (ll_file_nolock(file))
1396 spin_lock(&lli->lli_lock);
1397 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1398 CWARN("group lock already existed with gid %lu\n",
1399 fd->fd_grouplock.lg_gid);
1400 spin_unlock(&lli->lli_lock);
1403 LASSERT(!fd->fd_grouplock.lg_lock);
1404 spin_unlock(&lli->lli_lock);
1406 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1407 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1411 spin_lock(&lli->lli_lock);
1412 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1413 spin_unlock(&lli->lli_lock);
1414 CERROR("another thread just won the race\n");
1415 cl_put_grouplock(&grouplock);
1419 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1420 fd->fd_grouplock = grouplock;
1421 spin_unlock(&lli->lli_lock);
1423 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1427 static int ll_put_grouplock(struct inode *inode, struct file *file,
1430 struct ll_inode_info *lli = ll_i2info(inode);
1431 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1432 struct ll_grouplock grouplock;
1434 spin_lock(&lli->lli_lock);
1435 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1436 spin_unlock(&lli->lli_lock);
1437 CWARN("no group lock held\n");
1440 LASSERT(fd->fd_grouplock.lg_lock);
1442 if (fd->fd_grouplock.lg_gid != arg) {
1443 CWARN("group lock %lu doesn't match current id %lu\n",
1444 arg, fd->fd_grouplock.lg_gid);
1445 spin_unlock(&lli->lli_lock);
1449 grouplock = fd->fd_grouplock;
1450 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1451 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1452 spin_unlock(&lli->lli_lock);
1454 cl_put_grouplock(&grouplock);
1455 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1460 * Close inode open handle
1462 * \param inode [in] inode in question
1463 * \param it [in,out] intent which contains open info and result
1466 * \retval <0 failure
1468 int ll_release_openhandle(struct inode *inode, struct lookup_intent *it)
1470 struct obd_client_handle *och;
1475 /* Root ? Do nothing. */
1476 if (is_root_inode(inode))
1479 /* No open handle to close? Move away */
1480 if (!it_disposition(it, DISP_OPEN_OPEN))
1483 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1485 och = kzalloc(sizeof(*och), GFP_NOFS);
1491 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1493 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
1495 /* this one is in place of ll_file_open */
1496 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1497 ptlrpc_req_finished(it->it_request);
1498 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1504 * Get size for inode for which FIEMAP mapping is requested.
1505 * Make the FIEMAP get_info call and returns the result.
1507 * \param fiemap kernel buffer to hold extens
1508 * \param num_bytes kernel buffer size
1510 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
1513 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
1518 /* Checks for fiemap flags */
1519 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1520 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1524 /* Check for FIEMAP_FLAG_SYNC */
1525 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1526 rc = filemap_fdatawrite(inode->i_mapping);
1531 env = cl_env_get(&refcheck);
1533 return PTR_ERR(env);
1535 if (i_size_read(inode) == 0) {
1536 rc = ll_glimpse_size(inode);
1541 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1542 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
1543 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
1545 /* If filesize is 0, then there would be no objects for mapping */
1546 if (fmkey.lfik_oa.o_size == 0) {
1547 fiemap->fm_mapped_extents = 0;
1552 memcpy(&fmkey.lfik_fiemap, fiemap, sizeof(*fiemap));
1554 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
1555 &fmkey, fiemap, &num_bytes);
1557 cl_env_put(env, &refcheck);
1561 int ll_fid2path(struct inode *inode, void __user *arg)
1563 struct obd_export *exp = ll_i2mdexp(inode);
1564 const struct getinfo_fid2path __user *gfin = arg;
1565 struct getinfo_fid2path *gfout;
1570 if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
1571 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1574 /* Only need to get the buflen */
1575 if (get_user(pathlen, &gfin->gf_pathlen))
1578 if (pathlen > PATH_MAX)
1581 outsize = sizeof(*gfout) + pathlen;
1583 gfout = kzalloc(outsize, GFP_NOFS);
1587 if (copy_from_user(gfout, arg, sizeof(*gfout))) {
1592 /* Call mdc_iocontrol */
1593 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1597 if (copy_to_user(arg, gfout, outsize))
1606 * Read the data_version for inode.
1608 * This value is computed using stripe object version on OST.
1609 * Version is computed using server side locking.
1611 * @param flags if do sync on the OST side;
1613 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1614 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1616 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1618 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1624 /* If no file object initialized, we consider its version is 0. */
1630 env = cl_env_get(&refcheck);
1632 return PTR_ERR(env);
1634 io = vvp_env_thread_io(env);
1636 io->u.ci_data_version.dv_data_version = 0;
1637 io->u.ci_data_version.dv_flags = flags;
1640 if (!cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj))
1641 result = cl_io_loop(env, io);
1643 result = io->ci_result;
1645 *data_version = io->u.ci_data_version.dv_data_version;
1647 cl_io_fini(env, io);
1649 if (unlikely(io->ci_need_restart))
1652 cl_env_put(env, &refcheck);
1658 * Trigger a HSM release request for the provided inode.
1660 int ll_hsm_release(struct inode *inode)
1663 struct obd_client_handle *och = NULL;
1664 __u64 data_version = 0;
1668 CDEBUG(D_INODE, "%s: Releasing file " DFID ".\n",
1669 ll_get_fsname(inode->i_sb, NULL, 0),
1670 PFID(&ll_i2info(inode)->lli_fid));
1672 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
1678 /* Grab latest data_version and [am]time values */
1679 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
1683 env = cl_env_get(&refcheck);
1689 ll_merge_attr(env, inode);
1690 cl_env_put(env, &refcheck);
1692 /* Release the file.
1693 * NB: lease lock handle is released in mdc_hsm_release_pack() because
1694 * we still need it to pack l_remote_handle to MDT.
