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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
40 * Author: Andreas Dilger <adilger@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <lustre_dlm.h>
45 #include <lustre_lite.h>
46 #include <linux/pagemap.h>
47 #include <linux/file.h>
48 #include <linux/sched.h>
49 #include "llite_internal.h"
50 #include <lustre/ll_fiemap.h>
51 #include <lustre_ioctl.h>
53 #include "cl_object.h"
56 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
58 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
61 static enum llioc_iter
62 ll_iocontrol_call(struct inode *inode, struct file *file,
63 unsigned int cmd, unsigned long arg, int *rcp);
65 static struct ll_file_data *ll_file_data_get(void)
67 struct ll_file_data *fd;
69 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
73 fd->fd_write_failed = false;
78 static void ll_file_data_put(struct ll_file_data *fd)
81 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
84 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
85 struct lustre_handle *fh)
87 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
88 op_data->op_attr.ia_mode = inode->i_mode;
89 op_data->op_attr.ia_atime = inode->i_atime;
90 op_data->op_attr.ia_mtime = inode->i_mtime;
91 op_data->op_attr.ia_ctime = inode->i_ctime;
92 op_data->op_attr.ia_size = i_size_read(inode);
93 op_data->op_attr_blocks = inode->i_blocks;
94 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags =
95 ll_inode_to_ext_flags(inode->i_flags);
96 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
98 op_data->op_handle = *fh;
99 op_data->op_capa1 = ll_mdscapa_get(inode);
101 if (LLIF_DATA_MODIFIED & ll_i2info(inode)->lli_flags)
102 op_data->op_bias |= MDS_DATA_MODIFIED;
106 * Closes the IO epoch and packs all the attributes into @op_data for
109 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
110 struct obd_client_handle *och)
114 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
115 ATTR_MTIME | ATTR_MTIME_SET |
116 ATTR_CTIME | ATTR_CTIME_SET;
118 if (!(och->och_flags & FMODE_WRITE))
121 if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
122 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
124 ll_ioepoch_close(inode, op_data, &och, 0);
127 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
128 ll_prep_md_op_data(op_data, inode, NULL, NULL,
129 0, 0, LUSTRE_OPC_ANY, NULL);
133 static int ll_close_inode_openhandle(struct obd_export *md_exp,
135 struct obd_client_handle *och,
136 const __u64 *data_version)
138 struct obd_export *exp = ll_i2mdexp(inode);
139 struct md_op_data *op_data;
140 struct ptlrpc_request *req = NULL;
141 struct obd_device *obd = class_exp2obd(exp);
148 * XXX: in case of LMV, is this correct to access
151 CERROR("Invalid MDC connection handle "LPX64"\n",
152 ll_i2mdexp(inode)->exp_handle.h_cookie);
156 OBD_ALLOC_PTR(op_data);
158 GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here.
160 ll_prepare_close(inode, op_data, och);
161 if (data_version != NULL) {
162 /* Pass in data_version implies release. */
163 op_data->op_bias |= MDS_HSM_RELEASE;
164 op_data->op_data_version = *data_version;
165 op_data->op_lease_handle = och->och_lease_handle;
166 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
168 epoch_close = (op_data->op_flags & MF_EPOCH_CLOSE);
169 rc = md_close(md_exp, op_data, och->och_mod, &req);
171 /* This close must have the epoch closed. */
172 LASSERT(epoch_close);
173 /* MDS has instructed us to obtain Size-on-MDS attribute from
174 * OSTs and send setattr to back to MDS. */
175 rc = ll_som_update(inode, op_data);
177 CERROR("%s: inode "DFID" mdc Size-on-MDS update"
178 " failed: rc = %d\n",
179 ll_i2mdexp(inode)->exp_obd->obd_name,
180 PFID(ll_inode2fid(inode)), rc);
184 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
185 ll_i2mdexp(inode)->exp_obd->obd_name,
186 PFID(ll_inode2fid(inode)), rc);
189 /* DATA_MODIFIED flag was successfully sent on close, cancel data
190 * modification flag. */
191 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
192 struct ll_inode_info *lli = ll_i2info(inode);
194 spin_lock(&lli->lli_lock);
195 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
196 spin_unlock(&lli->lli_lock);
200 rc = ll_objects_destroy(req, inode);
202 CERROR("%s: inode "DFID
203 " ll_objects destroy: rc = %d\n",
204 ll_i2mdexp(inode)->exp_obd->obd_name,
205 PFID(ll_inode2fid(inode)), rc);
208 if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
209 struct mdt_body *body;
210 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
211 if (!(body->mbo_valid & OBD_MD_FLRELEASED))
215 ll_finish_md_op_data(op_data);
219 if (exp_connect_som(exp) && !epoch_close &&
220 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
221 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
223 md_clear_open_replay_data(md_exp, och);
224 /* Free @och if it is not waiting for DONE_WRITING. */
225 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
228 if (req) /* This is close request */
229 ptlrpc_req_finished(req);
233 int ll_md_real_close(struct inode *inode, fmode_t fmode)
235 struct ll_inode_info *lli = ll_i2info(inode);
236 struct obd_client_handle **och_p;
237 struct obd_client_handle *och;
242 if (fmode & FMODE_WRITE) {
243 och_p = &lli->lli_mds_write_och;
244 och_usecount = &lli->lli_open_fd_write_count;
245 } else if (fmode & FMODE_EXEC) {
246 och_p = &lli->lli_mds_exec_och;
247 och_usecount = &lli->lli_open_fd_exec_count;
249 LASSERT(fmode & FMODE_READ);
250 och_p = &lli->lli_mds_read_och;
251 och_usecount = &lli->lli_open_fd_read_count;
254 mutex_lock(&lli->lli_och_mutex);
255 if (*och_usecount > 0) {
256 /* There are still users of this handle, so skip
258 mutex_unlock(&lli->lli_och_mutex);
264 mutex_unlock(&lli->lli_och_mutex);
267 /* There might be a race and this handle may already
269 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
276 static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
279 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
280 struct ll_inode_info *lli = ll_i2info(inode);
284 /* clear group lock, if present */
285 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
286 ll_put_grouplock(inode, file, fd->fd_grouplock.cg_gid);
288 if (fd->fd_lease_och != NULL) {
291 /* Usually the lease is not released when the
292 * application crashed, we need to release here. */
293 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
294 CDEBUG(rc ? D_ERROR : D_INODE, "Clean up lease "DFID" %d/%d\n",
295 PFID(&lli->lli_fid), rc, lease_broken);
297 fd->fd_lease_och = NULL;
300 if (fd->fd_och != NULL) {
301 rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL);
306 /* Let's see if we have good enough OPEN lock on the file and if
307 we can skip talking to MDS */
308 if (file->f_dentry->d_inode) { /* Can this ever be false? */
310 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
311 struct lustre_handle lockh;
312 struct inode *inode = file->f_dentry->d_inode;
313 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
315 mutex_lock(&lli->lli_och_mutex);
316 if (fd->fd_omode & FMODE_WRITE) {
318 LASSERT(lli->lli_open_fd_write_count);
319 lli->lli_open_fd_write_count--;
320 } else if (fd->fd_omode & FMODE_EXEC) {
322 LASSERT(lli->lli_open_fd_exec_count);
323 lli->lli_open_fd_exec_count--;
326 LASSERT(lli->lli_open_fd_read_count);
327 lli->lli_open_fd_read_count--;
329 mutex_unlock(&lli->lli_och_mutex);
331 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
332 LDLM_IBITS, &policy, lockmode,
334 rc = ll_md_real_close(file->f_dentry->d_inode,
338 CERROR("released file has negative dentry: file = %p, "
339 "dentry = %p, name = %s\n",
340 file, file->f_dentry, file->f_dentry->d_name.name);
344 LUSTRE_FPRIVATE(file) = NULL;
345 ll_file_data_put(fd);
346 ll_capa_close(inode);
351 /* While this returns an error code, fput() the caller does not, so we need
352 * to make every effort to clean up all of our state here. Also, applications
353 * rarely check close errors and even if an error is returned they will not
354 * re-try the close call.
356 int ll_file_release(struct inode *inode, struct file *file)
358 struct ll_file_data *fd;
359 struct ll_sb_info *sbi = ll_i2sbi(inode);
360 struct ll_inode_info *lli = ll_i2info(inode);
364 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
365 PFID(ll_inode2fid(inode)), inode);
367 #ifdef CONFIG_FS_POSIX_ACL
368 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
369 inode == inode->i_sb->s_root->d_inode) {
370 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
373 if (unlikely(fd->fd_flags & LL_FILE_RMTACL)) {
374 fd->fd_flags &= ~LL_FILE_RMTACL;
375 rct_del(&sbi->ll_rct, current_pid());
376 et_search_free(&sbi->ll_et, current_pid());
381 if (inode->i_sb->s_root != file->f_dentry)
382 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
383 fd = LUSTRE_FPRIVATE(file);
386 /* The last ref on @file, maybe not the the owner pid of statahead,
387 * because parent and child process can share the same file handle. */
388 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
389 ll_deauthorize_statahead(inode, fd);
391 if (inode->i_sb->s_root == file->f_dentry) {
392 LUSTRE_FPRIVATE(file) = NULL;
393 ll_file_data_put(fd);
397 if (!S_ISDIR(inode->i_mode)) {
398 if (lli->lli_clob != NULL)
399 lov_read_and_clear_async_rc(lli->lli_clob);
400 lli->lli_async_rc = 0;
403 rc = ll_md_close(sbi->ll_md_exp, inode, file);
405 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
406 libcfs_debug_dumplog();
411 static int ll_intent_file_open(struct file *file, void *lmm, int lmmsize,
412 struct lookup_intent *itp)
414 struct dentry *de = file->f_dentry;
415 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
416 struct dentry *parent = de->d_parent;
417 const char *name = NULL;
419 struct md_op_data *op_data;
420 struct ptlrpc_request *req = NULL;
424 LASSERT(parent != NULL);
425 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
427 /* if server supports open-by-fid, or file name is invalid, don't pack
428 * name in open request */
429 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
430 lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
431 name = de->d_name.name;
432 len = de->d_name.len;
435 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
436 name, len, 0, LUSTRE_OPC_ANY, NULL);
438 RETURN(PTR_ERR(op_data));
439 op_data->op_data = lmm;
440 op_data->op_data_size = lmmsize;
442 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
443 &ll_md_blocking_ast, 0);
444 ll_finish_md_op_data(op_data);
446 /* reason for keep own exit path - don`t flood log
447 * with messages with -ESTALE errors.
449 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
450 it_open_error(DISP_OPEN_OPEN, itp))
452 ll_release_openhandle(de, itp);
456 if (it_disposition(itp, DISP_LOOKUP_NEG))
457 GOTO(out, rc = -ENOENT);
459 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
460 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
461 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
465 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
466 if (!rc && itp->d.lustre.it_lock_mode)
467 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
470 ptlrpc_req_finished(req);
471 ll_intent_drop_lock(itp);
477 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
478 * not believe attributes if a few ioepoch holders exist. Attributes for
479 * previous ioepoch if new one is opened are also skipped by MDS.
