4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
40 #include "../include/lustre_dlm.h"
41 #include "../include/lustre_lite.h"
42 #include <linux/pagemap.h>
43 #include <linux/file.h>
44 #include <linux/mount.h>
45 #include "llite_internal.h"
46 #include "../include/lustre/ll_fiemap.h"
48 #include "../include/cl_object.h"
51 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
53 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
56 static enum llioc_iter
57 ll_iocontrol_call(struct inode *inode, struct file *file,
58 unsigned int cmd, unsigned long arg, int *rcp);
60 static struct ll_file_data *ll_file_data_get(void)
62 struct ll_file_data *fd;
64 fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS);
67 fd->fd_write_failed = false;
71 static void ll_file_data_put(struct ll_file_data *fd)
74 kmem_cache_free(ll_file_data_slab, fd);
77 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
78 struct lustre_handle *fh)
80 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
81 op_data->op_attr.ia_mode = inode->i_mode;
82 op_data->op_attr.ia_atime = inode->i_atime;
83 op_data->op_attr.ia_mtime = inode->i_mtime;
84 op_data->op_attr.ia_ctime = inode->i_ctime;
85 op_data->op_attr.ia_size = i_size_read(inode);
86 op_data->op_attr_blocks = inode->i_blocks;
87 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
88 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
90 op_data->op_handle = *fh;
92 if (ll_i2info(inode)->lli_flags & LLIF_DATA_MODIFIED)
93 op_data->op_bias |= MDS_DATA_MODIFIED;
97 * Closes the IO epoch and packs all the attributes into @op_data for
100 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
101 struct obd_client_handle *och)
103 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
104 ATTR_MTIME | ATTR_MTIME_SET |
105 ATTR_CTIME | ATTR_CTIME_SET;
107 if (!(och->och_flags & FMODE_WRITE))
110 if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
111 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
113 ll_ioepoch_close(inode, op_data, &och, 0);
116 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
117 ll_prep_md_op_data(op_data, inode, NULL, NULL,
118 0, 0, LUSTRE_OPC_ANY, NULL);
121 static int ll_close_inode_openhandle(struct obd_export *md_exp,
123 struct obd_client_handle *och,
124 const __u64 *data_version)
126 struct obd_export *exp = ll_i2mdexp(inode);
127 struct md_op_data *op_data;
128 struct ptlrpc_request *req = NULL;
129 struct obd_device *obd = class_exp2obd(exp);
135 * XXX: in case of LMV, is this correct to access
138 CERROR("Invalid MDC connection handle %#llx\n",
139 ll_i2mdexp(inode)->exp_handle.h_cookie);
144 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
146 /* XXX We leak openhandle and request here. */
151 ll_prepare_close(inode, op_data, och);
153 /* Pass in data_version implies release. */
154 op_data->op_bias |= MDS_HSM_RELEASE;
155 op_data->op_data_version = *data_version;
156 op_data->op_lease_handle = och->och_lease_handle;
157 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
159 epoch_close = op_data->op_flags & MF_EPOCH_CLOSE;
160 rc = md_close(md_exp, op_data, och->och_mod, &req);
162 /* This close must have the epoch closed. */
163 LASSERT(epoch_close);
164 /* MDS has instructed us to obtain Size-on-MDS attribute from
165 * OSTs and send setattr to back to MDS.
167 rc = ll_som_update(inode, op_data);
169 CERROR("%s: inode "DFID" mdc Size-on-MDS update failed: rc = %d\n",
170 ll_i2mdexp(inode)->exp_obd->obd_name,
171 PFID(ll_inode2fid(inode)), rc);
175 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
176 ll_i2mdexp(inode)->exp_obd->obd_name,
177 PFID(ll_inode2fid(inode)), rc);
180 /* DATA_MODIFIED flag was successfully sent on close, cancel data
183 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
184 struct ll_inode_info *lli = ll_i2info(inode);
186 spin_lock(&lli->lli_lock);
187 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
188 spin_unlock(&lli->lli_lock);
192 rc = ll_objects_destroy(req, inode);
194 CERROR("inode %lu ll_objects destroy: rc = %d\n",
197 if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
198 struct mdt_body *body;
200 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
201 if (!(body->valid & OBD_MD_FLRELEASED))
205 ll_finish_md_op_data(op_data);
208 if (exp_connect_som(exp) && !epoch_close &&
209 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
210 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
212 md_clear_open_replay_data(md_exp, och);
213 /* Free @och if it is not waiting for DONE_WRITING. */
214 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
217 if (req) /* This is close request */
218 ptlrpc_req_finished(req);
222 int ll_md_real_close(struct inode *inode, fmode_t fmode)
224 struct ll_inode_info *lli = ll_i2info(inode);
225 struct obd_client_handle **och_p;
226 struct obd_client_handle *och;
230 if (fmode & FMODE_WRITE) {
231 och_p = &lli->lli_mds_write_och;
232 och_usecount = &lli->lli_open_fd_write_count;
233 } else if (fmode & FMODE_EXEC) {
234 och_p = &lli->lli_mds_exec_och;
235 och_usecount = &lli->lli_open_fd_exec_count;
237 LASSERT(fmode & FMODE_READ);
238 och_p = &lli->lli_mds_read_och;
239 och_usecount = &lli->lli_open_fd_read_count;
242 mutex_lock(&lli->lli_och_mutex);
243 if (*och_usecount > 0) {
244 /* There are still users of this handle, so skip
247 mutex_unlock(&lli->lli_och_mutex);
253 mutex_unlock(&lli->lli_och_mutex);
256 /* There might be a race and this handle may already
259 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
266 static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
269 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
270 struct ll_inode_info *lli = ll_i2info(inode);
272 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
273 struct lustre_handle lockh;
274 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN} };
277 /* clear group lock, if present */
278 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
279 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
281 if (fd->fd_lease_och) {
284 /* Usually the lease is not released when the
285 * application crashed, we need to release here.
287 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
288 CDEBUG(rc ? D_ERROR : D_INODE,
289 "Clean up lease " DFID " %d/%d\n",
290 PFID(&lli->lli_fid), rc, lease_broken);
292 fd->fd_lease_och = NULL;
296 rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL);
301 /* Let's see if we have good enough OPEN lock on the file and if
302 * we can skip talking to MDS
305 mutex_lock(&lli->lli_och_mutex);
306 if (fd->fd_omode & FMODE_WRITE) {
308 LASSERT(lli->lli_open_fd_write_count);
309 lli->lli_open_fd_write_count--;
310 } else if (fd->fd_omode & FMODE_EXEC) {
312 LASSERT(lli->lli_open_fd_exec_count);
313 lli->lli_open_fd_exec_count--;
316 LASSERT(lli->lli_open_fd_read_count);
317 lli->lli_open_fd_read_count--;
319 mutex_unlock(&lli->lli_och_mutex);
321 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
322 LDLM_IBITS, &policy, lockmode, &lockh))
323 rc = ll_md_real_close(inode, fd->fd_omode);
326 LUSTRE_FPRIVATE(file) = NULL;
327 ll_file_data_put(fd);
332 /* While this returns an error code, fput() the caller does not, so we need
333 * to make every effort to clean up all of our state here. Also, applications
334 * rarely check close errors and even if an error is returned they will not
335 * re-try the close call.
337 int ll_file_release(struct inode *inode, struct file *file)
339 struct ll_file_data *fd;
340 struct ll_sb_info *sbi = ll_i2sbi(inode);
341 struct ll_inode_info *lli = ll_i2info(inode);
344 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
345 PFID(ll_inode2fid(inode)), inode);
347 if (!is_root_inode(inode))
348 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
349 fd = LUSTRE_FPRIVATE(file);
352 /* The last ref on @file, maybe not be the owner pid of statahead.
353 * Different processes can open the same dir, "ll_opendir_key" means:
354 * it is me that should stop the statahead thread.
356 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
357 lli->lli_opendir_pid != 0)
358 ll_stop_statahead(inode, lli->lli_opendir_key);
360 if (is_root_inode(inode)) {
361 LUSTRE_FPRIVATE(file) = NULL;
362 ll_file_data_put(fd);
366 if (!S_ISDIR(inode->i_mode)) {
367 lov_read_and_clear_async_rc(lli->lli_clob);
368 lli->lli_async_rc = 0;
371 rc = ll_md_close(sbi->ll_md_exp, inode, file);
373 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
374 libcfs_debug_dumplog();
379 static int ll_intent_file_open(struct dentry *dentry, void *lmm,
380 int lmmsize, struct lookup_intent *itp)
382 struct inode *inode = d_inode(dentry);
383 struct ll_sb_info *sbi = ll_i2sbi(inode);
384 struct dentry *parent = dentry->d_parent;
385 const char *name = dentry->d_name.name;
386 const int len = dentry->d_name.len;
387 struct md_op_data *op_data;
388 struct ptlrpc_request *req;
389 __u32 opc = LUSTRE_OPC_ANY;
392 /* Usually we come here only for NFSD, and we want open lock. */
393 /* We can also get here if there was cached open handle in revalidate_it
394 * but it disappeared while we were getting from there to ll_file_open.
395 * But this means this file was closed and immediately opened which
396 * makes a good candidate for using OPEN lock
398 /* If lmmsize & lmm are not 0, we are just setting stripe info
399 * parameters. No need for the open lock
401 if (!lmm && lmmsize == 0) {
402 struct ll_dentry_data *ldd = ll_d2d(dentry);
404 * If we came via ll_iget_for_nfs, then we need to request
405 * struct ll_dentry_data *ldd = ll_d2d(file->f_dentry);
407 * NB: when ldd is NULL, it must have come via normal
408 * lookup path only, since ll_iget_for_nfs always calls
411 if (ldd && ldd->lld_nfs_dentry) {
412 ldd->lld_nfs_dentry = 0;
413 itp->it_flags |= MDS_OPEN_LOCK;
415 if (itp->it_flags & FMODE_WRITE)
416 opc = LUSTRE_OPC_CREATE;
419 op_data = ll_prep_md_op_data(NULL, d_inode(parent),
423 return PTR_ERR(op_data);
425 itp->it_flags |= MDS_OPEN_BY_FID;
426 rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
427 0 /*unused */, &req, ll_md_blocking_ast, 0);
428 ll_finish_md_op_data(op_data);
430 /* reason for keep own exit path - don`t flood log
431 * with messages with -ESTALE errors.