1696 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
1701 if (och && !IS_ERR(och)) /* close the file */
1702 ll_lease_close(och, inode, NULL);
1707 struct ll_swap_stack {
1710 struct inode *inode1;
1711 struct inode *inode2;
1716 static int ll_swap_layouts(struct file *file1, struct file *file2,
1717 struct lustre_swap_layouts *lsl)
1719 struct mdc_swap_layouts msl;
1720 struct md_op_data *op_data;
1723 struct ll_swap_stack *llss = NULL;
1726 llss = kzalloc(sizeof(*llss), GFP_NOFS);
1730 llss->inode1 = file_inode(file1);
1731 llss->inode2 = file_inode(file2);
1733 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
1737 /* we use 2 bool because it is easier to swap than 2 bits */
1738 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
1739 llss->check_dv1 = true;
1741 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
1742 llss->check_dv2 = true;
1744 /* we cannot use lsl->sl_dvX directly because we may swap them */
1745 llss->dv1 = lsl->sl_dv1;
1746 llss->dv2 = lsl->sl_dv2;
1748 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
1749 if (!rc) /* same file, done! */
1752 if (rc < 0) { /* sequentialize it */
1753 swap(llss->inode1, llss->inode2);
1755 swap(llss->dv1, llss->dv2);
1756 swap(llss->check_dv1, llss->check_dv2);
1760 if (gid != 0) { /* application asks to flush dirty cache */
1761 rc = ll_get_grouplock(llss->inode1, file1, gid);
1765 rc = ll_get_grouplock(llss->inode2, file2, gid);
1767 ll_put_grouplock(llss->inode1, file1, gid);
1772 /* ultimate check, before swapping the layouts we check if
1773 * dataversion has changed (if requested)
1775 if (llss->check_dv1) {
1776 rc = ll_data_version(llss->inode1, &dv, 0);
1779 if (dv != llss->dv1) {
1785 if (llss->check_dv2) {
1786 rc = ll_data_version(llss->inode2, &dv, 0);
1789 if (dv != llss->dv2) {
1795 /* struct md_op_data is used to send the swap args to the mdt
1796 * only flags is missing, so we use struct mdc_swap_layouts
1797 * through the md_op_data->op_data
1799 /* flags from user space have to be converted before they are send to
1800 * server, no flag is sent today, they are only used on the client
1804 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
1805 0, LUSTRE_OPC_ANY, &msl);
1806 if (IS_ERR(op_data)) {
1807 rc = PTR_ERR(op_data);
1811 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
1812 sizeof(*op_data), op_data, NULL);
1813 ll_finish_md_op_data(op_data);
1817 ll_put_grouplock(llss->inode2, file2, gid);
1818 ll_put_grouplock(llss->inode1, file1, gid);
1827 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
1829 struct md_op_data *op_data;
1832 /* Detect out-of range masks */
1833 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
1836 /* Non-root users are forbidden to set or clear flags which are
1837 * NOT defined in HSM_USER_MASK.
1839 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
1840 !capable(CFS_CAP_SYS_ADMIN))
1843 /* Detect out-of range archive id */
1844 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
1845 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
1848 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1849 LUSTRE_OPC_ANY, hss);
1850 if (IS_ERR(op_data))
1851 return PTR_ERR(op_data);
1853 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
1854 sizeof(*op_data), op_data, NULL);
1856 ll_finish_md_op_data(op_data);
1861 static int ll_hsm_import(struct inode *inode, struct file *file,
1862 struct hsm_user_import *hui)
1864 struct hsm_state_set *hss = NULL;
1865 struct iattr *attr = NULL;
1868 if (!S_ISREG(inode->i_mode))
1872 hss = kzalloc(sizeof(*hss), GFP_NOFS);
1876 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
1877 hss->hss_archive_id = hui->hui_archive_id;
1878 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
1879 rc = ll_hsm_state_set(inode, hss);
1883 attr = kzalloc(sizeof(*attr), GFP_NOFS);
1889 attr->ia_mode = hui->hui_mode & 0777;
1890 attr->ia_mode |= S_IFREG;
1891 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
1892 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
1893 attr->ia_size = hui->hui_size;
1894 attr->ia_mtime.tv_sec = hui->hui_mtime;
1895 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
1896 attr->ia_atime.tv_sec = hui->hui_atime;
1897 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
1899 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
1900 ATTR_UID | ATTR_GID |
1901 ATTR_MTIME | ATTR_MTIME_SET |
1902 ATTR_ATIME | ATTR_ATIME_SET;
1906 rc = ll_setattr_raw(file->f_path.dentry, attr, true);
1910 inode_unlock(inode);
1918 static inline long ll_lease_type_from_fmode(fmode_t fmode)
1920 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
1921 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
1925 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1927 struct inode *inode = file_inode(file);
1928 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1931 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p),cmd=%x\n",
1932 PFID(ll_inode2fid(inode)), inode, cmd);
1933 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1935 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1936 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1940 case LL_IOC_GETFLAGS:
1941 /* Get the current value of the file flags */
1942 return put_user(fd->fd_flags, (int __user *)arg);
1943 case LL_IOC_SETFLAGS:
1944 case LL_IOC_CLRFLAGS:
1945 /* Set or clear specific file flags */
1946 /* XXX This probably needs checks to ensure the flags are
1947 * not abused, and to handle any flag side effects.
1949 if (get_user(flags, (int __user *)arg))
1952 if (cmd == LL_IOC_SETFLAGS) {
1953 if ((flags & LL_FILE_IGNORE_LOCK) &&
1954 !(file->f_flags & O_DIRECT)) {
1955 CERROR("%s: unable to disable locking on non-O_DIRECT file\n",
1960 fd->fd_flags |= flags;
1962 fd->fd_flags &= ~flags;
1965 case LL_IOC_LOV_SETSTRIPE:
1966 return ll_lov_setstripe(inode, file, arg);
1967 case LL_IOC_LOV_SETEA:
1968 return ll_lov_setea(inode, file, arg);
1969 case LL_IOC_LOV_SWAP_LAYOUTS: {
1971 struct lustre_swap_layouts lsl;
1973 if (copy_from_user(&lsl, (char __user *)arg,
1974 sizeof(struct lustre_swap_layouts)))
1977 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
1980 file2 = fget(lsl.sl_fd);
1984 /* O_WRONLY or O_RDWR */
1985 if ((file2->f_flags & O_ACCMODE) == O_RDONLY) {
1990 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
1991 struct obd_client_handle *och = NULL;
1992 struct ll_inode_info *lli;
1993 struct inode *inode2;
1995 if (lsl.sl_flags != SWAP_LAYOUTS_CLOSE) {
2000 lli = ll_i2info(inode);
2001 mutex_lock(&lli->lli_och_mutex);
2002 if (fd->fd_lease_och) {
2003 och = fd->fd_lease_och;
2004 fd->fd_lease_och = NULL;
2006 mutex_unlock(&lli->lli_och_mutex);
2011 inode2 = file_inode(file2);
2012 rc = ll_swap_layouts_close(och, inode, inode2);
2014 rc = ll_swap_layouts(file, file2, &lsl);
2020 case LL_IOC_LOV_GETSTRIPE:
2021 return ll_file_getstripe(inode,
2022 (struct lov_user_md __user *)arg);
2023 case FSFILT_IOC_GETFLAGS:
2024 case FSFILT_IOC_SETFLAGS:
2025 return ll_iocontrol(inode, file, cmd, arg);
2026 case FSFILT_IOC_GETVERSION_OLD:
2027 case FSFILT_IOC_GETVERSION:
2028 return put_user(inode->i_generation, (int __user *)arg);
2029 case LL_IOC_GROUP_LOCK:
2030 return ll_get_grouplock(inode, file, arg);
2031 case LL_IOC_GROUP_UNLOCK:
2032 return ll_put_grouplock(inode, file, arg);
2033 case IOC_OBD_STATFS:
2034 return ll_obd_statfs(inode, (void __user *)arg);
2036 /* We need to special case any other ioctls we want to handle,
2037 * to send them to the MDS/OST as appropriate and to properly
2038 * network encode the arg field.