481 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
483 if (ioepoch && lli->lli_ioepoch != ioepoch) {
484 lli->lli_ioepoch = ioepoch;
485 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
486 ioepoch, PFID(&lli->lli_fid));
490 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
491 struct obd_client_handle *och)
493 struct ptlrpc_request *req = it->d.lustre.it_data;
494 struct mdt_body *body;
496 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
497 och->och_fh = body->mbo_handle;
498 och->och_fid = body->mbo_fid1;
499 och->och_lease_handle.cookie = it->d.lustre.it_lock_handle;
500 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
501 och->och_flags = it->it_flags;
503 return md_set_open_replay_data(md_exp, och, it);
506 static int ll_local_open(struct file *file, struct lookup_intent *it,
507 struct ll_file_data *fd, struct obd_client_handle *och)
509 struct inode *inode = file->f_dentry->d_inode;
510 struct ll_inode_info *lli = ll_i2info(inode);
513 LASSERT(!LUSTRE_FPRIVATE(file));
518 struct ptlrpc_request *req = it->d.lustre.it_data;
519 struct mdt_body *body;
522 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
526 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
527 ll_ioepoch_open(lli, body->mbo_ioepoch);
530 LUSTRE_FPRIVATE(file) = fd;
531 ll_readahead_init(inode, &fd->fd_ras);
532 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
534 /* ll_cl_context initialize */
535 rwlock_init(&fd->fd_lock);
536 INIT_LIST_HEAD(&fd->fd_lccs);
541 /* Open a file, and (for the very first open) create objects on the OSTs at
542 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
543 * creation or open until ll_lov_setstripe() ioctl is called.
545 * If we already have the stripe MD locally then we don't request it in
546 * md_open(), by passing a lmm_size = 0.
548 * It is up to the application to ensure no other processes open this file
549 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
550 * used. We might be able to avoid races of that sort by getting lli_open_sem
551 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
552 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
554 int ll_file_open(struct inode *inode, struct file *file)
556 struct ll_inode_info *lli = ll_i2info(inode);
557 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
558 .it_flags = file->f_flags };
559 struct obd_client_handle **och_p = NULL;
560 __u64 *och_usecount = NULL;
561 struct ll_file_data *fd;
565 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
566 PFID(ll_inode2fid(inode)), inode, file->f_flags);
568 it = file->private_data; /* XXX: compat macro */
569 file->private_data = NULL; /* prevent ll_local_open assertion */
571 fd = ll_file_data_get();
573 GOTO(out_openerr, rc = -ENOMEM);
576 if (S_ISDIR(inode->i_mode))
577 ll_authorize_statahead(inode, fd);
579 if (inode->i_sb->s_root == file->f_dentry) {
580 LUSTRE_FPRIVATE(file) = fd;
584 if (!it || !it->d.lustre.it_disposition) {
585 /* Convert f_flags into access mode. We cannot use file->f_mode,
586 * because everything but O_ACCMODE mask was stripped from
588 if ((oit.it_flags + 1) & O_ACCMODE)
590 if (file->f_flags & O_TRUNC)
591 oit.it_flags |= FMODE_WRITE;
593 /* kernel only call f_op->open in dentry_open. filp_open calls
594 * dentry_open after call to open_namei that checks permissions.
595 * Only nfsd_open call dentry_open directly without checking
596 * permissions and because of that this code below is safe. */
597 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
598 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
600 /* We do not want O_EXCL here, presumably we opened the file
601 * already? XXX - NFS implications? */
602 oit.it_flags &= ~O_EXCL;
604 /* bug20584, if "it_flags" contains O_CREAT, the file will be
605 * created if necessary, then "IT_CREAT" should be set to keep
606 * consistent with it */
607 if (oit.it_flags & O_CREAT)
608 oit.it_op |= IT_CREAT;
614 /* Let's see if we have file open on MDS already. */
615 if (it->it_flags & FMODE_WRITE) {
616 och_p = &lli->lli_mds_write_och;
617 och_usecount = &lli->lli_open_fd_write_count;
618 } else if (it->it_flags & FMODE_EXEC) {
619 och_p = &lli->lli_mds_exec_och;
620 och_usecount = &lli->lli_open_fd_exec_count;
622 och_p = &lli->lli_mds_read_och;
623 och_usecount = &lli->lli_open_fd_read_count;
626 mutex_lock(&lli->lli_och_mutex);
627 if (*och_p) { /* Open handle is present */
628 if (it_disposition(it, DISP_OPEN_OPEN)) {
629 /* Well, there's extra open request that we do not need,
630 let's close it somehow. This will decref request. */
631 rc = it_open_error(DISP_OPEN_OPEN, it);
633 mutex_unlock(&lli->lli_och_mutex);
634 GOTO(out_openerr, rc);
637 ll_release_openhandle(file->f_dentry, it);
641 rc = ll_local_open(file, it, fd, NULL);
644 mutex_unlock(&lli->lli_och_mutex);
645 GOTO(out_openerr, rc);
648 LASSERT(*och_usecount == 0);
649 if (!it->d.lustre.it_disposition) {
650 /* We cannot just request lock handle now, new ELC code
651 means that one of other OPEN locks for this file
652 could be cancelled, and since blocking ast handler
653 would attempt to grab och_mutex as well, that would
654 result in a deadlock */
655 mutex_unlock(&lli->lli_och_mutex);
657 * Normally called under two situations:
659 * 2. A race/condition on MDS resulting in no open
660 * handle to be returned from LOOKUP|OPEN request,
661 * for example if the target entry was a symlink.
663 * Always fetch MDS_OPEN_LOCK if this is not setstripe.
665 * Always specify MDS_OPEN_BY_FID because we don't want
666 * to get file with different fid.
668 it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
669 rc = ll_intent_file_open(file, NULL, 0, it);
671 GOTO(out_openerr, rc);
675 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
677 GOTO(out_och_free, rc = -ENOMEM);
681 /* md_intent_lock() didn't get a request ref if there was an
682 * open error, so don't do cleanup on the request here
684 /* XXX (green): Should not we bail out on any error here, not
685 * just open error? */
686 rc = it_open_error(DISP_OPEN_OPEN, it);
688 GOTO(out_och_free, rc);
690 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
691 "inode %p: disposition %x, status %d\n", inode,
692 it_disposition(it, ~0), it->d.lustre.it_status);
694 rc = ll_local_open(file, it, fd, *och_p);
696 GOTO(out_och_free, rc);
698 mutex_unlock(&lli->lli_och_mutex);
701 /* Must do this outside lli_och_mutex lock to prevent deadlock where
702 different kind of OPEN lock for this same inode gets cancelled
703 by ldlm_cancel_lru */
704 if (!S_ISREG(inode->i_mode))
705 GOTO(out_och_free, rc);
709 if (!lli->lli_has_smd &&
710 (cl_is_lov_delay_create(file->f_flags) ||
711 (file->f_mode & FMODE_WRITE) == 0)) {
712 CDEBUG(D_INODE, "object creation was delayed\n");
713 GOTO(out_och_free, rc);
715 cl_lov_delay_create_clear(&file->f_flags);
716 GOTO(out_och_free, rc);
720 if (och_p && *och_p) {
721 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
722 *och_p = NULL; /* OBD_FREE writes some magic there */
725 mutex_unlock(&lli->lli_och_mutex);
728 if (lli->lli_opendir_key == fd)
729 ll_deauthorize_statahead(inode, fd);
731 ll_file_data_put(fd);
733 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
736 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
737 ptlrpc_req_finished(it->d.lustre.it_data);
738 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
744 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
745 struct ldlm_lock_desc *desc, void *data, int flag)
748 struct lustre_handle lockh;
752 case LDLM_CB_BLOCKING:
753 ldlm_lock2handle(lock, &lockh);
754 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
756 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
760 case LDLM_CB_CANCELING:
768 * Acquire a lease and open the file.
770 static struct obd_client_handle *
771 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
774 struct lookup_intent it = { .it_op = IT_OPEN };
775 struct ll_sb_info *sbi = ll_i2sbi(inode);
776 struct md_op_data *op_data;
777 struct ptlrpc_request *req = NULL;
778 struct lustre_handle old_handle = { 0 };
779 struct obd_client_handle *och = NULL;
784 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
785 RETURN(ERR_PTR(-EINVAL));
788 struct ll_inode_info *lli = ll_i2info(inode);
789 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
790 struct obd_client_handle **och_p;
793 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
794 RETURN(ERR_PTR(-EPERM));
796 /* Get the openhandle of the file */
798 mutex_lock(&lli->lli_och_mutex);
799 if (fd->fd_lease_och != NULL) {
800 mutex_unlock(&lli->lli_och_mutex);
804 if (fd->fd_och == NULL) {
805 if (file->f_mode & FMODE_WRITE) {
806 LASSERT(lli->lli_mds_write_och != NULL);
807 och_p = &lli->lli_mds_write_och;
808 och_usecount = &lli->lli_open_fd_write_count;
810 LASSERT(lli->lli_mds_read_och != NULL);
811 och_p = &lli->lli_mds_read_och;
812 och_usecount = &lli->lli_open_fd_read_count;
814 if (*och_usecount == 1) {
821 mutex_unlock(&lli->lli_och_mutex);
822 if (rc < 0) /* more than 1 opener */
825 LASSERT(fd->fd_och != NULL);
826 old_handle = fd->fd_och->och_fh;
831 RETURN(ERR_PTR(-ENOMEM));
833 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
834 LUSTRE_OPC_ANY, NULL);
836 GOTO(out, rc = PTR_ERR(op_data));
838 /* To tell the MDT this openhandle is from the same owner */
839 op_data->op_handle = old_handle;
841 it.it_flags = fmode | open_flags;
842 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
843 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
844 &ll_md_blocking_lease_ast,
845 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
846 * it can be cancelled which may mislead applications that the lease is
848 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
849 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
850 * doesn't deal with openhandle, so normal openhandle will be leaked. */
851 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
852 ll_finish_md_op_data(op_data);
853 ptlrpc_req_finished(req);
855 GOTO(out_release_it, rc);
857 if (it_disposition(&it, DISP_LOOKUP_NEG))
858 GOTO(out_release_it, rc = -ENOENT);
860 rc = it_open_error(DISP_OPEN_OPEN, &it);
862 GOTO(out_release_it, rc);
864 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
865 ll_och_fill(sbi->ll_md_exp, &it, och);
867 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
868 GOTO(out_close, rc = -EOPNOTSUPP);
870 /* already get lease, handle lease lock */
871 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
872 if (it.d.lustre.it_lock_mode == 0 ||
873 it.d.lustre.it_lock_bits != MDS_INODELOCK_OPEN) {
874 /* open lock must return for lease */
875 CERROR(DFID "lease granted but no open lock, %d/"LPU64".\n",
876 PFID(ll_inode2fid(inode)), it.d.lustre.it_lock_mode,
877 it.d.lustre.it_lock_bits);
878 GOTO(out_close, rc = -EPROTO);
881 ll_intent_release(&it);
885 /* Cancel open lock */
886 if (it.d.lustre.it_lock_mode != 0) {
887 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
888 it.d.lustre.it_lock_mode);
889 it.d.lustre.it_lock_mode = 0;
890 och->och_lease_handle.cookie = 0ULL;
892 rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL);
894 CERROR("%s: error closing file "DFID": %d\n",
895 ll_get_fsname(inode->i_sb, NULL, 0),
896 PFID(&ll_i2info(inode)->lli_fid), rc2);
897 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
899 ll_intent_release(&it);
907 * Release lease and close the file.