433 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
434 it_open_error(DISP_OPEN_OPEN, itp))
436 ll_release_openhandle(inode, itp);
440 if (it_disposition(itp, DISP_LOOKUP_NEG)) {
445 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
446 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
447 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
451 rc = ll_prep_inode(&inode, req, NULL, itp);
452 if (!rc && itp->it_lock_mode)
453 ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL);
456 ptlrpc_req_finished(req);
457 ll_intent_drop_lock(itp);
463 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
464 * not believe attributes if a few ioepoch holders exist. Attributes for
465 * previous ioepoch if new one is opened are also skipped by MDS.
467 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
469 if (ioepoch && lli->lli_ioepoch != ioepoch) {
470 lli->lli_ioepoch = ioepoch;
471 CDEBUG(D_INODE, "Epoch %llu opened on "DFID"\n",
472 ioepoch, PFID(&lli->lli_fid));
476 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
477 struct obd_client_handle *och)
479 struct mdt_body *body;
481 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
482 och->och_fh = body->handle;
483 och->och_fid = body->fid1;
484 och->och_lease_handle.cookie = it->it_lock_handle;
485 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
486 och->och_flags = it->it_flags;
488 return md_set_open_replay_data(md_exp, och, it);
491 static int ll_local_open(struct file *file, struct lookup_intent *it,
492 struct ll_file_data *fd, struct obd_client_handle *och)
494 struct inode *inode = file_inode(file);
495 struct ll_inode_info *lli = ll_i2info(inode);
497 LASSERT(!LUSTRE_FPRIVATE(file));
502 struct mdt_body *body;
505 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
509 body = req_capsule_server_get(&it->it_request->rq_pill,
511 ll_ioepoch_open(lli, body->ioepoch);
514 LUSTRE_FPRIVATE(file) = fd;
515 ll_readahead_init(inode, &fd->fd_ras);
516 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
518 /* ll_cl_context initialize */
519 rwlock_init(&fd->fd_lock);
520 INIT_LIST_HEAD(&fd->fd_lccs);
525 /* Open a file, and (for the very first open) create objects on the OSTs at
526 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
527 * creation or open until ll_lov_setstripe() ioctl is called.
529 * If we already have the stripe MD locally then we don't request it in
530 * md_open(), by passing a lmm_size = 0.
532 * It is up to the application to ensure no other processes open this file
533 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
534 * used. We might be able to avoid races of that sort by getting lli_open_sem
535 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
536 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
538 int ll_file_open(struct inode *inode, struct file *file)
540 struct ll_inode_info *lli = ll_i2info(inode);
541 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
542 .it_flags = file->f_flags };
543 struct obd_client_handle **och_p = NULL;
544 __u64 *och_usecount = NULL;
545 struct ll_file_data *fd;
546 int rc = 0, opendir_set = 0;
548 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
549 PFID(ll_inode2fid(inode)), inode, file->f_flags);
551 it = file->private_data; /* XXX: compat macro */
552 file->private_data = NULL; /* prevent ll_local_open assertion */
554 fd = ll_file_data_get();
561 if (S_ISDIR(inode->i_mode)) {
562 spin_lock(&lli->lli_sa_lock);
563 if (!lli->lli_opendir_key && !lli->lli_sai &&
564 lli->lli_opendir_pid == 0) {
565 lli->lli_opendir_key = fd;
566 lli->lli_opendir_pid = current_pid();
569 spin_unlock(&lli->lli_sa_lock);
572 if (is_root_inode(inode)) {
573 LUSTRE_FPRIVATE(file) = fd;
577 if (!it || !it->it_disposition) {
578 /* Convert f_flags into access mode. We cannot use file->f_mode,
579 * because everything but O_ACCMODE mask was stripped from
582 if ((oit.it_flags + 1) & O_ACCMODE)
584 if (file->f_flags & O_TRUNC)
585 oit.it_flags |= FMODE_WRITE;
587 /* kernel only call f_op->open in dentry_open. filp_open calls
588 * dentry_open after call to open_namei that checks permissions.
589 * Only nfsd_open call dentry_open directly without checking
590 * permissions and because of that this code below is safe.
592 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
593 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
595 /* We do not want O_EXCL here, presumably we opened the file
596 * already? XXX - NFS implications?
598 oit.it_flags &= ~O_EXCL;
600 /* bug20584, if "it_flags" contains O_CREAT, the file will be
601 * created if necessary, then "IT_CREAT" should be set to keep
604 if (oit.it_flags & O_CREAT)
605 oit.it_op |= IT_CREAT;
611 /* Let's see if we have file open on MDS already. */
612 if (it->it_flags & FMODE_WRITE) {
613 och_p = &lli->lli_mds_write_och;
614 och_usecount = &lli->lli_open_fd_write_count;
615 } else if (it->it_flags & FMODE_EXEC) {
616 och_p = &lli->lli_mds_exec_och;
617 och_usecount = &lli->lli_open_fd_exec_count;
619 och_p = &lli->lli_mds_read_och;
620 och_usecount = &lli->lli_open_fd_read_count;
623 mutex_lock(&lli->lli_och_mutex);
624 if (*och_p) { /* Open handle is present */
625 if (it_disposition(it, DISP_OPEN_OPEN)) {
626 /* Well, there's extra open request that we do not need,
627 * let's close it somehow. This will decref request.
629 rc = it_open_error(DISP_OPEN_OPEN, it);
631 mutex_unlock(&lli->lli_och_mutex);
635 ll_release_openhandle(inode, it);
639 rc = ll_local_open(file, it, fd, NULL);
642 mutex_unlock(&lli->lli_och_mutex);
646 LASSERT(*och_usecount == 0);
647 if (!it->it_disposition) {
648 /* We cannot just request lock handle now, new ELC code
649 * means that one of other OPEN locks for this file
650 * could be cancelled, and since blocking ast handler
651 * would attempt to grab och_mutex as well, that would
652 * result in a deadlock
654 mutex_unlock(&lli->lli_och_mutex);
655 it->it_create_mode |= M_CHECK_STALE;
656 rc = ll_intent_file_open(file->f_path.dentry, NULL, 0, it);
657 it->it_create_mode &= ~M_CHECK_STALE;
663 *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS);
671 /* md_intent_lock() didn't get a request ref if there was an
672 * open error, so don't do cleanup on the request here
675 /* XXX (green): Should not we bail out on any error here, not
678 rc = it_open_error(DISP_OPEN_OPEN, it);
682 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
683 "inode %p: disposition %x, status %d\n", inode,
684 it_disposition(it, ~0), it->it_status);
686 rc = ll_local_open(file, it, fd, *och_p);
690 mutex_unlock(&lli->lli_och_mutex);
693 /* Must do this outside lli_och_mutex lock to prevent deadlock where
694 * different kind of OPEN lock for this same inode gets cancelled
697 if (!S_ISREG(inode->i_mode))
700 if (!lli->lli_has_smd &&
701 (cl_is_lov_delay_create(file->f_flags) ||
702 (file->f_mode & FMODE_WRITE) == 0)) {
703 CDEBUG(D_INODE, "object creation was delayed\n");
706 cl_lov_delay_create_clear(&file->f_flags);
711 if (och_p && *och_p) {
716 mutex_unlock(&lli->lli_och_mutex);
719 if (opendir_set != 0)
720 ll_stop_statahead(inode, lli->lli_opendir_key);
721 ll_file_data_put(fd);
723 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
726 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
727 ptlrpc_req_finished(it->it_request);
728 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
734 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
735 struct ldlm_lock_desc *desc,
736 void *data, int flag)
739 struct lustre_handle lockh;
742 case LDLM_CB_BLOCKING:
743 ldlm_lock2handle(lock, &lockh);
744 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
746 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
750 case LDLM_CB_CANCELING:
758 * Acquire a lease and open the file.
760 static struct obd_client_handle *
761 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
764 struct lookup_intent it = { .it_op = IT_OPEN };
765 struct ll_sb_info *sbi = ll_i2sbi(inode);
766 struct md_op_data *op_data;
767 struct ptlrpc_request *req;
768 struct lustre_handle old_handle = { 0 };
769 struct obd_client_handle *och = NULL;
773 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
774 return ERR_PTR(-EINVAL);
777 struct ll_inode_info *lli = ll_i2info(inode);
778 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
779 struct obd_client_handle **och_p;
782 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
783 return ERR_PTR(-EPERM);
785 /* Get the openhandle of the file */
787 mutex_lock(&lli->lli_och_mutex);
788 if (fd->fd_lease_och) {
789 mutex_unlock(&lli->lli_och_mutex);
794 if (file->f_mode & FMODE_WRITE) {
795 LASSERT(lli->lli_mds_write_och);
796 och_p = &lli->lli_mds_write_och;
797 och_usecount = &lli->lli_open_fd_write_count;
799 LASSERT(lli->lli_mds_read_och);
800 och_p = &lli->lli_mds_read_och;
801 och_usecount = &lli->lli_open_fd_read_count;
803 if (*och_usecount == 1) {
810 mutex_unlock(&lli->lli_och_mutex);
811 if (rc < 0) /* more than 1 opener */
815 old_handle = fd->fd_och->och_fh;
818 och = kzalloc(sizeof(*och), GFP_NOFS);
820 return ERR_PTR(-ENOMEM);
822 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
823 LUSTRE_OPC_ANY, NULL);
824 if (IS_ERR(op_data)) {
825 rc = PTR_ERR(op_data);
829 /* To tell the MDT this openhandle is from the same owner */
830 op_data->op_handle = old_handle;
832 it.it_flags = fmode | open_flags;
833 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
834 rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req,
835 ll_md_blocking_lease_ast,
836 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
837 * it can be cancelled which may mislead applications that the lease is
839 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
840 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
841 * doesn't deal with openhandle, so normal openhandle will be leaked.
843 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
844 ll_finish_md_op_data(op_data);
845 ptlrpc_req_finished(req);
849 if (it_disposition(&it, DISP_LOOKUP_NEG)) {
854 rc = it_open_error(DISP_OPEN_OPEN, &it);
858 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
859 ll_och_fill(sbi->ll_md_exp, &it, och);
861 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ {
866 /* already get lease, handle lease lock */
867 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
868 if (it.it_lock_mode == 0 ||
869 it.it_lock_bits != MDS_INODELOCK_OPEN) {
870 /* open lock must return for lease */
871 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
872 PFID(ll_inode2fid(inode)), it.it_lock_mode,
878 ll_intent_release(&it);
882 /* Cancel open lock */
883 if (it.it_lock_mode != 0) {
884 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
887 och->och_lease_handle.cookie = 0ULL;
889 rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL);
891 CERROR("%s: error closing file "DFID": %d\n",
892 ll_get_fsname(inode->i_sb, NULL, 0),
893 PFID(&ll_i2info(inode)->lli_fid), rc2);
894 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
896 ll_intent_release(&it);
903 * Release lease and close the file.