2039 case FSFILT_IOC_SETVERSION_OLD:
2040 case FSFILT_IOC_SETVERSION:
2042 case LL_IOC_FLUSHCTX:
2043 return ll_flush_ctx(inode);
2044 case LL_IOC_PATH2FID: {
2045 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2046 sizeof(struct lu_fid)))
2051 case LL_IOC_GETPARENT:
2052 return ll_getparent(file, (struct getparent __user *)arg);
2053 case OBD_IOC_FID2PATH:
2054 return ll_fid2path(inode, (void __user *)arg);
2055 case LL_IOC_DATA_VERSION: {
2056 struct ioc_data_version idv;
2059 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2062 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2063 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2064 if (rc == 0 && copy_to_user((char __user *)arg, &idv,
2071 case LL_IOC_GET_MDTIDX: {
2074 mdtidx = ll_get_mdt_idx(inode);
2078 if (put_user(mdtidx, (int __user *)arg))
2083 case OBD_IOC_GETDTNAME:
2084 case OBD_IOC_GETMDNAME:
2085 return ll_get_obd_name(inode, cmd, arg);
2086 case LL_IOC_HSM_STATE_GET: {
2087 struct md_op_data *op_data;
2088 struct hsm_user_state *hus;
2091 hus = kzalloc(sizeof(*hus), GFP_NOFS);
2095 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2096 LUSTRE_OPC_ANY, hus);
2097 if (IS_ERR(op_data)) {
2099 return PTR_ERR(op_data);
2102 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2105 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2108 ll_finish_md_op_data(op_data);
2112 case LL_IOC_HSM_STATE_SET: {
2113 struct hsm_state_set *hss;
2116 hss = memdup_user((char __user *)arg, sizeof(*hss));
2118 return PTR_ERR(hss);
2120 rc = ll_hsm_state_set(inode, hss);
2125 case LL_IOC_HSM_ACTION: {
2126 struct md_op_data *op_data;
2127 struct hsm_current_action *hca;
2130 hca = kzalloc(sizeof(*hca), GFP_NOFS);
2134 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2135 LUSTRE_OPC_ANY, hca);
2136 if (IS_ERR(op_data)) {
2138 return PTR_ERR(op_data);
2141 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2144 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2147 ll_finish_md_op_data(op_data);
2151 case LL_IOC_SET_LEASE: {
2152 struct ll_inode_info *lli = ll_i2info(inode);
2153 struct obd_client_handle *och = NULL;
2158 case LL_LEASE_WRLCK:
2159 if (!(file->f_mode & FMODE_WRITE))
2161 fmode = FMODE_WRITE;
2163 case LL_LEASE_RDLCK:
2164 if (!(file->f_mode & FMODE_READ))
2168 case LL_LEASE_UNLCK:
2169 mutex_lock(&lli->lli_och_mutex);
2170 if (fd->fd_lease_och) {
2171 och = fd->fd_lease_och;
2172 fd->fd_lease_och = NULL;
2174 mutex_unlock(&lli->lli_och_mutex);
2179 fmode = och->och_flags;
2180 rc = ll_lease_close(och, inode, &lease_broken);
2187 return ll_lease_type_from_fmode(fmode);
2192 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
2194 /* apply for lease */
2195 och = ll_lease_open(inode, file, fmode, 0);
2197 return PTR_ERR(och);
2200 mutex_lock(&lli->lli_och_mutex);
2201 if (!fd->fd_lease_och) {
2202 fd->fd_lease_och = och;
2205 mutex_unlock(&lli->lli_och_mutex);
2207 /* impossible now that only excl is supported for now */
2208 ll_lease_close(och, inode, &lease_broken);
2213 case LL_IOC_GET_LEASE: {
2214 struct ll_inode_info *lli = ll_i2info(inode);
2215 struct ldlm_lock *lock = NULL;
2218 mutex_lock(&lli->lli_och_mutex);
2219 if (fd->fd_lease_och) {
2220 struct obd_client_handle *och = fd->fd_lease_och;
2222 lock = ldlm_handle2lock(&och->och_lease_handle);
2224 lock_res_and_lock(lock);
2225 if (!ldlm_is_cancel(lock))
2226 fmode = och->och_flags;
2227 unlock_res_and_lock(lock);
2228 LDLM_LOCK_PUT(lock);
2231 mutex_unlock(&lli->lli_och_mutex);
2232 return ll_lease_type_from_fmode(fmode);
2234 case LL_IOC_HSM_IMPORT: {
2235 struct hsm_user_import *hui;
2237 hui = memdup_user((void __user *)arg, sizeof(*hui));
2239 return PTR_ERR(hui);
2241 rc = ll_hsm_import(inode, file, hui);
2249 if (ll_iocontrol_call(inode, file, cmd, arg, &err) ==
2253 return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2254 (void __user *)arg);
2259 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2261 struct inode *inode = file_inode(file);
2262 loff_t retval, eof = 0;
2264 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2265 (origin == SEEK_CUR) ? file->f_pos : 0);
2266 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p), to=%llu=%#llx(%d)\n",
2267 PFID(ll_inode2fid(inode)), inode, retval, retval, origin);
2268 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2270 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2271 retval = ll_glimpse_size(inode);
2274 eof = i_size_read(inode);
2277 return generic_file_llseek_size(file, offset, origin,
2278 ll_file_maxbytes(inode), eof);
2281 static int ll_flush(struct file *file, fl_owner_t id)
2283 struct inode *inode = file_inode(file);
2284 struct ll_inode_info *lli = ll_i2info(inode);
2285 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2288 LASSERT(!S_ISDIR(inode->i_mode));
2290 /* catch async errors that were recorded back when async writeback
2291 * failed for pages in this mapping.