908 * It will check if the lease has ever broken.
910 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
913 struct ldlm_lock *lock;
914 bool cancelled = true;
918 lock = ldlm_handle2lock(&och->och_lease_handle);
920 lock_res_and_lock(lock);
921 cancelled = ldlm_is_cancel(lock);
922 unlock_res_and_lock(lock);
926 CDEBUG(D_INODE, "lease for "DFID" broken? %d\n",
927 PFID(&ll_i2info(inode)->lli_fid), cancelled);
930 ldlm_cli_cancel(&och->och_lease_handle, 0);
931 if (lease_broken != NULL)
932 *lease_broken = cancelled;
934 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
939 /* Fills the obdo with the attributes for the lsm */
940 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
941 struct obd_capa *capa, struct obdo *obdo,
942 __u64 ioepoch, int dv_flags)
944 struct ptlrpc_request_set *set;
945 struct obd_info oinfo = { { { 0 } } };
950 LASSERT(lsm != NULL);
954 oinfo.oi_oa->o_oi = lsm->lsm_oi;
955 oinfo.oi_oa->o_mode = S_IFREG;
956 oinfo.oi_oa->o_ioepoch = ioepoch;
957 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
958 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
959 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
960 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
961 OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
962 OBD_MD_FLDATAVERSION;
963 oinfo.oi_capa = capa;
964 if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) {
965 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
966 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
967 if (dv_flags & LL_DV_WR_FLUSH)
968 oinfo.oi_oa->o_flags |= OBD_FL_FLUSH;
971 set = ptlrpc_prep_set();
973 CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM);
976 rc = obd_getattr_async(exp, &oinfo, set);
978 rc = ptlrpc_set_wait(set);
979 ptlrpc_set_destroy(set);
982 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
983 OBD_MD_FLATIME | OBD_MD_FLMTIME |
984 OBD_MD_FLCTIME | OBD_MD_FLSIZE |
985 OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS);
986 if (dv_flags & LL_DV_WR_FLUSH &&
987 !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
988 oinfo.oi_oa->o_flags & OBD_FL_FLUSH))
995 * Performs the getattr on the inode and updates its fields.
996 * If @sync != 0, perform the getattr under the server-side lock.
998 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
999 __u64 ioepoch, int sync)
1001 struct obd_capa *capa = ll_mdscapa_get(inode);
1002 struct lov_stripe_md *lsm;
1006 lsm = ccc_inode_lsm_get(inode);
1007 rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode),
1008 capa, obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0);
1011 struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi;
1013 obdo_refresh_inode(inode, obdo, obdo->o_valid);
1014 CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu,"
1015 " blksize %lu\n", POSTID(oi), i_size_read(inode),
1016 (unsigned long long)inode->i_blocks,
1017 1UL << inode->i_blkbits);
1019 ccc_inode_lsm_put(inode, lsm);
1023 int ll_merge_lvb(const struct lu_env *env, struct inode *inode)
1025 struct ll_inode_info *lli = ll_i2info(inode);
1026 struct cl_object *obj = lli->lli_clob;
1027 struct cl_attr *attr = ccc_env_thread_attr(env);
1033 ll_inode_size_lock(inode);
1034 /* merge timestamps the most recently obtained from mds with
1035 timestamps obtained from osts */
1036 LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime;
1037 LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime;
1038 LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime;
1039 inode_init_lvb(inode, &lvb);
1041 cl_object_attr_lock(obj);
1042 rc = cl_object_attr_get(env, obj, attr);
1043 cl_object_attr_unlock(obj);
1046 if (lvb.lvb_atime < attr->cat_atime)
1047 lvb.lvb_atime = attr->cat_atime;
1048 if (lvb.lvb_ctime < attr->cat_ctime)
1049 lvb.lvb_ctime = attr->cat_ctime;
1050 if (lvb.lvb_mtime < attr->cat_mtime)
1051 lvb.lvb_mtime = attr->cat_mtime;
1053 CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n",
1054 PFID(&lli->lli_fid), attr->cat_size);
1055 cl_isize_write_nolock(inode, attr->cat_size);
1057 inode->i_blocks = attr->cat_blocks;
1059 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
1060 LTIME_S(inode->i_atime) = lvb.lvb_atime;
1061 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
1063 ll_inode_size_unlock(inode);
1068 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
1071 struct obdo obdo = { 0 };
1074 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0);
1076 st->st_size = obdo.o_size;
1077 st->st_blocks = obdo.o_blocks;
1078 st->st_mtime = obdo.o_mtime;
1079 st->st_atime = obdo.o_atime;
1080 st->st_ctime = obdo.o_ctime;
1085 static bool file_is_noatime(const struct file *file)
1087 const struct vfsmount *mnt = file->f_path.mnt;
1088 const struct inode *inode = file->f_path.dentry->d_inode;
1090 /* Adapted from file_accessed() and touch_atime().*/
1091 if (file->f_flags & O_NOATIME)
1094 if (inode->i_flags & S_NOATIME)
1097 if (IS_NOATIME(inode))
1100 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1103 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1106 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1112 static void ll_io_init(struct cl_io *io, const struct file *file, int write)
1114 struct inode *inode = file->f_dentry->d_inode;
1116 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1118 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1119 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1120 file->f_flags & O_DIRECT ||
1123 io->ci_obj = ll_i2info(inode)->lli_clob;
1124 io->ci_lockreq = CILR_MAYBE;
1125 if (ll_file_nolock(file)) {
1126 io->ci_lockreq = CILR_NEVER;
1127 io->ci_no_srvlock = 1;
1128 } else if (file->f_flags & O_APPEND) {
1129 io->ci_lockreq = CILR_MANDATORY;
1132 io->ci_noatime = file_is_noatime(file);
1136 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1137 struct file *file, enum cl_io_type iot,
1138 loff_t *ppos, size_t count)
1140 struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
1141 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1144 struct range_lock range;
1147 CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: "LPU64", count: %zd\n",
1148 file->f_dentry->d_name.name, iot, *ppos, count);
1151 io = ccc_env_thread_io(env);
1152 ll_io_init(io, file, iot == CIT_WRITE);
1154 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1155 struct vvp_io *vio = vvp_env_io(env);
1156 struct ccc_io *cio = ccc_env_io(env);
1157 bool range_locked = false;
1159 if (file->f_flags & O_APPEND)
1160 range_lock_init(&range, 0, LUSTRE_EOF);
1162 range_lock_init(&range, *ppos, *ppos + count - 1);
1163 cio->cui_fd = LUSTRE_FPRIVATE(file);
1164 vio->cui_io_subtype = args->via_io_subtype;
1166 switch (vio->cui_io_subtype) {
1168 cio->cui_iov = args->u.normal.via_iov;
1169 cio->cui_nrsegs = args->u.normal.via_nrsegs;
1170 cio->cui_tot_nrsegs = cio->cui_nrsegs;
1171 cio->cui_iocb = args->u.normal.via_iocb;
1172 if ((iot == CIT_WRITE) &&
1173 !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1174 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1176 result = range_lock(&lli->lli_write_tree,
1181 range_locked = true;
1183 down_read(&lli->lli_trunc_sem);
1186 vio->u.splice.cui_pipe = args->u.splice.via_pipe;
1187 vio->u.splice.cui_flags = args->u.splice.via_flags;
1190 CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
1194 ll_cl_add(file, env, io);
1195 result = cl_io_loop(env, io);
1196 ll_cl_remove(file, env);
1198 if (args->via_io_subtype == IO_NORMAL)
1199 up_read(&lli->lli_trunc_sem);
1201 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1203 range_unlock(&lli->lli_write_tree, &range);
1206 /* cl_io_rw_init() handled IO */
1207 result = io->ci_result;
1210 if (io->ci_nob > 0) {
1211 result = io->ci_nob;
1212 *ppos = io->u.ci_wr.wr.crw_pos;
1216 cl_io_fini(env, io);
1217 /* If any bit been read/written (result != 0), we just return
1218 * short read/write instead of restart io. */
1219 if ((result == 0 || result == -ENODATA) && io->ci_need_restart) {
1220 CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n",
1221 iot == CIT_READ ? "read" : "write",
1222 file->f_dentry->d_name.name, *ppos, count);
1223 LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob);
1227 if (iot == CIT_READ) {
1229 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
1230 LPROC_LL_READ_BYTES, result);
1231 } else if (iot == CIT_WRITE) {
1233 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
1234 LPROC_LL_WRITE_BYTES, result);
1235 fd->fd_write_failed = false;
1236 } else if (result != -ERESTARTSYS) {
1237 fd->fd_write_failed = true;
1240 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1247 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1249 static int ll_file_get_iov_count(const struct iovec *iov,
1250 unsigned long *nr_segs, size_t *count)
1255 for (seg = 0; seg < *nr_segs; seg++) {
1256 const struct iovec *iv = &iov[seg];
1259 * If any segment has a negative length, or the cumulative
1260 * length ever wraps negative then return -EINVAL.
1263 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1265 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1270 cnt -= iv->iov_len; /* This segment is no good */
1277 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1278 unsigned long nr_segs, loff_t pos)
1281 struct vvp_io_args *args;
1287 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1291 env = cl_env_get(&refcheck);
1293 RETURN(PTR_ERR(env));
1295 args = vvp_env_args(env, IO_NORMAL);
1296 args->u.normal.via_iov = (struct iovec *)iov;
1297 args->u.normal.via_nrsegs = nr_segs;
1298 args->u.normal.via_iocb = iocb;
1300 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1301 &iocb->ki_pos, count);
1302 cl_env_put(env, &refcheck);
1306 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1310 struct iovec *local_iov;
1311 struct kiocb *kiocb;
1316 env = cl_env_get(&refcheck);
1318 RETURN(PTR_ERR(env));
1320 local_iov = &vvp_env_info(env)->vti_local_iov;
1321 kiocb = &vvp_env_info(env)->vti_kiocb;
1322 local_iov->iov_base = (void __user *)buf;
1323 local_iov->iov_len = count;
1324 init_sync_kiocb(kiocb, file);
1325 kiocb->ki_pos = *ppos;
1326 #ifdef HAVE_KIOCB_KI_LEFT
1327 kiocb->ki_left = count;
1329 kiocb->ki_nbytes = count;
1332 result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos);
1333 *ppos = kiocb->ki_pos;
1335 cl_env_put(env, &refcheck);
1340 * Write to a file (through the page cache).