904 * It will check if the lease has ever broken.
906 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
909 struct ldlm_lock *lock;
910 bool cancelled = true;
913 lock = ldlm_handle2lock(&och->och_lease_handle);
915 lock_res_and_lock(lock);
916 cancelled = ldlm_is_cancel(lock);
917 unlock_res_and_lock(lock);
921 CDEBUG(D_INODE, "lease for " DFID " broken? %d\n",
922 PFID(&ll_i2info(inode)->lli_fid), cancelled);
925 ldlm_cli_cancel(&och->och_lease_handle, 0);
927 *lease_broken = cancelled;
929 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
934 /* Fills the obdo with the attributes for the lsm */
935 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
936 struct obdo *obdo, __u64 ioepoch, int dv_flags)
938 struct ptlrpc_request_set *set;
939 struct obd_info oinfo = { };
946 oinfo.oi_oa->o_oi = lsm->lsm_oi;
947 oinfo.oi_oa->o_mode = S_IFREG;
948 oinfo.oi_oa->o_ioepoch = ioepoch;
949 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
950 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
951 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
952 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
953 OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
954 OBD_MD_FLDATAVERSION;
955 if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) {
956 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
957 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
958 if (dv_flags & LL_DV_WR_FLUSH)
959 oinfo.oi_oa->o_flags |= OBD_FL_FLUSH;
962 set = ptlrpc_prep_set();
964 CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM);
967 rc = obd_getattr_async(exp, &oinfo, set);
969 rc = ptlrpc_set_wait(set);
970 ptlrpc_set_destroy(set);
973 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
974 OBD_MD_FLATIME | OBD_MD_FLMTIME |
975 OBD_MD_FLCTIME | OBD_MD_FLSIZE |
976 OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS);
977 if (dv_flags & LL_DV_WR_FLUSH &&
978 !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
979 oinfo.oi_oa->o_flags & OBD_FL_FLUSH))
986 * Performs the getattr on the inode and updates its fields.
987 * If @sync != 0, perform the getattr under the server-side lock.
989 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
990 __u64 ioepoch, int sync)
992 struct lov_stripe_md *lsm;
995 lsm = ccc_inode_lsm_get(inode);
996 rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode),
997 obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0);
999 struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi;
1001 obdo_refresh_inode(inode, obdo, obdo->o_valid);
1002 CDEBUG(D_INODE, "objid " DOSTID " size %llu, blocks %llu, blksize %lu\n",
1003 POSTID(oi), i_size_read(inode),
1004 (unsigned long long)inode->i_blocks,
1005 1UL << inode->i_blkbits);
1007 ccc_inode_lsm_put(inode, lsm);
1011 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1013 struct ll_inode_info *lli = ll_i2info(inode);
1014 struct cl_object *obj = lli->lli_clob;
1015 struct cl_attr *attr = vvp_env_thread_attr(env);
1021 ll_inode_size_lock(inode);
1023 /* merge timestamps the most recently obtained from mds with
1024 * timestamps obtained from osts
1026 LTIME_S(inode->i_atime) = lli->lli_atime;
1027 LTIME_S(inode->i_mtime) = lli->lli_mtime;
1028 LTIME_S(inode->i_ctime) = lli->lli_ctime;
1030 mtime = LTIME_S(inode->i_mtime);
1031 atime = LTIME_S(inode->i_atime);
1032 ctime = LTIME_S(inode->i_ctime);
1034 cl_object_attr_lock(obj);
1035 rc = cl_object_attr_get(env, obj, attr);
1036 cl_object_attr_unlock(obj);
1039 goto out_size_unlock;
1041 if (atime < attr->cat_atime)
1042 atime = attr->cat_atime;
1044 if (ctime < attr->cat_ctime)
1045 ctime = attr->cat_ctime;
1047 if (mtime < attr->cat_mtime)
1048 mtime = attr->cat_mtime;
1050 CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n",
1051 PFID(&lli->lli_fid), attr->cat_size);
1053 i_size_write(inode, attr->cat_size);
1055 inode->i_blocks = attr->cat_blocks;
1057 LTIME_S(inode->i_mtime) = mtime;
1058 LTIME_S(inode->i_atime) = atime;
1059 LTIME_S(inode->i_ctime) = ctime;
1062 ll_inode_size_unlock(inode);
1067 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
1070 struct obdo obdo = { 0 };
1073 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, &obdo, 0, 0);
1075 st->st_size = obdo.o_size;
1076 st->st_blocks = obdo.o_blocks;
1077 st->st_mtime = obdo.o_mtime;
1078 st->st_atime = obdo.o_atime;
1079 st->st_ctime = obdo.o_ctime;
1084 static bool file_is_noatime(const struct file *file)
1086 const struct vfsmount *mnt = file->f_path.mnt;
1087 const struct inode *inode = file_inode(file);
1089 /* Adapted from file_accessed() and touch_atime().*/
1090 if (file->f_flags & O_NOATIME)
1093 if (inode->i_flags & S_NOATIME)
1096 if (IS_NOATIME(inode))
1099 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1102 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1105 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1111 void ll_io_init(struct cl_io *io, const struct file *file, int write)
1113 struct inode *inode = file_inode(file);
1115 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1117 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1118 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1119 file->f_flags & O_DIRECT ||
1122 io->ci_obj = ll_i2info(inode)->lli_clob;
1123 io->ci_lockreq = CILR_MAYBE;
1124 if (ll_file_nolock(file)) {
1125 io->ci_lockreq = CILR_NEVER;
1126 io->ci_no_srvlock = 1;
1127 } else if (file->f_flags & O_APPEND) {
1128 io->ci_lockreq = CILR_MANDATORY;
1131 io->ci_noatime = file_is_noatime(file);
1135 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1136 struct file *file, enum cl_io_type iot,
1137 loff_t *ppos, size_t count)
1139 struct ll_inode_info *lli = ll_i2info(file_inode(file));
1140 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1144 CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: %llu, count: %zd\n",
1145 file->f_path.dentry->d_name.name, iot, *ppos, count);
1148 io = vvp_env_thread_io(env);
1149 ll_io_init(io, file, iot == CIT_WRITE);
1151 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1152 struct vvp_io *vio = vvp_env_io(env);
1153 int write_mutex_locked = 0;
1155 vio->vui_fd = LUSTRE_FPRIVATE(file);
1156 vio->vui_iter = args->u.normal.via_iter;
1157 vio->vui_iocb = args->u.normal.via_iocb;
1158 if ((iot == CIT_WRITE) &&
1159 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1160 if (mutex_lock_interruptible(&lli->lli_write_mutex)) {
1161 result = -ERESTARTSYS;
1164 write_mutex_locked = 1;
1166 down_read(&lli->lli_trunc_sem);
1167 ll_cl_add(file, env, io);
1168 result = cl_io_loop(env, io);
1169 ll_cl_remove(file, env);
1170 up_read(&lli->lli_trunc_sem);
1171 if (write_mutex_locked)
1172 mutex_unlock(&lli->lli_write_mutex);
1174 /* cl_io_rw_init() handled IO */
1175 result = io->ci_result;
1178 if (io->ci_nob > 0) {
1179 result = io->ci_nob;
1180 *ppos = io->u.ci_wr.wr.crw_pos;
1184 cl_io_fini(env, io);
1185 /* If any bit been read/written (result != 0), we just return
1186 * short read/write instead of restart io.
1188 if ((result == 0 || result == -ENODATA) && io->ci_need_restart) {
1189 CDEBUG(D_VFSTRACE, "Restart %s on %pD from %lld, count:%zd\n",
1190 iot == CIT_READ ? "read" : "write",
1191 file, *ppos, count);
1192 LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob);
1196 if (iot == CIT_READ) {
1198 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1199 LPROC_LL_READ_BYTES, result);
1200 } else if (iot == CIT_WRITE) {
1202 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1203 LPROC_LL_WRITE_BYTES, result);
1204 fd->fd_write_failed = false;
1205 } else if (result != -ERESTARTSYS) {
1206 fd->fd_write_failed = true;
1209 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1214 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1217 struct vvp_io_args *args;
1221 env = cl_env_get(&refcheck);
1223 return PTR_ERR(env);
1225 args = ll_env_args(env);
1226 args->u.normal.via_iter = to;
1227 args->u.normal.via_iocb = iocb;
1229 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1230 &iocb->ki_pos, iov_iter_count(to));
1231 cl_env_put(env, &refcheck);
1236 * Write to a file (through the page cache).