2293 rc = lli->lli_async_rc;
2294 lli->lli_async_rc = 0;
2295 if (lli->lli_clob) {
2296 err = lov_read_and_clear_async_rc(lli->lli_clob);
2301 /* The application has been told about write failure already.
2302 * Do not report failure again.
2304 if (fd->fd_write_failed)
2306 return rc ? -EIO : 0;
2310 * Called to make sure a portion of file has been written out.
2311 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2313 * Return how many pages have been written.
2315 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2316 enum cl_fsync_mode mode, int ignore_layout)
2320 struct cl_fsync_io *fio;
2324 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2325 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2328 env = cl_env_get(&refcheck);
2330 return PTR_ERR(env);
2332 io = vvp_env_thread_io(env);
2333 io->ci_obj = ll_i2info(inode)->lli_clob;
2334 io->ci_ignore_layout = ignore_layout;
2336 /* initialize parameters for sync */
2337 fio = &io->u.ci_fsync;
2338 fio->fi_start = start;
2340 fio->fi_fid = ll_inode2fid(inode);
2341 fio->fi_mode = mode;
2342 fio->fi_nr_written = 0;
2344 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2345 result = cl_io_loop(env, io);
2347 result = io->ci_result;
2349 result = fio->fi_nr_written;
2350 cl_io_fini(env, io);
2351 cl_env_put(env, &refcheck);
2356 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2358 struct inode *inode = file_inode(file);
2359 struct ll_inode_info *lli = ll_i2info(inode);
2360 struct ptlrpc_request *req;
2363 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p)\n",
2364 PFID(ll_inode2fid(inode)), inode);
2365 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2367 rc = file_write_and_wait_range(file, start, end);
2370 /* catch async errors that were recorded back when async writeback
2371 * failed for pages in this mapping.
2373 if (!S_ISDIR(inode->i_mode)) {
2374 err = lli->lli_async_rc;
2375 lli->lli_async_rc = 0;
2378 if (lli->lli_clob) {
2379 err = lov_read_and_clear_async_rc(lli->lli_clob);
2385 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
2389 ptlrpc_req_finished(req);
2391 if (S_ISREG(inode->i_mode)) {
2392 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2394 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2395 if (rc == 0 && err < 0)
2398 fd->fd_write_failed = true;
2400 fd->fd_write_failed = false;
2403 inode_unlock(inode);
2408 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2410 struct inode *inode = file_inode(file);
2411 struct ll_sb_info *sbi = ll_i2sbi(inode);
2412 struct ldlm_enqueue_info einfo = {
2413 .ei_type = LDLM_FLOCK,
2414 .ei_cb_cp = ldlm_flock_completion_ast,
2415 .ei_cbdata = file_lock,
2417 struct md_op_data *op_data;
2418 struct lustre_handle lockh = {0};
2419 union ldlm_policy_data flock = { { 0 } };
2420 int fl_type = file_lock->fl_type;
2425 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID " file_lock=%p\n",
2426 PFID(ll_inode2fid(inode)), file_lock);
2428 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2430 if (file_lock->fl_flags & FL_FLOCK)
2431 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2432 else if (!(file_lock->fl_flags & FL_POSIX))
2435 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2436 flock.l_flock.pid = file_lock->fl_pid;
2437 flock.l_flock.start = file_lock->fl_start;
2438 flock.l_flock.end = file_lock->fl_end;
2440 /* Somewhat ugly workaround for svc lockd.
2441 * lockd installs custom fl_lmops->lm_compare_owner that checks
2442 * for the fl_owner to be the same (which it always is on local node
2443 * I guess between lockd processes) and then compares pid.
2444 * As such we assign pid to the owner field to make it all work,
2445 * conflict with normal locks is unlikely since pid space and
2446 * pointer space for current->files are not intersecting
2448 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2449 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2453 einfo.ei_mode = LCK_PR;
2456 /* An unlock request may or may not have any relation to
2457 * existing locks so we may not be able to pass a lock handle
2458 * via a normal ldlm_lock_cancel() request. The request may even
2459 * unlock a byte range in the middle of an existing lock. In
2460 * order to process an unlock request we need all of the same
2461 * information that is given with a normal read or write record
2462 * lock request. To avoid creating another ldlm unlock (cancel)
2463 * message we'll treat a LCK_NL flock request as an unlock.
2465 einfo.ei_mode = LCK_NL;
2468 einfo.ei_mode = LCK_PW;
2471 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
2486 flags = LDLM_FL_BLOCK_NOWAIT;
2492 flags = LDLM_FL_TEST_LOCK;
2495 CERROR("unknown fcntl lock command: %d\n", cmd);
2500 * Save the old mode so that if the mode in the lock changes we
2501 * can decrement the appropriate reader or writer refcount.