1343 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1344 unsigned long nr_segs, loff_t pos)
1347 struct vvp_io_args *args;
1353 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1357 env = cl_env_get(&refcheck);
1359 RETURN(PTR_ERR(env));
1361 args = vvp_env_args(env, IO_NORMAL);
1362 args->u.normal.via_iov = (struct iovec *)iov;
1363 args->u.normal.via_nrsegs = nr_segs;
1364 args->u.normal.via_iocb = iocb;
1366 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1367 &iocb->ki_pos, count);
1368 cl_env_put(env, &refcheck);
1372 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1376 struct iovec *local_iov;
1377 struct kiocb *kiocb;
1382 env = cl_env_get(&refcheck);
1384 RETURN(PTR_ERR(env));
1386 local_iov = &vvp_env_info(env)->vti_local_iov;
1387 kiocb = &vvp_env_info(env)->vti_kiocb;
1388 local_iov->iov_base = (void __user *)buf;
1389 local_iov->iov_len = count;
1390 init_sync_kiocb(kiocb, file);
1391 kiocb->ki_pos = *ppos;
1392 #ifdef HAVE_KIOCB_KI_LEFT
1393 kiocb->ki_left = count;
1395 kiocb->ki_nbytes = count;
1398 result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos);
1399 *ppos = kiocb->ki_pos;
1401 cl_env_put(env, &refcheck);
1406 * Send file content (through pagecache) somewhere with helper
1408 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1409 struct pipe_inode_info *pipe, size_t count,
1413 struct vvp_io_args *args;
1418 env = cl_env_get(&refcheck);
1420 RETURN(PTR_ERR(env));
1422 args = vvp_env_args(env, IO_SPLICE);
1423 args->u.splice.via_pipe = pipe;
1424 args->u.splice.via_flags = flags;
1426 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1427 cl_env_put(env, &refcheck);
1431 static int ll_lov_recreate(struct inode *inode, struct ost_id *oi,
1434 struct obd_export *exp = ll_i2dtexp(inode);
1435 struct obd_trans_info oti = { 0 };
1436 struct obdo *oa = NULL;
1439 struct lov_stripe_md *lsm = NULL, *lsm2;
1446 lsm = ccc_inode_lsm_get(inode);
1447 if (!lsm_has_objects(lsm))
1448 GOTO(out, rc = -ENOENT);
1450 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1451 (lsm->lsm_stripe_count));
1453 OBD_ALLOC_LARGE(lsm2, lsm_size);
1455 GOTO(out, rc = -ENOMEM);
1458 oa->o_nlink = ost_idx;
1459 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1460 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1461 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1462 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1463 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1464 memcpy(lsm2, lsm, lsm_size);
1465 ll_inode_size_lock(inode);
1466 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1467 ll_inode_size_unlock(inode);
1469 OBD_FREE_LARGE(lsm2, lsm_size);
1472 ccc_inode_lsm_put(inode, lsm);
1477 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1479 struct ll_recreate_obj ucreat;
1483 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1486 if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
1490 ostid_set_seq_mdt0(&oi);
1491 ostid_set_id(&oi, ucreat.lrc_id);
1492 RETURN(ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx));
1495 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1502 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1505 if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid)))
1508 fid_to_ostid(&fid, &oi);
1509 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1510 RETURN(ll_lov_recreate(inode, &oi, ost_idx));
1513 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1514 __u64 flags, struct lov_user_md *lum,
1517 struct lov_stripe_md *lsm = NULL;
1518 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1522 lsm = ccc_inode_lsm_get(inode);
1524 ccc_inode_lsm_put(inode, lsm);
1525 CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n",
1526 PFID(ll_inode2fid(inode)));
1527 GOTO(out, rc = -EEXIST);
1530 ll_inode_size_lock(inode);
1531 oit.it_flags |= MDS_OPEN_BY_FID;
1532 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1534 GOTO(out_unlock, rc);
1535 rc = oit.d.lustre.it_status;
1537 GOTO(out_req_free, rc);
1539 ll_release_openhandle(file->f_dentry, &oit);
1542 ll_inode_size_unlock(inode);
1543 ll_intent_release(&oit);
1544 ccc_inode_lsm_put(inode, lsm);
1546 cl_lov_delay_create_clear(&file->f_flags);
1549 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1553 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1554 struct lov_mds_md **lmmp, int *lmm_size,
1555 struct ptlrpc_request **request)
1557 struct ll_sb_info *sbi = ll_i2sbi(inode);
1558 struct mdt_body *body;
1559 struct lov_mds_md *lmm = NULL;
1560 struct ptlrpc_request *req = NULL;
1561 struct md_op_data *op_data;
1564 rc = ll_get_default_mdsize(sbi, &lmmsize);
1568 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1569 strlen(filename), lmmsize,
1570 LUSTRE_OPC_ANY, NULL);
1571 if (IS_ERR(op_data))
1572 RETURN(PTR_ERR(op_data));
1574 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1575 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1576 ll_finish_md_op_data(op_data);
1578 CDEBUG(D_INFO, "md_getattr_name failed "
1579 "on %s: rc %d\n", filename, rc);
1583 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1584 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1586 lmmsize = body->mbo_eadatasize;
1588 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1590 GOTO(out, rc = -ENODATA);
1593 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1594 LASSERT(lmm != NULL);
1596 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1597 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1598 GOTO(out, rc = -EPROTO);
1602 * This is coming from the MDS, so is probably in
1603 * little endian. We convert it to host endian before
1604 * passing it to userspace.
1606 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1609 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1610 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1613 /* if function called for directory - we should
1614 * avoid swab not existent lsm objects */
1615 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1616 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1617 if (S_ISREG(body->mbo_mode))
1618 lustre_swab_lov_user_md_objects(
1619 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1621 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1622 lustre_swab_lov_user_md_v3(
1623 (struct lov_user_md_v3 *)lmm);
1624 if (S_ISREG(body->mbo_mode))
1625 lustre_swab_lov_user_md_objects(
1626 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1633 *lmm_size = lmmsize;
1638 static int ll_lov_setea(struct inode *inode, struct file *file,
1641 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1642 struct lov_user_md *lump;
1643 int lum_size = sizeof(struct lov_user_md) +
1644 sizeof(struct lov_user_ost_data);
1648 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1651 OBD_ALLOC_LARGE(lump, lum_size);
1655 if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) {
1656 OBD_FREE_LARGE(lump, lum_size);
1660 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1662 OBD_FREE_LARGE(lump, lum_size);
1666 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1669 struct lov_user_md_v3 lumv3;
1670 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1671 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1672 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1674 __u64 flags = FMODE_WRITE;
1677 /* first try with v1 which is smaller than v3 */
1678 lum_size = sizeof(struct lov_user_md_v1);
1679 if (copy_from_user(lumv1, lumv1p, lum_size))
1682 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1683 lum_size = sizeof(struct lov_user_md_v3);
1684 if (copy_from_user(&lumv3, lumv3p, lum_size))
1688 rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size);
1690 struct lov_stripe_md *lsm;
1693 put_user(0, &lumv1p->lmm_stripe_count);
1695 ll_layout_refresh(inode, &gen);
1696 lsm = ccc_inode_lsm_get(inode);
1697 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1698 0, lsm, (void *)arg);
1699 ccc_inode_lsm_put(inode, lsm);
1704 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1706 struct lov_stripe_md *lsm;
1710 lsm = ccc_inode_lsm_get(inode);
1712 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1714 ccc_inode_lsm_put(inode, lsm);
1719 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1721 struct ll_inode_info *lli = ll_i2info(inode);
1722 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1723 struct ccc_grouplock grouplock;
1727 if (ll_file_nolock(file))
1728 RETURN(-EOPNOTSUPP);
1730 spin_lock(&lli->lli_lock);
1731 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1732 CWARN("group lock already existed with gid %lu\n",
1733 fd->fd_grouplock.cg_gid);
1734 spin_unlock(&lli->lli_lock);
1737 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1738 spin_unlock(&lli->lli_lock);
1740 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1741 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1745 spin_lock(&lli->lli_lock);
1746 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1747 spin_unlock(&lli->lli_lock);
1748 CERROR("another thread just won the race\n");
1749 cl_put_grouplock(&grouplock);
1753 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1754 fd->fd_grouplock = grouplock;
1755 spin_unlock(&lli->lli_lock);
1757 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1761 int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1763 struct ll_inode_info *lli = ll_i2info(inode);
1764 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1765 struct ccc_grouplock grouplock;
1768 spin_lock(&lli->lli_lock);
1769 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1770 spin_unlock(&lli->lli_lock);
1771 CWARN("no group lock held\n");
1774 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1776 if (fd->fd_grouplock.cg_gid != arg) {
1777 CWARN("group lock %lu doesn't match current id %lu\n",
1778 arg, fd->fd_grouplock.cg_gid);
1779 spin_unlock(&lli->lli_lock);
1783 grouplock = fd->fd_grouplock;
1784 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1785 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1786 spin_unlock(&lli->lli_lock);
1788 cl_put_grouplock(&grouplock);
1789 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1794 * Close inode open handle
1796 * \param dentry [in] dentry which contains the inode
1797 * \param it [in,out] intent which contains open info and result
1800 * \retval <0 failure
1802 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1804 struct inode *inode = dentry->d_inode;
1805 struct obd_client_handle *och;
1811 /* Root ? Do nothing. */
1812 if (dentry->d_inode->i_sb->s_root == dentry)
1815 /* No open handle to close? Move away */
1816 if (!it_disposition(it, DISP_OPEN_OPEN))
1819 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1821 OBD_ALLOC(och, sizeof(*och));
1823 GOTO(out, rc = -ENOMEM);
1825 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1827 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1830 /* this one is in place of ll_file_open */
1831 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1832 ptlrpc_req_finished(it->d.lustre.it_data);
1833 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1839 * Get size for inode for which FIEMAP mapping is requested.
1840 * Make the FIEMAP get_info call and returns the result.
1842 static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1845 struct obd_export *exp = ll_i2dtexp(inode);
1846 struct lov_stripe_md *lsm = NULL;
1847 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1848 __u32 vallen = num_bytes;
1852 /* Checks for fiemap flags */
1853 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1854 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1858 /* Check for FIEMAP_FLAG_SYNC */
1859 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1860 rc = filemap_fdatawrite(inode->i_mapping);
1865 lsm = ccc_inode_lsm_get(inode);
1869 /* If the stripe_count > 1 and the application does not understand
1870 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1872 if (lsm->lsm_stripe_count > 1 &&
1873 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER))
1874 GOTO(out, rc = -EOPNOTSUPP);
1876 fm_key.oa.o_oi = lsm->lsm_oi;
1877 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1879 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1880 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1881 /* If filesize is 0, then there would be no objects for mapping */
1882 if (fm_key.oa.o_size == 0) {
1883 fiemap->fm_mapped_extents = 0;
1887 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1889 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1892 CERROR("obd_get_info failed: rc = %d\n", rc);
1895 ccc_inode_lsm_put(inode, lsm);
1899 int ll_fid2path(struct inode *inode, void __user *arg)
1901 struct obd_export *exp = ll_i2mdexp(inode);
1902 const struct getinfo_fid2path __user *gfin = arg;
1904 struct getinfo_fid2path *gfout;
1910 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1911 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1914 /* Only need to get the buflen */
1915 if (get_user(pathlen, &gfin->gf_pathlen))
1918 if (pathlen > PATH_MAX)
1921 outsize = sizeof(*gfout) + pathlen;
1922 OBD_ALLOC(gfout, outsize);
1926 if (copy_from_user(gfout, arg, sizeof(*gfout)))
1927 GOTO(gf_free, rc = -EFAULT);
1929 /* Call mdc_iocontrol */
1930 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1934 if (copy_to_user(arg, gfout, outsize))
1938 OBD_FREE(gfout, outsize);
1942 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1944 struct ll_user_fiemap *fiemap_s;
1945 size_t num_bytes, ret_bytes;
1946 unsigned int extent_count;
1949 /* Get the extent count so we can calculate the size of
1950 * required fiemap buffer */
1951 if (get_user(extent_count,
1952 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1956 (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent))
1958 num_bytes = sizeof(*fiemap_s) + (extent_count *
1959 sizeof(struct ll_fiemap_extent));
1961 OBD_ALLOC_LARGE(fiemap_s, num_bytes);
1962 if (fiemap_s == NULL)
1965 /* get the fiemap value */
1966 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1968 GOTO(error, rc = -EFAULT);
1970 /* If fm_extent_count is non-zero, read the first extent since
1971 * it is used to calculate end_offset and device from previous
1974 if (copy_from_user(&fiemap_s->fm_extents[0],
1975 (char __user *)arg + sizeof(*fiemap_s),
1976 sizeof(struct ll_fiemap_extent)))
1977 GOTO(error, rc = -EFAULT);
1980 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1984 ret_bytes = sizeof(struct ll_user_fiemap);
1986 if (extent_count != 0)
1987 ret_bytes += (fiemap_s->fm_mapped_extents *
1988 sizeof(struct ll_fiemap_extent));
1990 if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
1994 OBD_FREE_LARGE(fiemap_s, num_bytes);
1999 * Read the data_version for inode.