1238 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1241 struct vvp_io_args *args;
1245 env = cl_env_get(&refcheck);
1247 return PTR_ERR(env);
1249 args = ll_env_args(env);
1250 args->u.normal.via_iter = from;
1251 args->u.normal.via_iocb = iocb;
1253 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1254 &iocb->ki_pos, iov_iter_count(from));
1255 cl_env_put(env, &refcheck);
1259 static int ll_lov_recreate(struct inode *inode, struct ost_id *oi, u32 ost_idx)
1261 struct obd_export *exp = ll_i2dtexp(inode);
1262 struct obd_trans_info oti = { 0 };
1263 struct obdo *oa = NULL;
1266 struct lov_stripe_md *lsm = NULL, *lsm2;
1268 oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
1272 lsm = ccc_inode_lsm_get(inode);
1273 if (!lsm_has_objects(lsm)) {
1278 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1279 (lsm->lsm_stripe_count));
1281 lsm2 = libcfs_kvzalloc(lsm_size, GFP_NOFS);
1288 oa->o_nlink = ost_idx;
1289 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1290 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1291 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1292 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1293 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1294 memcpy(lsm2, lsm, lsm_size);
1295 ll_inode_size_lock(inode);
1296 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1297 ll_inode_size_unlock(inode);
1302 ccc_inode_lsm_put(inode, lsm);
1303 kmem_cache_free(obdo_cachep, oa);
1307 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1309 struct ll_recreate_obj ucreat;
1312 if (!capable(CFS_CAP_SYS_ADMIN))
1315 if (copy_from_user(&ucreat, (struct ll_recreate_obj __user *)arg,
1319 ostid_set_seq_mdt0(&oi);
1320 ostid_set_id(&oi, ucreat.lrc_id);
1321 return ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx);
1324 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1330 if (!capable(CFS_CAP_SYS_ADMIN))
1333 if (copy_from_user(&fid, (struct lu_fid __user *)arg, sizeof(fid)))
1336 fid_to_ostid(&fid, &oi);
1337 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1338 return ll_lov_recreate(inode, &oi, ost_idx);
1341 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1342 __u64 flags, struct lov_user_md *lum,
1345 struct lov_stripe_md *lsm = NULL;
1346 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1349 lsm = ccc_inode_lsm_get(inode);
1351 ccc_inode_lsm_put(inode, lsm);
1352 CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n",
1353 PFID(ll_inode2fid(inode)));
1358 ll_inode_size_lock(inode);
1359 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
1366 ll_release_openhandle(inode, &oit);
1369 ll_inode_size_unlock(inode);
1370 ll_intent_release(&oit);
1371 ccc_inode_lsm_put(inode, lsm);
1375 ptlrpc_req_finished((struct ptlrpc_request *)oit.it_request);
1379 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1380 struct lov_mds_md **lmmp, int *lmm_size,
1381 struct ptlrpc_request **request)
1383 struct ll_sb_info *sbi = ll_i2sbi(inode);
1384 struct mdt_body *body;
1385 struct lov_mds_md *lmm = NULL;
1386 struct ptlrpc_request *req = NULL;
1387 struct md_op_data *op_data;
1390 rc = ll_get_default_mdsize(sbi, &lmmsize);
1394 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1395 strlen(filename), lmmsize,
1396 LUSTRE_OPC_ANY, NULL);
1397 if (IS_ERR(op_data))
1398 return PTR_ERR(op_data);
1400 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1401 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1402 ll_finish_md_op_data(op_data);
1404 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
1409 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1411 lmmsize = body->eadatasize;
1413 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1419 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1421 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1422 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1428 * This is coming from the MDS, so is probably in
1429 * little endian. We convert it to host endian before
1430 * passing it to userspace.
1432 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
1435 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1436 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1439 /* if function called for directory - we should
1440 * avoid swab not existent lsm objects
1442 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1443 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1444 if (S_ISREG(body->mode))
1445 lustre_swab_lov_user_md_objects(
1446 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1448 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1449 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1450 if (S_ISREG(body->mode))
1451 lustre_swab_lov_user_md_objects(
1452 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1459 *lmm_size = lmmsize;
1464 static int ll_lov_setea(struct inode *inode, struct file *file,
1467 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1468 struct lov_user_md *lump;
1469 int lum_size = sizeof(struct lov_user_md) +
1470 sizeof(struct lov_user_ost_data);
1473 if (!capable(CFS_CAP_SYS_ADMIN))
1476 lump = libcfs_kvzalloc(lum_size, GFP_NOFS);
1480 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) {
1485 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump,
1487 cl_lov_delay_create_clear(&file->f_flags);
1493 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1496 struct lov_user_md_v3 lumv3;
1497 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1498 struct lov_user_md_v1 __user *lumv1p = (void __user *)arg;
1499 struct lov_user_md_v3 __user *lumv3p = (void __user *)arg;
1501 __u64 flags = FMODE_WRITE;
1503 /* first try with v1 which is smaller than v3 */
1504 lum_size = sizeof(struct lov_user_md_v1);
1505 if (copy_from_user(lumv1, lumv1p, lum_size))
1508 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1509 lum_size = sizeof(struct lov_user_md_v3);
1510 if (copy_from_user(&lumv3, lumv3p, lum_size))
1514 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lumv1,
1516 cl_lov_delay_create_clear(&file->f_flags);
1518 struct lov_stripe_md *lsm;
1521 put_user(0, &lumv1p->lmm_stripe_count);
1523 ll_layout_refresh(inode, &gen);
1524 lsm = ccc_inode_lsm_get(inode);
1525 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1526 0, lsm, (void __user *)arg);
1527 ccc_inode_lsm_put(inode, lsm);
1532 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1534 struct lov_stripe_md *lsm;
1537 lsm = ccc_inode_lsm_get(inode);
1539 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1540 lsm, (void __user *)arg);
1541 ccc_inode_lsm_put(inode, lsm);
1546 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1548 struct ll_inode_info *lli = ll_i2info(inode);
1549 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1550 struct ll_grouplock grouplock;
1554 CWARN("group id for group lock must not be 0\n");
1558 if (ll_file_nolock(file))
1561 spin_lock(&lli->lli_lock);
1562 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1563 CWARN("group lock already existed with gid %lu\n",
1564 fd->fd_grouplock.lg_gid);
1565 spin_unlock(&lli->lli_lock);
1568 LASSERT(!fd->fd_grouplock.lg_lock);
1569 spin_unlock(&lli->lli_lock);
1571 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1572 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1576 spin_lock(&lli->lli_lock);
1577 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1578 spin_unlock(&lli->lli_lock);
1579 CERROR("another thread just won the race\n");
1580 cl_put_grouplock(&grouplock);
1584 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1585 fd->fd_grouplock = grouplock;
1586 spin_unlock(&lli->lli_lock);
1588 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1592 static int ll_put_grouplock(struct inode *inode, struct file *file,
1595 struct ll_inode_info *lli = ll_i2info(inode);
1596 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1597 struct ll_grouplock grouplock;
1599 spin_lock(&lli->lli_lock);
1600 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1601 spin_unlock(&lli->lli_lock);
1602 CWARN("no group lock held\n");
1605 LASSERT(fd->fd_grouplock.lg_lock);
1607 if (fd->fd_grouplock.lg_gid != arg) {
1608 CWARN("group lock %lu doesn't match current id %lu\n",
1609 arg, fd->fd_grouplock.lg_gid);
1610 spin_unlock(&lli->lli_lock);
1614 grouplock = fd->fd_grouplock;
1615 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1616 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1617 spin_unlock(&lli->lli_lock);
1619 cl_put_grouplock(&grouplock);
1620 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1625 * Close inode open handle
1627 * \param inode [in] inode in question
1628 * \param it [in,out] intent which contains open info and result
1631 * \retval <0 failure
1633 int ll_release_openhandle(struct inode *inode, struct lookup_intent *it)
1635 struct obd_client_handle *och;
1640 /* Root ? Do nothing. */
1641 if (is_root_inode(inode))
1644 /* No open handle to close? Move away */
1645 if (!it_disposition(it, DISP_OPEN_OPEN))
1648 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1650 och = kzalloc(sizeof(*och), GFP_NOFS);
1656 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1658 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1661 /* this one is in place of ll_file_open */
1662 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1663 ptlrpc_req_finished(it->it_request);
1664 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1670 * Get size for inode for which FIEMAP mapping is requested.
1671 * Make the FIEMAP get_info call and returns the result.
1673 static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1676 struct obd_export *exp = ll_i2dtexp(inode);
1677 struct lov_stripe_md *lsm = NULL;
1678 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1679 __u32 vallen = num_bytes;
1682 /* Checks for fiemap flags */
1683 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1684 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1688 /* Check for FIEMAP_FLAG_SYNC */
1689 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1690 rc = filemap_fdatawrite(inode->i_mapping);
1695 lsm = ccc_inode_lsm_get(inode);
1699 /* If the stripe_count > 1 and the application does not understand
1700 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1702 if (lsm->lsm_stripe_count > 1 &&
1703 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1708 fm_key.oa.o_oi = lsm->lsm_oi;
1709 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1711 if (i_size_read(inode) == 0) {
1712 rc = ll_glimpse_size(inode);
1717 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1718 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1719 /* If filesize is 0, then there would be no objects for mapping */
1720 if (fm_key.oa.o_size == 0) {
1721 fiemap->fm_mapped_extents = 0;
1726 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1728 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1731 CERROR("obd_get_info failed: rc = %d\n", rc);
1734 ccc_inode_lsm_put(inode, lsm);
1738 int ll_fid2path(struct inode *inode, void __user *arg)
1740 struct obd_export *exp = ll_i2mdexp(inode);
1741 const struct getinfo_fid2path __user *gfin = arg;
1742 struct getinfo_fid2path *gfout;
1747 if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
1748 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1751 /* Only need to get the buflen */
1752 if (get_user(pathlen, &gfin->gf_pathlen))
1755 if (pathlen > PATH_MAX)
1758 outsize = sizeof(*gfout) + pathlen;
1760 gfout = kzalloc(outsize, GFP_NOFS);
1764 if (copy_from_user(gfout, arg, sizeof(*gfout))) {
1769 /* Call mdc_iocontrol */
1770 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1774 if (copy_to_user(arg, gfout, outsize))
1782 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1784 struct ll_user_fiemap *fiemap_s;
1785 size_t num_bytes, ret_bytes;
1786 unsigned int extent_count;
1789 /* Get the extent count so we can calculate the size of
1790 * required fiemap buffer
1792 if (get_user(extent_count,
1793 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1797 (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent))
1799 num_bytes = sizeof(*fiemap_s) + (extent_count *
1800 sizeof(struct ll_fiemap_extent));
1802 fiemap_s = libcfs_kvzalloc(num_bytes, GFP_NOFS);
1806 /* get the fiemap value */
1807 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1808 sizeof(*fiemap_s))) {
1813 /* If fm_extent_count is non-zero, read the first extent since
1814 * it is used to calculate end_offset and device from previous
1818 if (copy_from_user(&fiemap_s->fm_extents[0],
1819 (char __user *)arg + sizeof(*fiemap_s),
1820 sizeof(struct ll_fiemap_extent))) {
1826 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1830 ret_bytes = sizeof(struct ll_user_fiemap);
1832 if (extent_count != 0)
1833 ret_bytes += (fiemap_s->fm_mapped_extents *
1834 sizeof(struct ll_fiemap_extent));
1836 if (copy_to_user((void __user *)arg, fiemap_s, ret_bytes))
1845 * Read the data_version for inode.
1847 * This value is computed using stripe object version on OST.
1848 * Version is computed using server side locking.