2503 file_lock->fl_type = einfo.ei_mode;
2505 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2506 LUSTRE_OPC_ANY, NULL);
2507 if (IS_ERR(op_data))
2508 return PTR_ERR(op_data);
2510 CDEBUG(D_DLMTRACE, "inode=" DFID ", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n",
2511 PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags,
2512 einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2514 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
2517 /* Restore the file lock type if not TEST lock. */
2518 if (!(flags & LDLM_FL_TEST_LOCK))
2519 file_lock->fl_type = fl_type;
2521 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
2522 !(flags & LDLM_FL_TEST_LOCK))
2523 rc2 = locks_lock_file_wait(file, file_lock);
2525 if (rc2 && file_lock->fl_type != F_UNLCK) {
2526 einfo.ei_mode = LCK_NL;
2527 md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
2532 ll_finish_md_op_data(op_data);
2537 int ll_get_fid_by_name(struct inode *parent, const char *name,
2538 int namelen, struct lu_fid *fid,
2539 struct inode **inode)
2541 struct md_op_data *op_data = NULL;
2542 struct ptlrpc_request *req;
2543 struct mdt_body *body;
2546 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
2547 LUSTRE_OPC_ANY, NULL);
2548 if (IS_ERR(op_data))
2549 return PTR_ERR(op_data);
2551 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
2552 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
2553 ll_finish_md_op_data(op_data);
2557 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2563 *fid = body->mbo_fid1;
2566 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
2568 ptlrpc_req_finished(req);
2572 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
2573 const char *name, int namelen)
2575 struct ptlrpc_request *request = NULL;
2576 struct obd_client_handle *och = NULL;
2577 struct inode *child_inode = NULL;
2578 struct dentry *dchild = NULL;
2579 struct md_op_data *op_data;
2580 struct mdt_body *body;
2581 u64 data_version = 0;
2585 CDEBUG(D_VFSTRACE, "migrate %s under " DFID " to MDT%d\n",
2586 name, PFID(ll_inode2fid(parent)), mdtidx);
2588 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
2589 0, LUSTRE_OPC_ANY, NULL);
2590 if (IS_ERR(op_data))
2591 return PTR_ERR(op_data);
2593 /* Get child FID first */
2594 qstr.hash = full_name_hash(parent, name, namelen);
2597 dchild = d_lookup(file_dentry(file), &qstr);
2599 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2600 if (dchild->d_inode)
2601 child_inode = igrab(dchild->d_inode);
2606 rc = ll_get_fid_by_name(parent, name, namelen,
2607 &op_data->op_fid3, &child_inode);
2617 inode_lock(child_inode);
2618 op_data->op_fid3 = *ll_inode2fid(child_inode);
2619 if (!fid_is_sane(&op_data->op_fid3)) {
2620 CERROR("%s: migrate %s, but fid " DFID " is insane\n",
2621 ll_get_fsname(parent->i_sb, NULL, 0), name,
2622 PFID(&op_data->op_fid3));
2627 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
2632 CDEBUG(D_INFO, "%s: " DFID " is already on MDT%d.\n", name,
2633 PFID(&op_data->op_fid3), mdtidx);
2638 if (S_ISREG(child_inode->i_mode)) {
2639 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
2646 rc = ll_data_version(child_inode, &data_version,
2651 op_data->op_handle = och->och_fh;
2652 op_data->op_data = och->och_mod;
2653 op_data->op_data_version = data_version;
2654 op_data->op_lease_handle = och->och_lease_handle;
2655 op_data->op_bias |= MDS_RENAME_MIGRATE;
2658 op_data->op_mds = mdtidx;
2659 op_data->op_cli_flags = CLI_MIGRATE;
2660 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
2661 namelen, name, namelen, &request);
2664 ll_update_times(request, parent);
2666 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2670 * If the server does release layout lock, then we cleanup
2671 * the client och here, otherwise release it in out_close:
2673 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
2674 obd_mod_put(och->och_mod);
2675 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
2677 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
2684 ptlrpc_req_finished(request);
2688 /* Try again if the file layout has changed. */
2689 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
2693 if (och) /* close the file */
2694 ll_lease_close(och, child_inode, NULL);
2696 clear_nlink(child_inode);
2698 inode_unlock(child_inode);
2701 ll_finish_md_op_data(op_data);
2706 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2712 * test if some locks matching bits and l_req_mode are acquired
2713 * - bits can be in different locks
2714 * - if found clear the common lock bits in *bits
2715 * - the bits not found, are kept in *bits
2717 * \param bits [IN] searched lock bits [IN]
2718 * \param l_req_mode [IN] searched lock mode
2719 * \retval boolean, true iff all bits are found
2721 int ll_have_md_lock(struct inode *inode, __u64 *bits,
2722 enum ldlm_mode l_req_mode)
2724 struct lustre_handle lockh;
2725 union ldlm_policy_data policy;
2726 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
2727 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
2735 fid = &ll_i2info(inode)->lli_fid;
2736 CDEBUG(D_INFO, "trying to match res " DFID " mode %s\n", PFID(fid),
2737 ldlm_lockname[mode]);
2739 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2740 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2741 policy.l_inodebits.bits = *bits & (1 << i);
2742 if (policy.l_inodebits.bits == 0)
2745 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2746 &policy, mode, &lockh)) {
2747 struct ldlm_lock *lock;
2749 lock = ldlm_handle2lock(&lockh);
2752 ~(lock->l_policy_data.l_inodebits.bits);
2753 LDLM_LOCK_PUT(lock);
2755 *bits &= ~policy.l_inodebits.bits;
2762 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
2763 struct lustre_handle *lockh, __u64 flags,
2764 enum ldlm_mode mode)
2766 union ldlm_policy_data policy = { .l_inodebits = { bits } };
2769 fid = &ll_i2info(inode)->lli_fid;
2770 CDEBUG(D_INFO, "trying to match res " DFID "\n", PFID(fid));
2772 return md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED,
2773 fid, LDLM_IBITS, &policy, mode, lockh);
2776 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2778 /* Already unlinked. Just update nlink and return success */
2779 if (rc == -ENOENT) {
2781 /* If it is striped directory, and there is bad stripe
2782 * Let's revalidate the dentry again, instead of returning
2785 if (S_ISDIR(inode->i_mode) && ll_i2info(inode)->lli_lsm_md)
2788 /* This path cannot be hit for regular files unless in
2789 * case of obscure races, so no need to validate size.
2791 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2793 } else if (rc != 0) {
2794 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
2795 "%s: revalidate FID " DFID " error: rc = %d\n",
2796 ll_get_fsname(inode->i_sb, NULL, 0),
2797 PFID(ll_inode2fid(inode)), rc);
2803 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2805 struct inode *inode = d_inode(dentry);
2806 struct ptlrpc_request *req = NULL;
2807 struct obd_export *exp;
2810 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p),name=%pd\n",
2811 PFID(ll_inode2fid(inode)), inode, dentry);
2813 exp = ll_i2mdexp(inode);
2815 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2816 * But under CMD case, it caused some lock issues, should be fixed
2817 * with new CMD ibits lock. See bug 12718
2819 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
2820 struct lookup_intent oit = { .it_op = IT_GETATTR };
2821 struct md_op_data *op_data;
2823 if (ibits == MDS_INODELOCK_LOOKUP)
2824 oit.it_op = IT_LOOKUP;
2826 /* Call getattr by fid, so do not provide name at all. */
2827 op_data = ll_prep_md_op_data(NULL, inode,
2829 LUSTRE_OPC_ANY, NULL);
2830 if (IS_ERR(op_data))
2831 return PTR_ERR(op_data);
2833 rc = md_intent_lock(exp, op_data, &oit, &req,
2834 &ll_md_blocking_ast, 0);
2835 ll_finish_md_op_data(op_data);
2837 rc = ll_inode_revalidate_fini(inode, rc);
2841 rc = ll_revalidate_it_finish(req, &oit, inode);
2843 ll_intent_release(&oit);
2847 /* Unlinked? Unhash dentry, so it is not picked up later by
2848 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2849 * here to preserve get_cwd functionality on 2.6.