2001 * This value is computed using stripe object version on OST.
2002 * Version is computed using server side locking.
2004 * @param sync if do sync on the OST side;
2006 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2007 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2009 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2011 struct lov_stripe_md *lsm = NULL;
2012 struct ll_sb_info *sbi = ll_i2sbi(inode);
2013 struct obdo *obdo = NULL;
2017 /* If no stripe, we consider version is 0. */
2018 lsm = ccc_inode_lsm_get(inode);
2019 if (!lsm_has_objects(lsm)) {
2021 CDEBUG(D_INODE, "No object for inode\n");
2025 OBD_ALLOC_PTR(obdo);
2027 GOTO(out, rc = -ENOMEM);
2029 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, flags);
2031 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
2034 *data_version = obdo->o_data_version;
2040 ccc_inode_lsm_put(inode, lsm);
2045 * Trigger a HSM release request for the provided inode.
2047 int ll_hsm_release(struct inode *inode)
2049 struct cl_env_nest nest;
2051 struct obd_client_handle *och = NULL;
2052 __u64 data_version = 0;
2056 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2057 ll_get_fsname(inode->i_sb, NULL, 0),
2058 PFID(&ll_i2info(inode)->lli_fid));
2060 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2062 GOTO(out, rc = PTR_ERR(och));
2064 /* Grab latest data_version and [am]time values */
2065 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2069 env = cl_env_nested_get(&nest);
2071 GOTO(out, rc = PTR_ERR(env));
2073 ll_merge_lvb(env, inode);
2074 cl_env_nested_put(&nest, env);
2076 /* Release the file.
2077 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2078 * we still need it to pack l_remote_handle to MDT. */
2079 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
2085 if (och != NULL && !IS_ERR(och)) /* close the file */
2086 ll_lease_close(och, inode, NULL);
2091 struct ll_swap_stack {
2092 struct iattr ia1, ia2;
2094 struct inode *inode1, *inode2;
2095 bool check_dv1, check_dv2;
2098 static int ll_swap_layouts(struct file *file1, struct file *file2,
2099 struct lustre_swap_layouts *lsl)
2101 struct mdc_swap_layouts msl;
2102 struct md_op_data *op_data;
2105 struct ll_swap_stack *llss = NULL;
2108 OBD_ALLOC_PTR(llss);
2112 llss->inode1 = file1->f_dentry->d_inode;
2113 llss->inode2 = file2->f_dentry->d_inode;
2115 if (!S_ISREG(llss->inode2->i_mode))
2116 GOTO(free, rc = -EINVAL);
2118 if (inode_permission(llss->inode1, MAY_WRITE) ||
2119 inode_permission(llss->inode2, MAY_WRITE))
2120 GOTO(free, rc = -EPERM);
2122 if (llss->inode2->i_sb != llss->inode1->i_sb)
2123 GOTO(free, rc = -EXDEV);
2125 /* we use 2 bool because it is easier to swap than 2 bits */
2126 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2127 llss->check_dv1 = true;
2129 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2130 llss->check_dv2 = true;
2132 /* we cannot use lsl->sl_dvX directly because we may swap them */
2133 llss->dv1 = lsl->sl_dv1;
2134 llss->dv2 = lsl->sl_dv2;
2136 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2137 if (rc == 0) /* same file, done! */
2140 if (rc < 0) { /* sequentialize it */
2141 swap(llss->inode1, llss->inode2);
2143 swap(llss->dv1, llss->dv2);
2144 swap(llss->check_dv1, llss->check_dv2);
2148 if (gid != 0) { /* application asks to flush dirty cache */
2149 rc = ll_get_grouplock(llss->inode1, file1, gid);
2153 rc = ll_get_grouplock(llss->inode2, file2, gid);
2155 ll_put_grouplock(llss->inode1, file1, gid);
2160 /* to be able to restore mtime and atime after swap
2161 * we need to first save them */
2163 (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
2164 llss->ia1.ia_mtime = llss->inode1->i_mtime;
2165 llss->ia1.ia_atime = llss->inode1->i_atime;
2166 llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
2167 llss->ia2.ia_mtime = llss->inode2->i_mtime;
2168 llss->ia2.ia_atime = llss->inode2->i_atime;
2169 llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
2172 /* ultimate check, before swaping the layouts we check if
2173 * dataversion has changed (if requested) */
2174 if (llss->check_dv1) {
2175 rc = ll_data_version(llss->inode1, &dv, 0);
2178 if (dv != llss->dv1)
2179 GOTO(putgl, rc = -EAGAIN);
2182 if (llss->check_dv2) {
2183 rc = ll_data_version(llss->inode2, &dv, 0);
2186 if (dv != llss->dv2)
2187 GOTO(putgl, rc = -EAGAIN);
2190 /* struct md_op_data is used to send the swap args to the mdt
2191 * only flags is missing, so we use struct mdc_swap_layouts
2192 * through the md_op_data->op_data */
2193 /* flags from user space have to be converted before they are send to
2194 * server, no flag is sent today, they are only used on the client */
2197 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2198 0, LUSTRE_OPC_ANY, &msl);
2199 if (IS_ERR(op_data))
2200 GOTO(free, rc = PTR_ERR(op_data));
2202 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2203 sizeof(*op_data), op_data, NULL);
2204 ll_finish_md_op_data(op_data);
2208 ll_put_grouplock(llss->inode2, file2, gid);
2209 ll_put_grouplock(llss->inode1, file1, gid);
2212 /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
2216 /* clear useless flags */
2217 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
2218 llss->ia1.ia_valid &= ~ATTR_MTIME;
2219 llss->ia2.ia_valid &= ~ATTR_MTIME;
2222 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
2223 llss->ia1.ia_valid &= ~ATTR_ATIME;
2224 llss->ia2.ia_valid &= ~ATTR_ATIME;
2227 /* update time if requested */
2229 if (llss->ia2.ia_valid != 0) {
2230 mutex_lock(&llss->inode1->i_mutex);
2231 rc = ll_setattr(file1->f_dentry, &llss->ia2);
2232 mutex_unlock(&llss->inode1->i_mutex);
2235 if (llss->ia1.ia_valid != 0) {
2238 mutex_lock(&llss->inode2->i_mutex);
2239 rc1 = ll_setattr(file2->f_dentry, &llss->ia1);
2240 mutex_unlock(&llss->inode2->i_mutex);
2252 static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2254 struct md_op_data *op_data;
2257 /* Non-root users are forbidden to set or clear flags which are
2258 * NOT defined in HSM_USER_MASK. */
2259 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2260 !cfs_capable(CFS_CAP_SYS_ADMIN))
2263 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2264 LUSTRE_OPC_ANY, hss);
2265 if (IS_ERR(op_data))
2266 RETURN(PTR_ERR(op_data));
2268 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2269 sizeof(*op_data), op_data, NULL);
2271 ll_finish_md_op_data(op_data);
2276 static int ll_hsm_import(struct inode *inode, struct file *file,
2277 struct hsm_user_import *hui)
2279 struct hsm_state_set *hss = NULL;
2280 struct iattr *attr = NULL;
2284 if (!S_ISREG(inode->i_mode))
2290 GOTO(out, rc = -ENOMEM);
2292 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2293 hss->hss_archive_id = hui->hui_archive_id;
2294 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2295 rc = ll_hsm_state_set(inode, hss);
2299 OBD_ALLOC_PTR(attr);
2301 GOTO(out, rc = -ENOMEM);
2303 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2304 attr->ia_mode |= S_IFREG;
2305 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2306 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2307 attr->ia_size = hui->hui_size;
2308 attr->ia_mtime.tv_sec = hui->hui_mtime;
2309 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2310 attr->ia_atime.tv_sec = hui->hui_atime;
2311 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2313 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2314 ATTR_UID | ATTR_GID |
2315 ATTR_MTIME | ATTR_MTIME_SET |
2316 ATTR_ATIME | ATTR_ATIME_SET;
2318 mutex_lock(&inode->i_mutex);
2320 rc = ll_setattr_raw(file->f_dentry, attr, true);
2324 mutex_unlock(&inode->i_mutex);
2336 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2338 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2339 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2343 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2345 struct inode *inode = file->f_dentry->d_inode;
2346 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2350 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
2351 PFID(ll_inode2fid(inode)), inode, cmd);
2352 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2354 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2355 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2359 case LL_IOC_GETFLAGS:
2360 /* Get the current value of the file flags */
2361 return put_user(fd->fd_flags, (int *)arg);
2362 case LL_IOC_SETFLAGS:
2363 case LL_IOC_CLRFLAGS:
2364 /* Set or clear specific file flags */
2365 /* XXX This probably needs checks to ensure the flags are
2366 * not abused, and to handle any flag side effects.
2368 if (get_user(flags, (int *) arg))
2371 if (cmd == LL_IOC_SETFLAGS) {
2372 if ((flags & LL_FILE_IGNORE_LOCK) &&
2373 !(file->f_flags & O_DIRECT)) {
2374 CERROR("%s: unable to disable locking on "
2375 "non-O_DIRECT file\n", current->comm);
2379 fd->fd_flags |= flags;
2381 fd->fd_flags &= ~flags;
2384 case LL_IOC_LOV_SETSTRIPE:
2385 RETURN(ll_lov_setstripe(inode, file, arg));
2386 case LL_IOC_LOV_SETEA:
2387 RETURN(ll_lov_setea(inode, file, arg));
2388 case LL_IOC_LOV_SWAP_LAYOUTS: {
2390 struct lustre_swap_layouts lsl;
2392 if (copy_from_user(&lsl, (char *)arg,
2393 sizeof(struct lustre_swap_layouts)))
2396 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
2399 file2 = fget(lsl.sl_fd);
2404 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
2405 rc = ll_swap_layouts(file, file2, &lsl);
2409 case LL_IOC_LOV_GETSTRIPE:
2410 RETURN(ll_lov_getstripe(inode, arg));
2411 case LL_IOC_RECREATE_OBJ:
2412 RETURN(ll_lov_recreate_obj(inode, arg));
2413 case LL_IOC_RECREATE_FID:
2414 RETURN(ll_lov_recreate_fid(inode, arg));
2415 case FSFILT_IOC_FIEMAP:
2416 RETURN(ll_ioctl_fiemap(inode, arg));
2417 case FSFILT_IOC_GETFLAGS:
2418 case FSFILT_IOC_SETFLAGS:
2419 RETURN(ll_iocontrol(inode, file, cmd, arg));
2420 case FSFILT_IOC_GETVERSION_OLD:
2421 case FSFILT_IOC_GETVERSION:
2422 RETURN(put_user(inode->i_generation, (int *)arg));
2423 case LL_IOC_GROUP_LOCK:
2424 RETURN(ll_get_grouplock(inode, file, arg));
2425 case LL_IOC_GROUP_UNLOCK:
2426 RETURN(ll_put_grouplock(inode, file, arg));
2427 case IOC_OBD_STATFS:
2428 RETURN(ll_obd_statfs(inode, (void *)arg));
2430 /* We need to special case any other ioctls we want to handle,
2431 * to send them to the MDS/OST as appropriate and to properly
2432 * network encode the arg field.