1850 * @param sync if do sync on the OST side;
1852 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1853 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1855 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1857 struct lov_stripe_md *lsm = NULL;
1858 struct ll_sb_info *sbi = ll_i2sbi(inode);
1859 struct obdo *obdo = NULL;
1862 /* If no stripe, we consider version is 0. */
1863 lsm = ccc_inode_lsm_get(inode);
1864 if (!lsm_has_objects(lsm)) {
1866 CDEBUG(D_INODE, "No object for inode\n");
1871 obdo = kzalloc(sizeof(*obdo), GFP_NOFS);
1877 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, 0, flags);
1879 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
1882 *data_version = obdo->o_data_version;
1887 ccc_inode_lsm_put(inode, lsm);
1892 * Trigger a HSM release request for the provided inode.
1894 int ll_hsm_release(struct inode *inode)
1896 struct cl_env_nest nest;
1898 struct obd_client_handle *och = NULL;
1899 __u64 data_version = 0;
1902 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
1903 ll_get_fsname(inode->i_sb, NULL, 0),
1904 PFID(&ll_i2info(inode)->lli_fid));
1906 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
1912 /* Grab latest data_version and [am]time values */
1913 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
1917 env = cl_env_nested_get(&nest);
1923 ll_merge_attr(env, inode);
1924 cl_env_nested_put(&nest, env);
1926 /* Release the file.
1927 * NB: lease lock handle is released in mdc_hsm_release_pack() because
1928 * we still need it to pack l_remote_handle to MDT.
1930 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
1935 if (och && !IS_ERR(och)) /* close the file */
1936 ll_lease_close(och, inode, NULL);
1941 struct ll_swap_stack {
1942 struct iattr ia1, ia2;
1944 struct inode *inode1, *inode2;
1945 bool check_dv1, check_dv2;
1948 static int ll_swap_layouts(struct file *file1, struct file *file2,
1949 struct lustre_swap_layouts *lsl)
1951 struct mdc_swap_layouts msl;
1952 struct md_op_data *op_data;
1955 struct ll_swap_stack *llss = NULL;
1958 llss = kzalloc(sizeof(*llss), GFP_NOFS);
1962 llss->inode1 = file_inode(file1);
1963 llss->inode2 = file_inode(file2);
1965 if (!S_ISREG(llss->inode2->i_mode)) {
1970 if (inode_permission(llss->inode1, MAY_WRITE) ||
1971 inode_permission(llss->inode2, MAY_WRITE)) {
1976 if (llss->inode2->i_sb != llss->inode1->i_sb) {
1981 /* we use 2 bool because it is easier to swap than 2 bits */
1982 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
1983 llss->check_dv1 = true;
1985 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
1986 llss->check_dv2 = true;
1988 /* we cannot use lsl->sl_dvX directly because we may swap them */
1989 llss->dv1 = lsl->sl_dv1;
1990 llss->dv2 = lsl->sl_dv2;
1992 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
1993 if (rc == 0) /* same file, done! */ {
1998 if (rc < 0) { /* sequentialize it */
1999 swap(llss->inode1, llss->inode2);
2001 swap(llss->dv1, llss->dv2);
2002 swap(llss->check_dv1, llss->check_dv2);
2006 if (gid != 0) { /* application asks to flush dirty cache */
2007 rc = ll_get_grouplock(llss->inode1, file1, gid);
2011 rc = ll_get_grouplock(llss->inode2, file2, gid);
2013 ll_put_grouplock(llss->inode1, file1, gid);
2018 /* to be able to restore mtime and atime after swap
2019 * we need to first save them
2022 (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
2023 llss->ia1.ia_mtime = llss->inode1->i_mtime;
2024 llss->ia1.ia_atime = llss->inode1->i_atime;
2025 llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
2026 llss->ia2.ia_mtime = llss->inode2->i_mtime;
2027 llss->ia2.ia_atime = llss->inode2->i_atime;
2028 llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
2031 /* ultimate check, before swapping the layouts we check if
2032 * dataversion has changed (if requested)
2034 if (llss->check_dv1) {
2035 rc = ll_data_version(llss->inode1, &dv, 0);
2038 if (dv != llss->dv1) {
2044 if (llss->check_dv2) {
2045 rc = ll_data_version(llss->inode2, &dv, 0);
2048 if (dv != llss->dv2) {
2054 /* struct md_op_data is used to send the swap args to the mdt
2055 * only flags is missing, so we use struct mdc_swap_layouts
2056 * through the md_op_data->op_data
2058 /* flags from user space have to be converted before they are send to
2059 * server, no flag is sent today, they are only used on the client
2063 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2064 0, LUSTRE_OPC_ANY, &msl);
2065 if (IS_ERR(op_data)) {
2066 rc = PTR_ERR(op_data);
2070 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2071 sizeof(*op_data), op_data, NULL);
2072 ll_finish_md_op_data(op_data);
2076 ll_put_grouplock(llss->inode2, file2, gid);
2077 ll_put_grouplock(llss->inode1, file1, gid);
2080 /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
2084 /* clear useless flags */
2085 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
2086 llss->ia1.ia_valid &= ~ATTR_MTIME;
2087 llss->ia2.ia_valid &= ~ATTR_MTIME;
2090 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
2091 llss->ia1.ia_valid &= ~ATTR_ATIME;
2092 llss->ia2.ia_valid &= ~ATTR_ATIME;
2095 /* update time if requested */
2097 if (llss->ia2.ia_valid != 0) {
2098 inode_lock(llss->inode1);
2099 rc = ll_setattr(file1->f_path.dentry, &llss->ia2);
2100 inode_unlock(llss->inode1);
2103 if (llss->ia1.ia_valid != 0) {
2106 inode_lock(llss->inode2);
2107 rc1 = ll_setattr(file2->f_path.dentry, &llss->ia1);
2108 inode_unlock(llss->inode2);
2119 static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2121 struct md_op_data *op_data;
2124 /* Detect out-of range masks */
2125 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2128 /* Non-root users are forbidden to set or clear flags which are
2129 * NOT defined in HSM_USER_MASK.
2131 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2132 !capable(CFS_CAP_SYS_ADMIN))
2135 /* Detect out-of range archive id */
2136 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2137 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
2140 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2141 LUSTRE_OPC_ANY, hss);
2142 if (IS_ERR(op_data))
2143 return PTR_ERR(op_data);
2145 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2146 sizeof(*op_data), op_data, NULL);
2148 ll_finish_md_op_data(op_data);
2153 static int ll_hsm_import(struct inode *inode, struct file *file,
2154 struct hsm_user_import *hui)
2156 struct hsm_state_set *hss = NULL;
2157 struct iattr *attr = NULL;
2160 if (!S_ISREG(inode->i_mode))
2164 hss = kzalloc(sizeof(*hss), GFP_NOFS);
2168 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2169 hss->hss_archive_id = hui->hui_archive_id;
2170 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2171 rc = ll_hsm_state_set(inode, hss);
2175 attr = kzalloc(sizeof(*attr), GFP_NOFS);
2181 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2182 attr->ia_mode |= S_IFREG;
2183 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2184 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2185 attr->ia_size = hui->hui_size;
2186 attr->ia_mtime.tv_sec = hui->hui_mtime;
2187 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2188 attr->ia_atime.tv_sec = hui->hui_atime;
2189 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2191 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2192 ATTR_UID | ATTR_GID |
2193 ATTR_MTIME | ATTR_MTIME_SET |
2194 ATTR_ATIME | ATTR_ATIME_SET;
2198 rc = ll_setattr_raw(file->f_path.dentry, attr, true);
2202 inode_unlock(inode);
2211 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2213 struct inode *inode = file_inode(file);
2214 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2217 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),cmd=%x\n",
2218 PFID(ll_inode2fid(inode)), inode, cmd);
2219 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2221 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2222 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2226 case LL_IOC_GETFLAGS:
2227 /* Get the current value of the file flags */
2228 return put_user(fd->fd_flags, (int __user *)arg);
2229 case LL_IOC_SETFLAGS:
2230 case LL_IOC_CLRFLAGS:
2231 /* Set or clear specific file flags */
2232 /* XXX This probably needs checks to ensure the flags are
2233 * not abused, and to handle any flag side effects.
2235 if (get_user(flags, (int __user *)arg))
2238 if (cmd == LL_IOC_SETFLAGS) {
2239 if ((flags & LL_FILE_IGNORE_LOCK) &&
2240 !(file->f_flags & O_DIRECT)) {
2241 CERROR("%s: unable to disable locking on non-O_DIRECT file\n",
2246 fd->fd_flags |= flags;
2248 fd->fd_flags &= ~flags;
2251 case LL_IOC_LOV_SETSTRIPE:
2252 return ll_lov_setstripe(inode, file, arg);
2253 case LL_IOC_LOV_SETEA:
2254 return ll_lov_setea(inode, file, arg);
2255 case LL_IOC_LOV_SWAP_LAYOUTS: {
2257 struct lustre_swap_layouts lsl;
2259 if (copy_from_user(&lsl, (char __user *)arg,
2260 sizeof(struct lustre_swap_layouts)))
2263 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
2266 file2 = fget(lsl.sl_fd);
2271 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
2272 rc = ll_swap_layouts(file, file2, &lsl);
2276 case LL_IOC_LOV_GETSTRIPE:
2277 return ll_lov_getstripe(inode, arg);
2278 case LL_IOC_RECREATE_OBJ:
2279 return ll_lov_recreate_obj(inode, arg);
2280 case LL_IOC_RECREATE_FID:
2281 return ll_lov_recreate_fid(inode, arg);
2282 case FSFILT_IOC_FIEMAP:
2283 return ll_ioctl_fiemap(inode, arg);
2284 case FSFILT_IOC_GETFLAGS:
2285 case FSFILT_IOC_SETFLAGS:
2286 return ll_iocontrol(inode, file, cmd, arg);
2287 case FSFILT_IOC_GETVERSION_OLD:
2288 case FSFILT_IOC_GETVERSION:
2289 return put_user(inode->i_generation, (int __user *)arg);
2290 case LL_IOC_GROUP_LOCK:
2291 return ll_get_grouplock(inode, file, arg);
2292 case LL_IOC_GROUP_UNLOCK:
2293 return ll_put_grouplock(inode, file, arg);
2294 case IOC_OBD_STATFS:
2295 return ll_obd_statfs(inode, (void __user *)arg);
2297 /* We need to special case any other ioctls we want to handle,
2298 * to send them to the MDS/OST as appropriate and to properly
2299 * network encode the arg field.