2852 if (!d_inode(dentry)->i_nlink) {
2853 spin_lock(&inode->i_lock);
2854 d_lustre_invalidate(dentry, 0);
2855 spin_unlock(&inode->i_lock);
2858 ll_lookup_finish_locks(&oit, inode);
2859 } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) {
2860 struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry));
2861 u64 valid = OBD_MD_FLGETATTR;
2862 struct md_op_data *op_data;
2865 if (S_ISREG(inode->i_mode)) {
2866 rc = ll_get_default_mdsize(sbi, &ealen);
2869 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2872 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2873 0, ealen, LUSTRE_OPC_ANY,
2875 if (IS_ERR(op_data))
2876 return PTR_ERR(op_data);
2878 op_data->op_valid = valid;
2879 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2880 ll_finish_md_op_data(op_data);
2882 return ll_inode_revalidate_fini(inode, rc);
2884 rc = ll_prep_inode(&inode, req, NULL, NULL);
2887 ptlrpc_req_finished(req);
2891 static int ll_merge_md_attr(struct inode *inode)
2893 struct cl_attr attr = { 0 };
2896 LASSERT(ll_i2info(inode)->lli_lsm_md);
2897 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
2898 &attr, ll_md_blocking_ast);
2902 set_nlink(inode, attr.cat_nlink);
2903 inode->i_blocks = attr.cat_blocks;
2904 i_size_write(inode, attr.cat_size);
2906 ll_i2info(inode)->lli_atime = attr.cat_atime;
2907 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
2908 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
2913 static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2915 struct inode *inode = d_inode(dentry);
2918 rc = __ll_inode_revalidate(dentry, ibits);
2922 /* if object isn't regular file, don't validate size */
2923 if (!S_ISREG(inode->i_mode)) {
2924 if (S_ISDIR(inode->i_mode) &&
2925 ll_i2info(inode)->lli_lsm_md) {
2926 rc = ll_merge_md_attr(inode);
2931 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
2932 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
2933 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
2935 struct ll_inode_info *lli = ll_i2info(inode);
2937 /* In case of restore, the MDT has the right size and has
2938 * already send it back without granting the layout lock,
2939 * inode is up-to-date so glimpse is useless.
2940 * Also to glimpse we need the layout, in case of a running
2941 * restore the MDT holds the layout lock so the glimpse will
2942 * block up to the end of restore (getattr will block)
2944 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags))
2945 rc = ll_glimpse_size(inode);
2950 int ll_getattr(const struct path *path, struct kstat *stat,
2951 u32 request_mask, unsigned int flags)
2953 struct inode *inode = d_inode(path->dentry);
2954 struct ll_sb_info *sbi = ll_i2sbi(inode);
2955 struct ll_inode_info *lli = ll_i2info(inode);
2958 res = ll_inode_revalidate(path->dentry,
2959 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP);
2960 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
2965 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
2967 stat->dev = inode->i_sb->s_dev;
2968 if (ll_need_32bit_api(sbi))
2969 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
2971 stat->ino = inode->i_ino;
2972 stat->mode = inode->i_mode;
2973 stat->uid = inode->i_uid;
2974 stat->gid = inode->i_gid;
2975 stat->rdev = inode->i_rdev;
2976 stat->atime = inode->i_atime;
2977 stat->mtime = inode->i_mtime;
2978 stat->ctime = inode->i_ctime;
2979 stat->blksize = 1 << inode->i_blkbits;
2981 stat->nlink = inode->i_nlink;
2982 stat->size = i_size_read(inode);
2983 stat->blocks = inode->i_blocks;
2988 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2989 __u64 start, __u64 len)
2993 struct fiemap *fiemap;
2994 unsigned int extent_count = fieinfo->fi_extents_max;
2996 num_bytes = sizeof(*fiemap) + (extent_count *
2997 sizeof(struct fiemap_extent));
2998 fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS);
3002 fiemap->fm_flags = fieinfo->fi_flags;
3003 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3004 fiemap->fm_start = start;
3005 fiemap->fm_length = len;
3007 if (extent_count > 0 &&
3008 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3009 sizeof(struct fiemap_extent))) {
3014 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3016 fieinfo->fi_flags = fiemap->fm_flags;
3017 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3018 if (extent_count > 0 &&
3019 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3020 fiemap->fm_mapped_extents *
3021 sizeof(struct fiemap_extent))) {
3030 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3032 struct ll_inode_info *lli = ll_i2info(inode);
3033 struct posix_acl *acl = NULL;
3035 spin_lock(&lli->lli_lock);
3036 /* VFS' acl_permission_check->check_acl will release the refcount */
3037 acl = posix_acl_dup(lli->lli_posix_acl);
3038 spin_unlock(&lli->lli_lock);
3043 int ll_inode_permission(struct inode *inode, int mask)
3045 struct ll_sb_info *sbi;
3046 struct root_squash_info *squash;
3047 const struct cred *old_cred = NULL;
3048 struct cred *cred = NULL;
3049 bool squash_id = false;
3053 if (mask & MAY_NOT_BLOCK)
3056 /* as root inode are NOT getting validated in lookup operation,
3057 * need to do it before permission check.