2433 case FSFILT_IOC_SETVERSION_OLD:
2434 case FSFILT_IOC_SETVERSION:
2436 case LL_IOC_FLUSHCTX:
2437 RETURN(ll_flush_ctx(inode));
2438 case LL_IOC_PATH2FID: {
2439 if (copy_to_user((void *)arg, ll_inode2fid(inode),
2440 sizeof(struct lu_fid)))
2445 case OBD_IOC_FID2PATH:
2446 RETURN(ll_fid2path(inode, (void *)arg));
2447 case LL_IOC_DATA_VERSION: {
2448 struct ioc_data_version idv;
2451 if (copy_from_user(&idv, (char *)arg, sizeof(idv)))
2454 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2455 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2457 if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv)))
2463 case LL_IOC_GET_MDTIDX: {
2466 mdtidx = ll_get_mdt_idx(inode);
2470 if (put_user((int)mdtidx, (int*)arg))
2475 case OBD_IOC_GETDTNAME:
2476 case OBD_IOC_GETMDNAME:
2477 RETURN(ll_get_obd_name(inode, cmd, arg));
2478 case LL_IOC_HSM_STATE_GET: {
2479 struct md_op_data *op_data;
2480 struct hsm_user_state *hus;
2487 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2488 LUSTRE_OPC_ANY, hus);
2489 if (IS_ERR(op_data)) {
2491 RETURN(PTR_ERR(op_data));
2494 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2497 if (copy_to_user((void *)arg, hus, sizeof(*hus)))
2500 ll_finish_md_op_data(op_data);
2504 case LL_IOC_HSM_STATE_SET: {
2505 struct hsm_state_set *hss;
2512 if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
2517 rc = ll_hsm_state_set(inode, hss);
2522 case LL_IOC_HSM_ACTION: {
2523 struct md_op_data *op_data;
2524 struct hsm_current_action *hca;
2531 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2532 LUSTRE_OPC_ANY, hca);
2533 if (IS_ERR(op_data)) {
2535 RETURN(PTR_ERR(op_data));
2538 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2541 if (copy_to_user((char *)arg, hca, sizeof(*hca)))
2544 ll_finish_md_op_data(op_data);
2548 case LL_IOC_SET_LEASE: {
2549 struct ll_inode_info *lli = ll_i2info(inode);
2550 struct obd_client_handle *och = NULL;
2555 case LL_LEASE_WRLCK:
2556 if (!(file->f_mode & FMODE_WRITE))
2558 fmode = FMODE_WRITE;
2560 case LL_LEASE_RDLCK:
2561 if (!(file->f_mode & FMODE_READ))
2565 case LL_LEASE_UNLCK:
2566 mutex_lock(&lli->lli_och_mutex);
2567 if (fd->fd_lease_och != NULL) {
2568 och = fd->fd_lease_och;
2569 fd->fd_lease_och = NULL;
2571 mutex_unlock(&lli->lli_och_mutex);
2576 fmode = och->och_flags;
2577 rc = ll_lease_close(och, inode, &lease_broken);
2584 RETURN(ll_lease_type_from_fmode(fmode));
2589 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
2591 /* apply for lease */
2592 och = ll_lease_open(inode, file, fmode, 0);
2594 RETURN(PTR_ERR(och));
2597 mutex_lock(&lli->lli_och_mutex);
2598 if (fd->fd_lease_och == NULL) {
2599 fd->fd_lease_och = och;
2602 mutex_unlock(&lli->lli_och_mutex);
2604 /* impossible now that only excl is supported for now */
2605 ll_lease_close(och, inode, &lease_broken);
2610 case LL_IOC_GET_LEASE: {
2611 struct ll_inode_info *lli = ll_i2info(inode);
2612 struct ldlm_lock *lock = NULL;
2615 mutex_lock(&lli->lli_och_mutex);
2616 if (fd->fd_lease_och != NULL) {
2617 struct obd_client_handle *och = fd->fd_lease_och;
2619 lock = ldlm_handle2lock(&och->och_lease_handle);
2621 lock_res_and_lock(lock);
2622 if (!ldlm_is_cancel(lock))
2623 fmode = och->och_flags;
2625 unlock_res_and_lock(lock);
2626 LDLM_LOCK_PUT(lock);
2629 mutex_unlock(&lli->lli_och_mutex);
2631 RETURN(ll_lease_type_from_fmode(fmode));
2633 case LL_IOC_HSM_IMPORT: {
2634 struct hsm_user_import *hui;
2640 if (copy_from_user(hui, (void *)arg, sizeof(*hui))) {
2645 rc = ll_hsm_import(inode, file, hui);
2655 ll_iocontrol_call(inode, file, cmd, arg, &err))
2658 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2664 #ifndef HAVE_FILE_LLSEEK_SIZE
2665 static inline loff_t
2666 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2668 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2670 if (offset > maxsize)
2673 if (offset != file->f_pos) {
2674 file->f_pos = offset;
2675 file->f_version = 0;
2681 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2682 loff_t maxsize, loff_t eof)
2684 struct inode *inode = file->f_dentry->d_inode;
2692 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2693 * position-querying operation. Avoid rewriting the "same"
2694 * f_pos value back to the file because a concurrent read(),
2695 * write() or lseek() might have altered it
2700 * f_lock protects against read/modify/write race with other
2701 * SEEK_CURs. Note that parallel writes and reads behave
2704 mutex_lock(&inode->i_mutex);
2705 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2706 mutex_unlock(&inode->i_mutex);
2710 * In the generic case the entire file is data, so as long as
2711 * offset isn't at the end of the file then the offset is data.
2718 * There is a virtual hole at the end of the file, so as long as
2719 * offset isn't i_size or larger, return i_size.
2727 return llseek_execute(file, offset, maxsize);
2731 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2733 struct inode *inode = file->f_dentry->d_inode;
2734 loff_t retval, eof = 0;
2737 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2738 (origin == SEEK_CUR) ? file->f_pos : 0);
2739 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2740 PFID(ll_inode2fid(inode)), inode, retval, retval,
2742 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2744 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2745 retval = ll_glimpse_size(inode);
2748 eof = i_size_read(inode);
2751 retval = ll_generic_file_llseek_size(file, offset, origin,
2752 ll_file_maxbytes(inode), eof);
2756 static int ll_flush(struct file *file, fl_owner_t id)
2758 struct inode *inode = file->f_dentry->d_inode;
2759 struct ll_inode_info *lli = ll_i2info(inode);
2760 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2763 LASSERT(!S_ISDIR(inode->i_mode));
2765 /* catch async errors that were recorded back when async writeback
2766 * failed for pages in this mapping. */
2767 rc = lli->lli_async_rc;
2768 lli->lli_async_rc = 0;
2769 if (lli->lli_clob != NULL) {
2770 err = lov_read_and_clear_async_rc(lli->lli_clob);
2775 /* The application has been told write failure already.
2776 * Do not report failure again. */
2777 if (fd->fd_write_failed)
2779 return rc ? -EIO : 0;
2783 * Called to make sure a portion of file has been written out.
2784 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2786 * Return how many pages have been written.
2788 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2789 enum cl_fsync_mode mode, int ignore_layout)
2791 struct cl_env_nest nest;
2794 struct obd_capa *capa = NULL;
2795 struct cl_fsync_io *fio;
2799 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2800 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2803 env = cl_env_nested_get(&nest);
2805 RETURN(PTR_ERR(env));
2807 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
2809 io = ccc_env_thread_io(env);
2810 io->ci_obj = cl_i2info(inode)->lli_clob;
2811 io->ci_ignore_layout = ignore_layout;
2813 /* initialize parameters for sync */
2814 fio = &io->u.ci_fsync;
2815 fio->fi_capa = capa;
2816 fio->fi_start = start;
2818 fio->fi_fid = ll_inode2fid(inode);
2819 fio->fi_mode = mode;
2820 fio->fi_nr_written = 0;
2822 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2823 result = cl_io_loop(env, io);
2825 result = io->ci_result;
2827 result = fio->fi_nr_written;
2828 cl_io_fini(env, io);
2829 cl_env_nested_put(&nest, env);
2837 * When dentry is provided (the 'else' case), *file->f_dentry may be
2838 * null and dentry must be used directly rather than pulled from
2839 * *file->f_dentry as is done otherwise.
2842 #ifdef HAVE_FILE_FSYNC_4ARGS
2843 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2845 struct dentry *dentry = file->f_dentry;
2846 #elif defined(HAVE_FILE_FSYNC_2ARGS)
2847 int ll_fsync(struct file *file, int datasync)
2849 struct dentry *dentry = file->f_dentry;
2851 loff_t end = LLONG_MAX;
2853 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
2856 loff_t end = LLONG_MAX;
2858 struct inode *inode = dentry->d_inode;
2859 struct ll_inode_info *lli = ll_i2info(inode);
2860 struct ptlrpc_request *req;
2861 struct obd_capa *oc;
2865 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2866 PFID(ll_inode2fid(inode)), inode);
2867 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2869 #ifdef HAVE_FILE_FSYNC_4ARGS
2870 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2871 mutex_lock(&inode->i_mutex);
2873 /* fsync's caller has already called _fdata{sync,write}, we want
2874 * that IO to finish before calling the osc and mdc sync methods */
2875 rc = filemap_fdatawait(inode->i_mapping);
2878 /* catch async errors that were recorded back when async writeback
2879 * failed for pages in this mapping. */
2880 if (!S_ISDIR(inode->i_mode)) {
2881 err = lli->lli_async_rc;
2882 lli->lli_async_rc = 0;
2885 err = lov_read_and_clear_async_rc(lli->lli_clob);
2890 oc = ll_mdscapa_get(inode);
2891 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2897 ptlrpc_req_finished(req);
2899 if (S_ISREG(inode->i_mode)) {
2900 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2902 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2903 if (rc == 0 && err < 0)
2906 fd->fd_write_failed = true;
2908 fd->fd_write_failed = false;
2911 #ifdef HAVE_FILE_FSYNC_4ARGS
2912 mutex_unlock(&inode->i_mutex);
2918 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2920 struct inode *inode = file->f_dentry->d_inode;
2921 struct ll_sb_info *sbi = ll_i2sbi(inode);
2922 struct ldlm_enqueue_info einfo = {
2923 .ei_type = LDLM_FLOCK,
2924 .ei_cb_cp = ldlm_flock_completion_ast,
2925 .ei_cbdata = file_lock,
2927 struct md_op_data *op_data;
2928 struct lustre_handle lockh = {0};
2929 ldlm_policy_data_t flock = {{0}};
2930 int fl_type = file_lock->fl_type;
2936 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2937 PFID(ll_inode2fid(inode)), file_lock);
2939 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2941 if (file_lock->fl_flags & FL_FLOCK) {
2942 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2943 /* flocks are whole-file locks */
2944 flock.l_flock.end = OFFSET_MAX;
2945 /* For flocks owner is determined by the local file desctiptor*/
2946 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
2947 } else if (file_lock->fl_flags & FL_POSIX) {
2948 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2949 flock.l_flock.start = file_lock->fl_start;
2950 flock.l_flock.end = file_lock->fl_end;
2954 flock.l_flock.pid = file_lock->fl_pid;
2956 /* Somewhat ugly workaround for svc lockd.