2300 case FSFILT_IOC_SETVERSION_OLD:
2301 case FSFILT_IOC_SETVERSION:
2303 case LL_IOC_FLUSHCTX:
2304 return ll_flush_ctx(inode);
2305 case LL_IOC_PATH2FID: {
2306 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2307 sizeof(struct lu_fid)))
2312 case OBD_IOC_FID2PATH:
2313 return ll_fid2path(inode, (void __user *)arg);
2314 case LL_IOC_DATA_VERSION: {
2315 struct ioc_data_version idv;
2318 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2321 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2322 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2323 if (rc == 0 && copy_to_user((char __user *)arg, &idv,
2330 case LL_IOC_GET_MDTIDX: {
2333 mdtidx = ll_get_mdt_idx(inode);
2337 if (put_user(mdtidx, (int __user *)arg))
2342 case OBD_IOC_GETDTNAME:
2343 case OBD_IOC_GETMDNAME:
2344 return ll_get_obd_name(inode, cmd, arg);
2345 case LL_IOC_HSM_STATE_GET: {
2346 struct md_op_data *op_data;
2347 struct hsm_user_state *hus;
2350 hus = kzalloc(sizeof(*hus), GFP_NOFS);
2354 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2355 LUSTRE_OPC_ANY, hus);
2356 if (IS_ERR(op_data)) {
2358 return PTR_ERR(op_data);
2361 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2364 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2367 ll_finish_md_op_data(op_data);
2371 case LL_IOC_HSM_STATE_SET: {
2372 struct hsm_state_set *hss;
2375 hss = memdup_user((char __user *)arg, sizeof(*hss));
2377 return PTR_ERR(hss);
2379 rc = ll_hsm_state_set(inode, hss);
2384 case LL_IOC_HSM_ACTION: {
2385 struct md_op_data *op_data;
2386 struct hsm_current_action *hca;
2389 hca = kzalloc(sizeof(*hca), GFP_NOFS);
2393 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2394 LUSTRE_OPC_ANY, hca);
2395 if (IS_ERR(op_data)) {
2397 return PTR_ERR(op_data);
2400 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2403 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2406 ll_finish_md_op_data(op_data);
2410 case LL_IOC_SET_LEASE: {
2411 struct ll_inode_info *lli = ll_i2info(inode);
2412 struct obd_client_handle *och = NULL;
2418 if (!(file->f_mode & FMODE_WRITE))
2423 if (!(file->f_mode & FMODE_READ))
2428 mutex_lock(&lli->lli_och_mutex);
2429 if (fd->fd_lease_och) {
2430 och = fd->fd_lease_och;
2431 fd->fd_lease_och = NULL;
2433 mutex_unlock(&lli->lli_och_mutex);
2436 mode = och->och_flags &
2437 (FMODE_READ|FMODE_WRITE);
2438 rc = ll_lease_close(och, inode, &lease_broken);
2439 if (rc == 0 && lease_broken)
2445 /* return the type of lease or error */
2446 return rc < 0 ? rc : (int)mode;
2451 CDEBUG(D_INODE, "Set lease with mode %d\n", mode);
2453 /* apply for lease */
2454 och = ll_lease_open(inode, file, mode, 0);
2456 return PTR_ERR(och);
2459 mutex_lock(&lli->lli_och_mutex);
2460 if (!fd->fd_lease_och) {
2461 fd->fd_lease_och = och;
2464 mutex_unlock(&lli->lli_och_mutex);
2466 /* impossible now that only excl is supported for now */
2467 ll_lease_close(och, inode, &lease_broken);
2472 case LL_IOC_GET_LEASE: {
2473 struct ll_inode_info *lli = ll_i2info(inode);
2474 struct ldlm_lock *lock = NULL;
2477 mutex_lock(&lli->lli_och_mutex);
2478 if (fd->fd_lease_och) {
2479 struct obd_client_handle *och = fd->fd_lease_och;
2481 lock = ldlm_handle2lock(&och->och_lease_handle);
2483 lock_res_and_lock(lock);
2484 if (!ldlm_is_cancel(lock))
2485 rc = och->och_flags &
2486 (FMODE_READ | FMODE_WRITE);
2487 unlock_res_and_lock(lock);
2488 LDLM_LOCK_PUT(lock);
2491 mutex_unlock(&lli->lli_och_mutex);
2494 case LL_IOC_HSM_IMPORT: {
2495 struct hsm_user_import *hui;
2497 hui = memdup_user((void __user *)arg, sizeof(*hui));
2499 return PTR_ERR(hui);
2501 rc = ll_hsm_import(inode, file, hui);
2509 if (ll_iocontrol_call(inode, file, cmd, arg, &err) ==
2513 return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2514 (void __user *)arg);
2519 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2521 struct inode *inode = file_inode(file);
2522 loff_t retval, eof = 0;
2524 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2525 (origin == SEEK_CUR) ? file->f_pos : 0);
2526 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2527 PFID(ll_inode2fid(inode)), inode, retval, retval, origin);
2528 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2530 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2531 retval = ll_glimpse_size(inode);
2534 eof = i_size_read(inode);
2537 retval = generic_file_llseek_size(file, offset, origin,
2538 ll_file_maxbytes(inode), eof);
2542 static int ll_flush(struct file *file, fl_owner_t id)
2544 struct inode *inode = file_inode(file);
2545 struct ll_inode_info *lli = ll_i2info(inode);
2546 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2549 LASSERT(!S_ISDIR(inode->i_mode));
2551 /* catch async errors that were recorded back when async writeback
2552 * failed for pages in this mapping.
2554 rc = lli->lli_async_rc;
2555 lli->lli_async_rc = 0;
2556 err = lov_read_and_clear_async_rc(lli->lli_clob);
2560 /* The application has been told about write failure already.
2561 * Do not report failure again.
2563 if (fd->fd_write_failed)
2565 return rc ? -EIO : 0;
2569 * Called to make sure a portion of file has been written out.
2570 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2572 * Return how many pages have been written.
2574 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2575 enum cl_fsync_mode mode, int ignore_layout)
2577 struct cl_env_nest nest;
2580 struct cl_fsync_io *fio;
2583 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2584 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2587 env = cl_env_nested_get(&nest);
2589 return PTR_ERR(env);
2591 io = vvp_env_thread_io(env);
2592 io->ci_obj = ll_i2info(inode)->lli_clob;
2593 io->ci_ignore_layout = ignore_layout;
2595 /* initialize parameters for sync */
2596 fio = &io->u.ci_fsync;
2597 fio->fi_start = start;
2599 fio->fi_fid = ll_inode2fid(inode);
2600 fio->fi_mode = mode;
2601 fio->fi_nr_written = 0;
2603 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2604 result = cl_io_loop(env, io);
2606 result = io->ci_result;
2608 result = fio->fi_nr_written;
2609 cl_io_fini(env, io);
2610 cl_env_nested_put(&nest, env);
2615 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2617 struct inode *inode = file_inode(file);
2618 struct ll_inode_info *lli = ll_i2info(inode);
2619 struct ptlrpc_request *req;
2622 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2623 PFID(ll_inode2fid(inode)), inode);
2624 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2626 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2629 /* catch async errors that were recorded back when async writeback
2630 * failed for pages in this mapping.
2632 if (!S_ISDIR(inode->i_mode)) {
2633 err = lli->lli_async_rc;
2634 lli->lli_async_rc = 0;
2637 err = lov_read_and_clear_async_rc(lli->lli_clob);
2642 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
2646 ptlrpc_req_finished(req);
2648 if (S_ISREG(inode->i_mode)) {
2649 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2651 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2652 if (rc == 0 && err < 0)
2655 fd->fd_write_failed = true;
2657 fd->fd_write_failed = false;
2660 inode_unlock(inode);
2665 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2667 struct inode *inode = file_inode(file);
2668 struct ll_sb_info *sbi = ll_i2sbi(inode);
2669 struct ldlm_enqueue_info einfo = {
2670 .ei_type = LDLM_FLOCK,
2671 .ei_cb_cp = ldlm_flock_completion_ast,
2672 .ei_cbdata = file_lock,
2674 struct md_op_data *op_data;
2675 struct lustre_handle lockh = {0};
2676 ldlm_policy_data_t flock = { {0} };
2681 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2682 PFID(ll_inode2fid(inode)), file_lock);
2684 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2686 if (file_lock->fl_flags & FL_FLOCK)
2687 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2688 else if (!(file_lock->fl_flags & FL_POSIX))
2691 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2692 flock.l_flock.pid = file_lock->fl_pid;
2693 flock.l_flock.start = file_lock->fl_start;
2694 flock.l_flock.end = file_lock->fl_end;
2696 /* Somewhat ugly workaround for svc lockd.
2697 * lockd installs custom fl_lmops->lm_compare_owner that checks
2698 * for the fl_owner to be the same (which it always is on local node
2699 * I guess between lockd processes) and then compares pid.
2700 * As such we assign pid to the owner field to make it all work,
2701 * conflict with normal locks is unlikely since pid space and
2702 * pointer space for current->files are not intersecting
2704 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2705 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2707 switch (file_lock->fl_type) {
2709 einfo.ei_mode = LCK_PR;
2712 /* An unlock request may or may not have any relation to
2713 * existing locks so we may not be able to pass a lock handle
2714 * via a normal ldlm_lock_cancel() request. The request may even
2715 * unlock a byte range in the middle of an existing lock. In
2716 * order to process an unlock request we need all of the same
2717 * information that is given with a normal read or write record
2718 * lock request. To avoid creating another ldlm unlock (cancel)
2719 * message we'll treat a LCK_NL flock request as an unlock.
2721 einfo.ei_mode = LCK_NL;
2724 einfo.ei_mode = LCK_PW;
2727 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
2728 file_lock->fl_type);
2743 flags = LDLM_FL_BLOCK_NOWAIT;
2749 flags = LDLM_FL_TEST_LOCK;
2750 /* Save the old mode so that if the mode in the lock changes we
2751 * can decrement the appropriate reader or writer refcount.