3060 if (is_root_inode(inode)) {
3061 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3062 MDS_INODELOCK_LOOKUP);
3067 CDEBUG(D_VFSTRACE, "VFS Op:inode=" DFID "(%p), inode mode %x mask %o\n",
3068 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3070 /* squash fsuid/fsgid if needed */
3071 sbi = ll_i2sbi(inode);
3072 squash = &sbi->ll_squash;
3073 if (unlikely(squash->rsi_uid &&
3074 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3075 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3080 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3081 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3082 squash->rsi_uid, squash->rsi_gid);
3085 * update current process's credentials
3088 cred = prepare_creds();
3092 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3093 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3094 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3095 if ((1 << cap) & CFS_CAP_FS_MASK)
3096 cap_lower(cred->cap_effective, cap);
3098 old_cred = override_creds(cred);
3101 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
3102 rc = generic_permission(inode, mask);
3104 /* restore current process's credentials and FS capability */
3106 revert_creds(old_cred);
3113 /* -o localflock - only provides locally consistent flock locks */
3114 const struct file_operations ll_file_operations = {
3115 .read_iter = ll_file_read_iter,
3116 .write_iter = ll_file_write_iter,
3117 .unlocked_ioctl = ll_file_ioctl,
3118 .open = ll_file_open,
3119 .release = ll_file_release,
3120 .mmap = ll_file_mmap,
3121 .llseek = ll_file_seek,
3122 .splice_read = generic_file_splice_read,
3127 const struct file_operations ll_file_operations_flock = {
3128 .read_iter = ll_file_read_iter,
3129 .write_iter = ll_file_write_iter,
3130 .unlocked_ioctl = ll_file_ioctl,
3131 .open = ll_file_open,
3132 .release = ll_file_release,
3133 .mmap = ll_file_mmap,
3134 .llseek = ll_file_seek,
3135 .splice_read = generic_file_splice_read,
3138 .flock = ll_file_flock,
3139 .lock = ll_file_flock
3142 /* These are for -o noflock - to return ENOSYS on flock calls */
3143 const struct file_operations ll_file_operations_noflock = {
3144 .read_iter = ll_file_read_iter,
3145 .write_iter = ll_file_write_iter,
3146 .unlocked_ioctl = ll_file_ioctl,
3147 .open = ll_file_open,
3148 .release = ll_file_release,
3149 .mmap = ll_file_mmap,
3150 .llseek = ll_file_seek,
3151 .splice_read = generic_file_splice_read,
3154 .flock = ll_file_noflock,
3155 .lock = ll_file_noflock
3158 const struct inode_operations ll_file_inode_operations = {
3159 .setattr = ll_setattr,
3160 .getattr = ll_getattr,
3161 .permission = ll_inode_permission,
3162 .listxattr = ll_listxattr,
3163 .fiemap = ll_fiemap,
3164 .get_acl = ll_get_acl,
3167 /* dynamic ioctl number support routines */
3168 static struct llioc_ctl_data {
3169 struct rw_semaphore ioc_sem;
3170 struct list_head ioc_head;
3172 __RWSEM_INITIALIZER(llioc.ioc_sem),
3173 LIST_HEAD_INIT(llioc.ioc_head)
3177 struct list_head iocd_list;
3178 unsigned int iocd_size;
3179 llioc_callback_t iocd_cb;
3180 unsigned int iocd_count;
3181 unsigned int iocd_cmd[0];
3184 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3187 struct llioc_data *in_data = NULL;
3189 if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0)
3192 size = sizeof(*in_data) + count * sizeof(unsigned int);
3193 in_data = kzalloc(size, GFP_NOFS);
3197 in_data->iocd_size = size;
3198 in_data->iocd_cb = cb;
3199 in_data->iocd_count = count;
3200 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3202 down_write(&llioc.ioc_sem);
3203 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3204 up_write(&llioc.ioc_sem);
3208 EXPORT_SYMBOL(ll_iocontrol_register);
3210 void ll_iocontrol_unregister(void *magic)
3212 struct llioc_data *tmp;
3217 down_write(&llioc.ioc_sem);
3218 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3220 list_del(&tmp->iocd_list);
3221 up_write(&llioc.ioc_sem);
3227 up_write(&llioc.ioc_sem);
3229 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3231 EXPORT_SYMBOL(ll_iocontrol_unregister);
3233 static enum llioc_iter
3234 ll_iocontrol_call(struct inode *inode, struct file *file,
3235 unsigned int cmd, unsigned long arg, int *rcp)
3237 enum llioc_iter ret = LLIOC_CONT;
3238 struct llioc_data *data;
3239 int rc = -EINVAL, i;
3241 down_read(&llioc.ioc_sem);
3242 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3243 for (i = 0; i < data->iocd_count; i++) {
3244 if (cmd != data->iocd_cmd[i])
3247 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3251 if (ret == LLIOC_STOP)
3254 up_read(&llioc.ioc_sem);
3261 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3263 struct ll_inode_info *lli = ll_i2info(inode);
3264 struct cl_object *obj = lli->lli_clob;
3272 env = cl_env_get(&refcheck);
3274 return PTR_ERR(env);
3276 rc = cl_conf_set(env, obj, conf);
3280 if (conf->coc_opc == OBJECT_CONF_SET) {
3281 struct ldlm_lock *lock = conf->coc_lock;
3282 struct cl_layout cl = {
3287 LASSERT(ldlm_has_layout(lock));
3289 /* it can only be allowed to match after layout is
3290 * applied to inode otherwise false layout would be
3291 * seen. Applying layout should happen before dropping
3294 ldlm_lock_allow_match(lock);
3296 rc = cl_object_layout_get(env, obj, &cl);
3300 CDEBUG(D_VFSTRACE, DFID ": layout version change: %u -> %u\n",
3301 PFID(&lli->lli_fid), ll_layout_version_get(lli),
3303 ll_layout_version_set(lli, cl.cl_layout_gen);
3306 cl_env_put(env, &refcheck);
3310 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3311 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3314 struct ll_sb_info *sbi = ll_i2sbi(inode);
3315 struct ptlrpc_request *req;
3316 struct mdt_body *body;
3322 CDEBUG(D_INODE, DFID " LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3323 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3324 lock->l_lvb_data, lock->l_lvb_len);
3326 if (lock->l_lvb_data && ldlm_is_lvb_ready(lock))
3329 /* if layout lock was granted right away, the layout is returned
3330 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3331 * blocked and then granted via completion ast, we have to fetch
3332 * layout here. Please note that we can't use the LVB buffer in
3333 * completion AST because it doesn't have a large enough buffer
3335 rc = ll_get_default_mdsize(sbi, &lmmsize);
3337 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
3338 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3343 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3349 lmmsize = body->mbo_eadatasize;
3350 if (lmmsize == 0) /* empty layout */ {
3355 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3361 lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS);
3367 memcpy(lvbdata, lmm, lmmsize);
3368 lock_res_and_lock(lock);
3369 if (lock->l_lvb_data)
3370 kvfree(lock->l_lvb_data);
3372 lock->l_lvb_data = lvbdata;
3373 lock->l_lvb_len = lmmsize;
3374 unlock_res_and_lock(lock);
3377 ptlrpc_req_finished(req);
3382 * Apply the layout to the inode. Layout lock is held and will be released
3385 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
3386 struct inode *inode)
3388 struct ll_inode_info *lli = ll_i2info(inode);
3389 struct ll_sb_info *sbi = ll_i2sbi(inode);
3390 struct ldlm_lock *lock;
3391 struct cl_object_conf conf;
3394 bool wait_layout = false;
3396 LASSERT(lustre_handle_is_used(lockh));
3398 lock = ldlm_handle2lock(lockh);
3400 LASSERT(ldlm_has_layout(lock));
3402 LDLM_DEBUG(lock, "File " DFID "(%p) being reconfigured",
3403 PFID(&lli->lli_fid), inode);
3405 /* in case this is a caching lock and reinstate with new inode */
3406 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
3408 lock_res_and_lock(lock);
3409 lvb_ready = ldlm_is_lvb_ready(lock);
3410 unlock_res_and_lock(lock);
3411 /* checking lvb_ready is racy but this is okay. The worst case is
3412 * that multi processes may configure the file on the same time.