2957 * lockd installs custom fl_lmops->lm_compare_owner that checks
2958 * for the fl_owner to be the same (which it always is on local node
2959 * I guess between lockd processes) and then compares pid.
2960 * As such we assign pid to the owner field to make it all work,
2961 * conflict with normal locks is unlikely since pid space and
2962 * pointer space for current->files are not intersecting */
2963 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2964 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2968 einfo.ei_mode = LCK_PR;
2971 /* An unlock request may or may not have any relation to
2972 * existing locks so we may not be able to pass a lock handle
2973 * via a normal ldlm_lock_cancel() request. The request may even
2974 * unlock a byte range in the middle of an existing lock. In
2975 * order to process an unlock request we need all of the same
2976 * information that is given with a normal read or write record
2977 * lock request. To avoid creating another ldlm unlock (cancel)
2978 * message we'll treat a LCK_NL flock request as an unlock. */
2979 einfo.ei_mode = LCK_NL;
2982 einfo.ei_mode = LCK_PW;
2985 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
3000 flags = LDLM_FL_BLOCK_NOWAIT;
3006 flags = LDLM_FL_TEST_LOCK;
3009 CERROR("unknown fcntl lock command: %d\n", cmd);
3013 /* Save the old mode so that if the mode in the lock changes we
3014 * can decrement the appropriate reader or writer refcount. */
3015 file_lock->fl_type = einfo.ei_mode;
3017 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3018 LUSTRE_OPC_ANY, NULL);
3019 if (IS_ERR(op_data))
3020 RETURN(PTR_ERR(op_data));
3022 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags="LPX64", mode=%u, "
3023 "start="LPU64", end="LPU64"\n", PFID(ll_inode2fid(inode)),
3024 flock.l_flock.pid, flags, einfo.ei_mode,
3025 flock.l_flock.start, flock.l_flock.end);
3027 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
3030 /* Restore the file lock type if not TEST lock. */
3031 if (!(flags & LDLM_FL_TEST_LOCK))
3032 file_lock->fl_type = fl_type;
3034 if ((file_lock->fl_flags & FL_FLOCK) &&
3035 (rc == 0 || file_lock->fl_type == F_UNLCK))
3036 rc2 = flock_lock_file_wait(file, file_lock);
3037 if ((file_lock->fl_flags & FL_POSIX) &&
3038 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
3039 !(flags & LDLM_FL_TEST_LOCK))
3040 rc2 = posix_lock_file_wait(file, file_lock);
3042 if (rc2 && file_lock->fl_type != F_UNLCK) {
3043 einfo.ei_mode = LCK_NL;
3044 md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
3049 ll_finish_md_op_data(op_data);
3054 int ll_get_fid_by_name(struct inode *parent, const char *name,
3055 int namelen, struct lu_fid *fid)
3057 struct md_op_data *op_data = NULL;
3058 struct mdt_body *body;
3059 struct ptlrpc_request *req;
3063 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
3064 LUSTRE_OPC_ANY, NULL);
3065 if (IS_ERR(op_data))
3066 RETURN(PTR_ERR(op_data));
3068 op_data->op_valid = OBD_MD_FLID;
3069 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
3070 ll_finish_md_op_data(op_data);
3074 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3076 GOTO(out_req, rc = -EFAULT);
3078 *fid = body->mbo_fid1;
3080 ptlrpc_req_finished(req);
3084 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
3085 const char *name, int namelen)
3087 struct dentry *dchild = NULL;
3088 struct inode *child_inode = NULL;
3089 struct md_op_data *op_data;
3090 struct ptlrpc_request *request = NULL;
3095 CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%04x\n",
3096 name, PFID(ll_inode2fid(parent)), mdtidx);
3098 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
3099 0, LUSTRE_OPC_ANY, NULL);
3100 if (IS_ERR(op_data))
3101 RETURN(PTR_ERR(op_data));
3103 /* Get child FID first */
3104 qstr.hash = full_name_hash(name, namelen);
3107 dchild = d_lookup(file->f_dentry, &qstr);
3108 if (dchild != NULL && dchild->d_inode != NULL) {
3109 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
3110 if (dchild->d_inode != NULL) {
3111 child_inode = igrab(dchild->d_inode);
3112 ll_invalidate_aliases(child_inode);
3116 rc = ll_get_fid_by_name(parent, name, namelen,
3122 if (!fid_is_sane(&op_data->op_fid3)) {
3123 CERROR("%s: migrate %s , but fid "DFID" is insane\n",
3124 ll_get_fsname(parent->i_sb, NULL, 0), name,
3125 PFID(&op_data->op_fid3));
3126 GOTO(out_free, rc = -EINVAL);
3129 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
3134 CDEBUG(D_INFO, "%s:"DFID" is already on MDT%d.\n", name,
3135 PFID(&op_data->op_fid3), mdtidx);
3136 GOTO(out_free, rc = 0);
3139 op_data->op_mds = mdtidx;
3140 op_data->op_cli_flags = CLI_MIGRATE;
3141 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
3142 namelen, name, namelen, &request);
3144 ll_update_times(request, parent);
3146 ptlrpc_req_finished(request);
3151 if (child_inode != NULL) {
3152 clear_nlink(child_inode);
3156 ll_finish_md_op_data(op_data);
3161 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
3169 * test if some locks matching bits and l_req_mode are acquired
3170 * - bits can be in different locks
3171 * - if found clear the common lock bits in *bits
3172 * - the bits not found, are kept in *bits
3174 * \param bits [IN] searched lock bits [IN]
3175 * \param l_req_mode [IN] searched lock mode
3176 * \retval boolean, true iff all bits are found
3178 int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode)
3180 struct lustre_handle lockh;
3181 ldlm_policy_data_t policy;
3182 ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
3183 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
3192 fid = &ll_i2info(inode)->lli_fid;
3193 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
3194 ldlm_lockname[mode]);
3196 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
3197 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
3198 policy.l_inodebits.bits = *bits & (1 << i);
3199 if (policy.l_inodebits.bits == 0)
3202 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
3203 &policy, mode, &lockh)) {
3204 struct ldlm_lock *lock;
3206 lock = ldlm_handle2lock(&lockh);
3209 ~(lock->l_policy_data.l_inodebits.bits);
3210 LDLM_LOCK_PUT(lock);
3212 *bits &= ~policy.l_inodebits.bits;
3219 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
3220 struct lustre_handle *lockh, __u64 flags,
3223 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
3228 fid = &ll_i2info(inode)->lli_fid;
3229 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
3231 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
3232 fid, LDLM_IBITS, &policy, mode, lockh);
3237 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
3239 /* Already unlinked. Just update nlink and return success */
3240 if (rc == -ENOENT) {
3242 /* This path cannot be hit for regular files unless in
3243 * case of obscure races, so no need to to validate
3245 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
3247 } else if (rc != 0) {
3248 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
3249 "%s: revalidate FID "DFID" error: rc = %d\n",
3250 ll_get_fsname(inode->i_sb, NULL, 0),
3251 PFID(ll_inode2fid(inode)), rc);
3257 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3259 struct inode *inode = dentry->d_inode;
3260 struct ptlrpc_request *req = NULL;
3261 struct obd_export *exp;
3265 LASSERT(inode != NULL);
3267 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
3268 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
3270 exp = ll_i2mdexp(inode);
3272 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
3273 * But under CMD case, it caused some lock issues, should be fixed
3274 * with new CMD ibits lock. See bug 12718 */
3275 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
3276 struct lookup_intent oit = { .it_op = IT_GETATTR };
3277 struct md_op_data *op_data;
3279 if (ibits == MDS_INODELOCK_LOOKUP)
3280 oit.it_op = IT_LOOKUP;
3282 /* Call getattr by fid, so do not provide name at all. */
3283 op_data = ll_prep_md_op_data(NULL, dentry->d_inode,
3284 dentry->d_inode, NULL, 0, 0,
3285 LUSTRE_OPC_ANY, NULL);
3286 if (IS_ERR(op_data))
3287 RETURN(PTR_ERR(op_data));
3289 rc = md_intent_lock(exp, op_data, &oit, &req,
3290 &ll_md_blocking_ast, 0);
3291 ll_finish_md_op_data(op_data);
3293 rc = ll_inode_revalidate_fini(inode, rc);
3297 rc = ll_revalidate_it_finish(req, &oit, dentry);
3299 ll_intent_release(&oit);
3303 /* Unlinked? Unhash dentry, so it is not picked up later by
3304 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
3305 here to preserve get_cwd functionality on 2.6.
3307 if (!dentry->d_inode->i_nlink)
3308 d_lustre_invalidate(dentry, 0);
3310 ll_lookup_finish_locks(&oit, dentry);
3311 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
3312 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
3313 obd_valid valid = OBD_MD_FLGETATTR;
3314 struct md_op_data *op_data;
3317 if (S_ISREG(inode->i_mode)) {
3318 rc = ll_get_default_mdsize(sbi, &ealen);
3321 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
3324 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
3325 0, ealen, LUSTRE_OPC_ANY,
3327 if (IS_ERR(op_data))
3328 RETURN(PTR_ERR(op_data));
3330 op_data->op_valid = valid;
3331 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
3332 * capa for this inode. Because we only keep capas of dirs
3334 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3335 ll_finish_md_op_data(op_data);
3337 rc = ll_inode_revalidate_fini(inode, rc);
3341 rc = ll_prep_inode(&inode, req, NULL, NULL);
3344 ptlrpc_req_finished(req);
3348 static int ll_merge_md_attr(struct inode *inode)
3350 struct cl_attr attr = { 0 };
3353 LASSERT(ll_i2info(inode)->lli_lsm_md != NULL);
3354 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
3359 ll_i2info(inode)->lli_stripe_dir_size = attr.cat_size;
3360 ll_i2info(inode)->lli_stripe_dir_nlink = attr.cat_nlink;
3362 ll_i2info(inode)->lli_lvb.lvb_atime = attr.cat_atime;
3363 ll_i2info(inode)->lli_lvb.lvb_mtime = attr.cat_mtime;
3364 ll_i2info(inode)->lli_lvb.lvb_ctime = attr.cat_ctime;
3370 ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3372 struct inode *inode = dentry->d_inode;
3376 rc = __ll_inode_revalidate(dentry, ibits);
3380 /* if object isn't regular file, don't validate size */
3381 if (!S_ISREG(inode->i_mode)) {
3382 if (S_ISDIR(inode->i_mode) &&
3383 ll_i2info(inode)->lli_lsm_md != NULL) {
3384 rc = ll_merge_md_attr(inode);
3389 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
3390 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
3391 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
3393 /* In case of restore, the MDT has the right size and has
3394 * already send it back without granting the layout lock,
3395 * inode is up-to-date so glimpse is useless.