2753 file_lock->fl_type = einfo.ei_mode;
2756 CERROR("unknown fcntl lock command: %d\n", cmd);
2760 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2761 LUSTRE_OPC_ANY, NULL);
2762 if (IS_ERR(op_data))
2763 return PTR_ERR(op_data);
2765 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n",
2766 PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags,
2767 einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2769 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2770 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2772 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
2773 !(flags & LDLM_FL_TEST_LOCK))
2774 rc2 = locks_lock_file_wait(file, file_lock);
2776 if (rc2 && file_lock->fl_type != F_UNLCK) {
2777 einfo.ei_mode = LCK_NL;
2778 md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2779 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2783 ll_finish_md_op_data(op_data);
2789 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2795 * test if some locks matching bits and l_req_mode are acquired
2796 * - bits can be in different locks
2797 * - if found clear the common lock bits in *bits
2798 * - the bits not found, are kept in *bits
2800 * \param bits [IN] searched lock bits [IN]
2801 * \param l_req_mode [IN] searched lock mode
2802 * \retval boolean, true iff all bits are found
2804 int ll_have_md_lock(struct inode *inode, __u64 *bits,
2805 enum ldlm_mode l_req_mode)
2807 struct lustre_handle lockh;
2808 ldlm_policy_data_t policy;
2809 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
2810 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
2818 fid = &ll_i2info(inode)->lli_fid;
2819 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
2820 ldlm_lockname[mode]);
2822 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2823 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2824 policy.l_inodebits.bits = *bits & (1 << i);
2825 if (policy.l_inodebits.bits == 0)
2828 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2829 &policy, mode, &lockh)) {
2830 struct ldlm_lock *lock;
2832 lock = ldlm_handle2lock(&lockh);
2835 ~(lock->l_policy_data.l_inodebits.bits);
2836 LDLM_LOCK_PUT(lock);
2838 *bits &= ~policy.l_inodebits.bits;
2845 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
2846 struct lustre_handle *lockh, __u64 flags,
2847 enum ldlm_mode mode)
2849 ldlm_policy_data_t policy = { .l_inodebits = {bits} };
2853 fid = &ll_i2info(inode)->lli_fid;
2854 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2856 rc = md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED,
2857 fid, LDLM_IBITS, &policy, mode, lockh);
2862 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2864 /* Already unlinked. Just update nlink and return success */
2865 if (rc == -ENOENT) {
2867 /* This path cannot be hit for regular files unless in
2868 * case of obscure races, so no need to validate size.
2870 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2872 } else if (rc != 0) {
2873 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
2874 "%s: revalidate FID "DFID" error: rc = %d\n",
2875 ll_get_fsname(inode->i_sb, NULL, 0),
2876 PFID(ll_inode2fid(inode)), rc);
2882 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2884 struct inode *inode = d_inode(dentry);
2885 struct ptlrpc_request *req = NULL;
2886 struct obd_export *exp;
2889 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%pd\n",
2890 PFID(ll_inode2fid(inode)), inode, dentry);
2892 exp = ll_i2mdexp(inode);
2894 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2895 * But under CMD case, it caused some lock issues, should be fixed
2896 * with new CMD ibits lock. See bug 12718
2898 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
2899 struct lookup_intent oit = { .it_op = IT_GETATTR };
2900 struct md_op_data *op_data;
2902 if (ibits == MDS_INODELOCK_LOOKUP)
2903 oit.it_op = IT_LOOKUP;
2905 /* Call getattr by fid, so do not provide name at all. */
2906 op_data = ll_prep_md_op_data(NULL, inode,
2908 LUSTRE_OPC_ANY, NULL);
2909 if (IS_ERR(op_data))
2910 return PTR_ERR(op_data);
2912 oit.it_create_mode |= M_CHECK_STALE;
2913 rc = md_intent_lock(exp, op_data, NULL, 0,
2914 /* we are not interested in name
2918 ll_md_blocking_ast, 0);
2919 ll_finish_md_op_data(op_data);
2920 oit.it_create_mode &= ~M_CHECK_STALE;
2922 rc = ll_inode_revalidate_fini(inode, rc);
2926 rc = ll_revalidate_it_finish(req, &oit, inode);
2928 ll_intent_release(&oit);
2932 /* Unlinked? Unhash dentry, so it is not picked up later by
2933 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2934 * here to preserve get_cwd functionality on 2.6.
2937 if (!d_inode(dentry)->i_nlink) {
2938 spin_lock(&inode->i_lock);
2939 d_lustre_invalidate(dentry, 0);
2940 spin_unlock(&inode->i_lock);
2943 ll_lookup_finish_locks(&oit, inode);
2944 } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) {
2945 struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry));
2946 u64 valid = OBD_MD_FLGETATTR;
2947 struct md_op_data *op_data;
2950 if (S_ISREG(inode->i_mode)) {
2951 rc = ll_get_default_mdsize(sbi, &ealen);
2954 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2957 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2958 0, ealen, LUSTRE_OPC_ANY,
2960 if (IS_ERR(op_data))
2961 return PTR_ERR(op_data);
2963 op_data->op_valid = valid;
2964 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2965 ll_finish_md_op_data(op_data);
2967 rc = ll_inode_revalidate_fini(inode, rc);
2971 rc = ll_prep_inode(&inode, req, NULL, NULL);
2974 ptlrpc_req_finished(req);
2978 static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2980 struct inode *inode = d_inode(dentry);
2983 rc = __ll_inode_revalidate(dentry, ibits);
2987 /* if object isn't regular file, don't validate size */
2988 if (!S_ISREG(inode->i_mode)) {
2989 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
2990 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
2991 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
2993 /* In case of restore, the MDT has the right size and has
2994 * already send it back without granting the layout lock,
2995 * inode is up-to-date so glimpse is useless.
2996 * Also to glimpse we need the layout, in case of a running
2997 * restore the MDT holds the layout lock so the glimpse will
2998 * block up to the end of restore (getattr will block)
3000 if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING))
3001 rc = ll_glimpse_size(inode);
3006 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3008 struct inode *inode = d_inode(de);
3009 struct ll_sb_info *sbi = ll_i2sbi(inode);
3010 struct ll_inode_info *lli = ll_i2info(inode);
3013 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3014 MDS_INODELOCK_LOOKUP);
3015 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3020 stat->dev = inode->i_sb->s_dev;
3021 if (ll_need_32bit_api(sbi))
3022 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3024 stat->ino = inode->i_ino;
3025 stat->mode = inode->i_mode;
3026 stat->nlink = inode->i_nlink;
3027 stat->uid = inode->i_uid;
3028 stat->gid = inode->i_gid;
3029 stat->rdev = inode->i_rdev;
3030 stat->atime = inode->i_atime;
3031 stat->mtime = inode->i_mtime;
3032 stat->ctime = inode->i_ctime;
3033 stat->blksize = 1 << inode->i_blkbits;
3035 stat->size = i_size_read(inode);
3036 stat->blocks = inode->i_blocks;
3041 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3042 __u64 start, __u64 len)
3046 struct ll_user_fiemap *fiemap;
3047 unsigned int extent_count = fieinfo->fi_extents_max;
3049 num_bytes = sizeof(*fiemap) + (extent_count *
3050 sizeof(struct ll_fiemap_extent));
3051 fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS);
3056 fiemap->fm_flags = fieinfo->fi_flags;
3057 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3058 fiemap->fm_start = start;
3059 fiemap->fm_length = len;
3060 if (extent_count > 0 &&
3061 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3062 sizeof(struct ll_fiemap_extent)) != 0) {
3067 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3069 fieinfo->fi_flags = fiemap->fm_flags;
3070 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3071 if (extent_count > 0 &&
3072 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3073 fiemap->fm_mapped_extents *
3074 sizeof(struct ll_fiemap_extent)) != 0) {
3084 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3086 struct ll_inode_info *lli = ll_i2info(inode);
3087 struct posix_acl *acl = NULL;
3089 spin_lock(&lli->lli_lock);
3090 /* VFS' acl_permission_check->check_acl will release the refcount */
3091 acl = posix_acl_dup(lli->lli_posix_acl);
3092 #ifdef CONFIG_FS_POSIX_ACL
3093 forget_cached_acl(inode, type);
3095 spin_unlock(&lli->lli_lock);
3100 int ll_inode_permission(struct inode *inode, int mask)
3104 if (mask & MAY_NOT_BLOCK)
3107 /* as root inode are NOT getting validated in lookup operation,
3108 * need to do it before permission check.