3419 rc = ll_layout_fetch(inode, lock);
3423 /* for layout lock, lmm is returned in lock's lvb.
3424 * lvb_data is immutable if the lock is held so it's safe to access it
3427 * set layout to file. Unlikely this will fail as old layout was
3430 memset(&conf, 0, sizeof(conf));
3431 conf.coc_opc = OBJECT_CONF_SET;
3432 conf.coc_inode = inode;
3433 conf.coc_lock = lock;
3434 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
3435 conf.u.coc_layout.lb_len = lock->l_lvb_len;
3436 rc = ll_layout_conf(inode, &conf);
3438 /* refresh layout failed, need to wait */
3439 wait_layout = rc == -EBUSY;
3442 LDLM_LOCK_PUT(lock);
3443 ldlm_lock_decref(lockh, mode);
3445 /* wait for IO to complete if it's still being used. */
3447 CDEBUG(D_INODE, "%s: " DFID "(%p) wait for layout reconf\n",
3448 ll_get_fsname(inode->i_sb, NULL, 0),
3449 PFID(&lli->lli_fid), inode);
3451 memset(&conf, 0, sizeof(conf));
3452 conf.coc_opc = OBJECT_CONF_WAIT;
3453 conf.coc_inode = inode;
3454 rc = ll_layout_conf(inode, &conf);
3458 CDEBUG(D_INODE, "%s: file=" DFID " waiting layout return: %d.\n",
3459 ll_get_fsname(inode->i_sb, NULL, 0),
3460 PFID(&lli->lli_fid), rc);
3465 static int ll_layout_refresh_locked(struct inode *inode)
3467 struct ll_inode_info *lli = ll_i2info(inode);
3468 struct ll_sb_info *sbi = ll_i2sbi(inode);
3469 struct md_op_data *op_data;
3470 struct lookup_intent it;
3471 struct lustre_handle lockh;
3472 enum ldlm_mode mode;
3473 struct ldlm_enqueue_info einfo = {
3474 .ei_type = LDLM_IBITS,
3476 .ei_cb_bl = &ll_md_blocking_ast,
3477 .ei_cb_cp = &ldlm_completion_ast,
3482 /* mostly layout lock is caching on the local side, so try to match
3483 * it before grabbing layout lock mutex.
3485 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
3486 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
3487 if (mode != 0) { /* hit cached lock */
3488 rc = ll_layout_lock_set(&lockh, mode, inode);
3494 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3495 0, 0, LUSTRE_OPC_ANY, NULL);
3496 if (IS_ERR(op_data))
3497 return PTR_ERR(op_data);
3499 /* have to enqueue one */
3500 memset(&it, 0, sizeof(it));
3501 it.it_op = IT_LAYOUT;
3502 lockh.cookie = 0ULL;
3504 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file " DFID "(%p)",
3505 ll_get_fsname(inode->i_sb, NULL, 0),
3506 PFID(&lli->lli_fid), inode);
3508 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
3509 ptlrpc_req_finished(it.it_request);
3510 it.it_request = NULL;
3512 ll_finish_md_op_data(op_data);
3514 mode = it.it_lock_mode;
3515 it.it_lock_mode = 0;
3516 ll_intent_drop_lock(&it);
3519 /* set lock data in case this is a new lock */
3520 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3521 rc = ll_layout_lock_set(&lockh, mode, inode);
3530 * This function checks if there exists a LAYOUT lock on the client side,
3531 * or enqueues it if it doesn't have one in cache.
3533 * This function will not hold layout lock so it may be revoked any time after
3534 * this function returns. Any operations depend on layout should be redone
3537 * This function should be called before lov_io_init() to get an uptodate
3538 * layout version, the caller should save the version number and after IO
3539 * is finished, this function should be called again to verify that layout
3540 * is not changed during IO time.
3542 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3544 struct ll_inode_info *lli = ll_i2info(inode);
3545 struct ll_sb_info *sbi = ll_i2sbi(inode);
3548 *gen = ll_layout_version_get(lli);
3549 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
3553 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3554 LASSERT(S_ISREG(inode->i_mode));
3556 /* take layout lock mutex to enqueue layout lock exclusively. */
3557 mutex_lock(&lli->lli_layout_mutex);
3559 rc = ll_layout_refresh_locked(inode);
3563 *gen = ll_layout_version_get(lli);
3565 mutex_unlock(&lli->lli_layout_mutex);
3571 * This function send a restore request to the MDT
3573 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
3575 struct hsm_user_request *hur;
3578 len = sizeof(struct hsm_user_request) +
3579 sizeof(struct hsm_user_item);
3580 hur = kzalloc(len, GFP_NOFS);
3584 hur->hur_request.hr_action = HUA_RESTORE;
3585 hur->hur_request.hr_archive_id = 0;
3586 hur->hur_request.hr_flags = 0;
3587 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
3588 sizeof(hur->hur_user_item[0].hui_fid));
3589 hur->hur_user_item[0].hui_extent.offset = offset;
3590 hur->hur_user_item[0].hui_extent.length = length;
3591 hur->hur_request.hr_itemcount = 1;
3592 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,