3111 if (is_root_inode(inode)) {
3112 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3113 MDS_INODELOCK_LOOKUP);
3118 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3119 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3121 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
3122 rc = generic_permission(inode, mask);
3127 /* -o localflock - only provides locally consistent flock locks */
3128 struct file_operations ll_file_operations = {
3129 .read_iter = ll_file_read_iter,
3130 .write_iter = ll_file_write_iter,
3131 .unlocked_ioctl = ll_file_ioctl,
3132 .open = ll_file_open,
3133 .release = ll_file_release,
3134 .mmap = ll_file_mmap,
3135 .llseek = ll_file_seek,
3136 .splice_read = generic_file_splice_read,
3141 struct file_operations ll_file_operations_flock = {
3142 .read_iter = ll_file_read_iter,
3143 .write_iter = ll_file_write_iter,
3144 .unlocked_ioctl = ll_file_ioctl,
3145 .open = ll_file_open,
3146 .release = ll_file_release,
3147 .mmap = ll_file_mmap,
3148 .llseek = ll_file_seek,
3149 .splice_read = generic_file_splice_read,
3152 .flock = ll_file_flock,
3153 .lock = ll_file_flock
3156 /* These are for -o noflock - to return ENOSYS on flock calls */
3157 struct file_operations ll_file_operations_noflock = {
3158 .read_iter = ll_file_read_iter,
3159 .write_iter = ll_file_write_iter,
3160 .unlocked_ioctl = ll_file_ioctl,
3161 .open = ll_file_open,
3162 .release = ll_file_release,
3163 .mmap = ll_file_mmap,
3164 .llseek = ll_file_seek,
3165 .splice_read = generic_file_splice_read,
3168 .flock = ll_file_noflock,
3169 .lock = ll_file_noflock
3172 const struct inode_operations ll_file_inode_operations = {
3173 .setattr = ll_setattr,
3174 .getattr = ll_getattr,
3175 .permission = ll_inode_permission,
3176 .setxattr = ll_setxattr,
3177 .getxattr = ll_getxattr,
3178 .listxattr = ll_listxattr,
3179 .removexattr = ll_removexattr,
3180 .fiemap = ll_fiemap,
3181 .get_acl = ll_get_acl,
3184 /* dynamic ioctl number support routines */
3185 static struct llioc_ctl_data {
3186 struct rw_semaphore ioc_sem;
3187 struct list_head ioc_head;
3189 __RWSEM_INITIALIZER(llioc.ioc_sem),
3190 LIST_HEAD_INIT(llioc.ioc_head)
3194 struct list_head iocd_list;
3195 unsigned int iocd_size;
3196 llioc_callback_t iocd_cb;
3197 unsigned int iocd_count;
3198 unsigned int iocd_cmd[0];
3201 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3204 struct llioc_data *in_data = NULL;
3206 if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0)
3209 size = sizeof(*in_data) + count * sizeof(unsigned int);
3210 in_data = kzalloc(size, GFP_NOFS);
3214 memset(in_data, 0, sizeof(*in_data));
3215 in_data->iocd_size = size;
3216 in_data->iocd_cb = cb;
3217 in_data->iocd_count = count;
3218 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3220 down_write(&llioc.ioc_sem);
3221 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3222 up_write(&llioc.ioc_sem);
3226 EXPORT_SYMBOL(ll_iocontrol_register);
3228 void ll_iocontrol_unregister(void *magic)
3230 struct llioc_data *tmp;
3235 down_write(&llioc.ioc_sem);
3236 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3238 list_del(&tmp->iocd_list);
3239 up_write(&llioc.ioc_sem);
3245 up_write(&llioc.ioc_sem);
3247 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3249 EXPORT_SYMBOL(ll_iocontrol_unregister);
3251 static enum llioc_iter
3252 ll_iocontrol_call(struct inode *inode, struct file *file,
3253 unsigned int cmd, unsigned long arg, int *rcp)
3255 enum llioc_iter ret = LLIOC_CONT;
3256 struct llioc_data *data;
3257 int rc = -EINVAL, i;
3259 down_read(&llioc.ioc_sem);
3260 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3261 for (i = 0; i < data->iocd_count; i++) {
3262 if (cmd != data->iocd_cmd[i])
3265 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3269 if (ret == LLIOC_STOP)
3272 up_read(&llioc.ioc_sem);
3279 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3281 struct ll_inode_info *lli = ll_i2info(inode);
3282 struct cl_env_nest nest;
3289 env = cl_env_nested_get(&nest);
3291 return PTR_ERR(env);
3293 result = cl_conf_set(env, lli->lli_clob, conf);
3294 cl_env_nested_put(&nest, env);
3296 if (conf->coc_opc == OBJECT_CONF_SET) {
3297 struct ldlm_lock *lock = conf->coc_lock;
3300 LASSERT(ldlm_has_layout(lock));
3302 /* it can only be allowed to match after layout is
3303 * applied to inode otherwise false layout would be
3304 * seen. Applying layout should happen before dropping
3307 ldlm_lock_allow_match(lock);
3313 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3314 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3317 struct ll_sb_info *sbi = ll_i2sbi(inode);
3318 struct ptlrpc_request *req;
3319 struct mdt_body *body;
3325 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3326 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3327 lock->l_lvb_data, lock->l_lvb_len);
3329 if (lock->l_lvb_data && ldlm_is_lvb_ready(lock))
3332 /* if layout lock was granted right away, the layout is returned
3333 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3334 * blocked and then granted via completion ast, we have to fetch
3335 * layout here. Please note that we can't use the LVB buffer in
3336 * completion AST because it doesn't have a large enough buffer
3338 rc = ll_get_default_mdsize(sbi, &lmmsize);
3340 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
3341 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3346 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3352 lmmsize = body->eadatasize;
3353 if (lmmsize == 0) /* empty layout */ {
3358 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3364 lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS);
3370 memcpy(lvbdata, lmm, lmmsize);
3371 lock_res_and_lock(lock);
3372 if (lock->l_lvb_data)
3373 kvfree(lock->l_lvb_data);
3375 lock->l_lvb_data = lvbdata;
3376 lock->l_lvb_len = lmmsize;
3377 unlock_res_and_lock(lock);
3380 ptlrpc_req_finished(req);
3385 * Apply the layout to the inode. Layout lock is held and will be released
3388 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
3389 struct inode *inode, __u32 *gen, bool reconf)
3391 struct ll_inode_info *lli = ll_i2info(inode);
3392 struct ll_sb_info *sbi = ll_i2sbi(inode);
3393 struct ldlm_lock *lock;
3394 struct lustre_md md = { NULL };
3395 struct cl_object_conf conf;
3398 bool wait_layout = false;
3400 LASSERT(lustre_handle_is_used(lockh));
3402 lock = ldlm_handle2lock(lockh);
3404 LASSERT(ldlm_has_layout(lock));
3406 LDLM_DEBUG(lock, "File "DFID"(%p) being reconfigured: %d",
3407 PFID(&lli->lli_fid), inode, reconf);
3409 /* in case this is a caching lock and reinstate with new inode */
3410 md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
3412 lock_res_and_lock(lock);
3413 lvb_ready = ldlm_is_lvb_ready(lock);
3414 unlock_res_and_lock(lock);
3415 /* checking lvb_ready is racy but this is okay. The worst case is
3416 * that multi processes may configure the file on the same time.
3418 if (lvb_ready || !reconf) {
3421 /* layout_gen must be valid if layout lock is not
3422 * cancelled and stripe has already set
3424 *gen = ll_layout_version_get(lli);
3430 rc = ll_layout_fetch(inode, lock);
3434 /* for layout lock, lmm is returned in lock's lvb.
3435 * lvb_data is immutable if the lock is held so it's safe to access it
3436 * without res lock. See the description in ldlm_lock_decref_internal()
3437 * for the condition to free lvb_data of layout lock
3439 if (lock->l_lvb_data) {
3440 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3441 lock->l_lvb_data, lock->l_lvb_len);
3443 *gen = LL_LAYOUT_GEN_EMPTY;
3445 *gen = md.lsm->lsm_layout_gen;
3448 CERROR("%s: file " DFID " unpackmd error: %d\n",
3449 ll_get_fsname(inode->i_sb, NULL, 0),
3450 PFID(&lli->lli_fid), rc);
3456 /* set layout to file. Unlikely this will fail as old layout was
3459 memset(&conf, 0, sizeof(conf));
3460 conf.coc_opc = OBJECT_CONF_SET;
3461 conf.coc_inode = inode;
3462 conf.coc_lock = lock;
3463 conf.u.coc_md = &md;
3464 rc = ll_layout_conf(inode, &conf);
3467 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3469 /* refresh layout failed, need to wait */
3470 wait_layout = rc == -EBUSY;
3473 LDLM_LOCK_PUT(lock);
3474 ldlm_lock_decref(lockh, mode);
3476 /* wait for IO to complete if it's still being used. */
3478 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3479 ll_get_fsname(inode->i_sb, NULL, 0),
3480 PFID(&lli->lli_fid), inode);
3482 memset(&conf, 0, sizeof(conf));
3483 conf.coc_opc = OBJECT_CONF_WAIT;
3484 conf.coc_inode = inode;
3485 rc = ll_layout_conf(inode, &conf);
3489 CDEBUG(D_INODE, "%s: file="DFID" waiting layout return: %d.\n",
3490 ll_get_fsname(inode->i_sb, NULL, 0),
3491 PFID(&lli->lli_fid), rc);
3497 * This function checks if there exists a LAYOUT lock on the client side,
3498 * or enqueues it if it doesn't have one in cache.
3500 * This function will not hold layout lock so it may be revoked any time after
3501 * this function returns. Any operations depend on layout should be redone
3504 * This function should be called before lov_io_init() to get an uptodate
3505 * layout version, the caller should save the version number and after IO
3506 * is finished, this function should be called again to verify that layout
3507 * is not changed during IO time.
3509 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3511 struct ll_inode_info *lli = ll_i2info(inode);
3512 struct ll_sb_info *sbi = ll_i2sbi(inode);
3513 struct md_op_data *op_data;
3514 struct lookup_intent it;
3515 struct lustre_handle lockh;
3516 enum ldlm_mode mode;
3517 struct ldlm_enqueue_info einfo = {
3518 .ei_type = LDLM_IBITS,
3520 .ei_cb_bl = ll_md_blocking_ast,
3521 .ei_cb_cp = ldlm_completion_ast,
3525 *gen = ll_layout_version_get(lli);
3526 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
3530 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3531 LASSERT(S_ISREG(inode->i_mode));
3533 /* take layout lock mutex to enqueue layout lock exclusively. */
3534 mutex_lock(&lli->lli_layout_mutex);
3537 /* mostly layout lock is caching on the local side, so try to match
3538 * it before grabbing layout lock mutex.
3540 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
3541 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
3542 if (mode != 0) { /* hit cached lock */
3543 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3547 mutex_unlock(&lli->lli_layout_mutex);
3551 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3552 0, 0, LUSTRE_OPC_ANY, NULL);
3553 if (IS_ERR(op_data)) {
3554 mutex_unlock(&lli->lli_layout_mutex);
3555 return PTR_ERR(op_data);
3558 /* have to enqueue one */
3559 memset(&it, 0, sizeof(it));
3560 it.it_op = IT_LAYOUT;
3561 lockh.cookie = 0ULL;
3563 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
3564 ll_get_fsname(inode->i_sb, NULL, 0),
3565 PFID(&lli->lli_fid), inode);
3567 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
3569 ptlrpc_req_finished(it.it_request);
3570 it.it_request = NULL;
3572 ll_finish_md_op_data(op_data);
3574 mode = it.it_lock_mode;
3575 it.it_lock_mode = 0;
3576 ll_intent_drop_lock(&it);
3579 /* set lock data in case this is a new lock */
3580 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3581 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3585 mutex_unlock(&lli->lli_layout_mutex);
3591 * This function send a restore request to the MDT
3593 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
3595 struct hsm_user_request *hur;
3598 len = sizeof(struct hsm_user_request) +
3599 sizeof(struct hsm_user_item);
3600 hur = kzalloc(len, GFP_NOFS);
3604 hur->hur_request.hr_action = HUA_RESTORE;
3605 hur->hur_request.hr_archive_id = 0;
3606 hur->hur_request.hr_flags = 0;
3607 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
3608 sizeof(hur->hur_user_item[0].hui_fid));
3609 hur->hur_user_item[0].hui_extent.offset = offset;
3610 hur->hur_user_item[0].hui_extent.length = length;
3611 hur->hur_request.hr_itemcount = 1;
3612 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,