3396 * Also to glimpse we need the layout, in case of a running
3397 * restore the MDT holds the layout lock so the glimpse will
3398 * block up to the end of restore (getattr will block)
3400 if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING))
3401 rc = ll_glimpse_size(inode);
3406 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3408 struct inode *inode = de->d_inode;
3409 struct ll_sb_info *sbi = ll_i2sbi(inode);
3410 struct ll_inode_info *lli = ll_i2info(inode);
3413 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3414 MDS_INODELOCK_LOOKUP);
3415 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3420 stat->dev = inode->i_sb->s_dev;
3421 if (ll_need_32bit_api(sbi))
3422 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3424 stat->ino = inode->i_ino;
3425 stat->mode = inode->i_mode;
3426 stat->uid = inode->i_uid;
3427 stat->gid = inode->i_gid;
3428 stat->rdev = inode->i_rdev;
3429 stat->atime = inode->i_atime;
3430 stat->mtime = inode->i_mtime;
3431 stat->ctime = inode->i_ctime;
3432 stat->blksize = 1 << inode->i_blkbits;
3433 stat->blocks = inode->i_blocks;
3435 if (S_ISDIR(inode->i_mode) &&
3436 ll_i2info(inode)->lli_lsm_md != NULL) {
3437 stat->nlink = lli->lli_stripe_dir_nlink;
3438 stat->size = lli->lli_stripe_dir_size;
3440 stat->nlink = inode->i_nlink;
3441 stat->size = i_size_read(inode);
3447 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3448 __u64 start, __u64 len)
3452 struct ll_user_fiemap *fiemap;
3453 unsigned int extent_count = fieinfo->fi_extents_max;
3455 num_bytes = sizeof(*fiemap) + (extent_count *
3456 sizeof(struct ll_fiemap_extent));
3457 OBD_ALLOC_LARGE(fiemap, num_bytes);
3462 fiemap->fm_flags = fieinfo->fi_flags;
3463 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3464 fiemap->fm_start = start;
3465 fiemap->fm_length = len;
3466 if (extent_count > 0)
3467 memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3468 sizeof(struct ll_fiemap_extent));
3470 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3472 fieinfo->fi_flags = fiemap->fm_flags;
3473 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3474 if (extent_count > 0)
3475 memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3476 fiemap->fm_mapped_extents *
3477 sizeof(struct ll_fiemap_extent));
3479 OBD_FREE_LARGE(fiemap, num_bytes);
3483 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3485 struct ll_inode_info *lli = ll_i2info(inode);
3486 struct posix_acl *acl = NULL;
3489 spin_lock(&lli->lli_lock);
3490 /* VFS' acl_permission_check->check_acl will release the refcount */
3491 acl = posix_acl_dup(lli->lli_posix_acl);
3492 spin_unlock(&lli->lli_lock);
3497 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
3499 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3500 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
3502 ll_check_acl(struct inode *inode, int mask)
3505 # ifdef CONFIG_FS_POSIX_ACL
3506 struct posix_acl *acl;
3510 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3511 if (flags & IPERM_FLAG_RCU)
3514 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
3519 rc = posix_acl_permission(inode, acl, mask);
3520 posix_acl_release(acl);
3523 # else /* !CONFIG_FS_POSIX_ACL */
3525 # endif /* CONFIG_FS_POSIX_ACL */
3527 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
3529 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
3530 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
3532 # ifdef HAVE_INODE_PERMISION_2ARGS
3533 int ll_inode_permission(struct inode *inode, int mask)
3535 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
3540 struct ll_sb_info *sbi;
3541 struct root_squash_info *squash;
3542 struct cred *cred = NULL;
3543 const struct cred *old_cred = NULL;
3545 bool squash_id = false;
3548 #ifdef MAY_NOT_BLOCK
3549 if (mask & MAY_NOT_BLOCK)
3551 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
3552 if (flags & IPERM_FLAG_RCU)
3556 /* as root inode are NOT getting validated in lookup operation,
3557 * need to do it before permission check. */
3559 if (inode == inode->i_sb->s_root->d_inode) {
3560 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3561 MDS_INODELOCK_LOOKUP);
3566 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3567 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3569 /* squash fsuid/fsgid if needed */
3570 sbi = ll_i2sbi(inode);
3571 squash = &sbi->ll_squash;
3572 if (unlikely(squash->rsi_uid != 0 &&
3573 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3574 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3578 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3579 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3580 squash->rsi_uid, squash->rsi_gid);
3582 /* update current process's credentials
3583 * and FS capability */
3584 cred = prepare_creds();
3588 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3589 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3590 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3591 if ((1 << cap) & CFS_CAP_FS_MASK)
3592 cap_lower(cred->cap_effective, cap);
3594 old_cred = override_creds(cred);
3597 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
3599 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
3600 rc = lustre_check_remote_perm(inode, mask);
3602 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
3604 /* restore current process's credentials and FS capability */
3606 revert_creds(old_cred);
3613 /* -o localflock - only provides locally consistent flock locks */
3614 struct file_operations ll_file_operations = {
3615 .read = ll_file_read,
3616 .aio_read = ll_file_aio_read,
3617 .write = ll_file_write,
3618 .aio_write = ll_file_aio_write,
3619 .unlocked_ioctl = ll_file_ioctl,
3620 .open = ll_file_open,
3621 .release = ll_file_release,
3622 .mmap = ll_file_mmap,
3623 .llseek = ll_file_seek,
3624 .splice_read = ll_file_splice_read,
3629 struct file_operations ll_file_operations_flock = {
3630 .read = ll_file_read,
3631 .aio_read = ll_file_aio_read,
3632 .write = ll_file_write,
3633 .aio_write = ll_file_aio_write,
3634 .unlocked_ioctl = ll_file_ioctl,
3635 .open = ll_file_open,
3636 .release = ll_file_release,
3637 .mmap = ll_file_mmap,
3638 .llseek = ll_file_seek,
3639 .splice_read = ll_file_splice_read,
3642 .flock = ll_file_flock,
3643 .lock = ll_file_flock
3646 /* These are for -o noflock - to return ENOSYS on flock calls */
3647 struct file_operations ll_file_operations_noflock = {
3648 .read = ll_file_read,
3649 .aio_read = ll_file_aio_read,
3650 .write = ll_file_write,
3651 .aio_write = ll_file_aio_write,
3652 .unlocked_ioctl = ll_file_ioctl,
3653 .open = ll_file_open,
3654 .release = ll_file_release,
3655 .mmap = ll_file_mmap,
3656 .llseek = ll_file_seek,
3657 .splice_read = ll_file_splice_read,
3660 .flock = ll_file_noflock,
3661 .lock = ll_file_noflock
3664 struct inode_operations ll_file_inode_operations = {
3665 .setattr = ll_setattr,
3666 .getattr = ll_getattr,
3667 .permission = ll_inode_permission,
3668 .setxattr = ll_setxattr,
3669 .getxattr = ll_getxattr,
3670 .listxattr = ll_listxattr,
3671 .removexattr = ll_removexattr,
3672 .fiemap = ll_fiemap,
3673 #ifdef HAVE_IOP_GET_ACL
3674 .get_acl = ll_get_acl,
3678 /* dynamic ioctl number support routins */
3679 static struct llioc_ctl_data {
3680 struct rw_semaphore ioc_sem;
3681 struct list_head ioc_head;
3683 __RWSEM_INITIALIZER(llioc.ioc_sem),
3684 LIST_HEAD_INIT(llioc.ioc_head)
3689 struct list_head iocd_list;
3690 unsigned int iocd_size;
3691 llioc_callback_t iocd_cb;
3692 unsigned int iocd_count;
3693 unsigned int iocd_cmd[0];
3696 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3699 struct llioc_data *in_data = NULL;
3702 if (cb == NULL || cmd == NULL ||
3703 count > LLIOC_MAX_CMD || count < 0)
3706 size = sizeof(*in_data) + count * sizeof(unsigned int);
3707 OBD_ALLOC(in_data, size);
3708 if (in_data == NULL)
3711 memset(in_data, 0, sizeof(*in_data));
3712 in_data->iocd_size = size;
3713 in_data->iocd_cb = cb;
3714 in_data->iocd_count = count;
3715 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3717 down_write(&llioc.ioc_sem);
3718 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3719 up_write(&llioc.ioc_sem);
3724 void ll_iocontrol_unregister(void *magic)
3726 struct llioc_data *tmp;
3731 down_write(&llioc.ioc_sem);
3732 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3734 unsigned int size = tmp->iocd_size;
3736 list_del(&tmp->iocd_list);
3737 up_write(&llioc.ioc_sem);
3739 OBD_FREE(tmp, size);
3743 up_write(&llioc.ioc_sem);
3745 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3748 EXPORT_SYMBOL(ll_iocontrol_register);
3749 EXPORT_SYMBOL(ll_iocontrol_unregister);
3751 static enum llioc_iter
3752 ll_iocontrol_call(struct inode *inode, struct file *file,
3753 unsigned int cmd, unsigned long arg, int *rcp)
3755 enum llioc_iter ret = LLIOC_CONT;
3756 struct llioc_data *data;
3757 int rc = -EINVAL, i;
3759 down_read(&llioc.ioc_sem);
3760 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3761 for (i = 0; i < data->iocd_count; i++) {
3762 if (cmd != data->iocd_cmd[i])
3765 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3769 if (ret == LLIOC_STOP)
3772 up_read(&llioc.ioc_sem);
3779 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3781 struct ll_inode_info *lli = ll_i2info(inode);
3782 struct cl_env_nest nest;
3787 if (lli->lli_clob == NULL)
3790 env = cl_env_nested_get(&nest);
3792 RETURN(PTR_ERR(env));
3794 result = cl_conf_set(env, lli->lli_clob, conf);
3795 cl_env_nested_put(&nest, env);
3797 if (conf->coc_opc == OBJECT_CONF_SET) {
3798 struct ldlm_lock *lock = conf->coc_lock;
3800 LASSERT(lock != NULL);
3801 LASSERT(ldlm_has_layout(lock));
3803 struct lustre_md *md = conf->u.coc_md;
3804 __u32 gen = LL_LAYOUT_GEN_EMPTY;
3806 /* it can only be allowed to match after layout is
3807 * applied to inode otherwise false layout would be
3808 * seen. Applying layout shoud happen before dropping
3809 * the intent lock. */
3810 ldlm_lock_allow_match(lock);
3812 lli->lli_has_smd = lsm_has_objects(md->lsm);
3813 if (md->lsm != NULL)
3814 gen = md->lsm->lsm_layout_gen;
3817 DFID ": layout version change: %u -> %u\n",
3818 PFID(&lli->lli_fid), ll_layout_version_get(lli),
3820 ll_layout_version_set(lli, gen);
3826 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3827 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3830 struct ll_sb_info *sbi = ll_i2sbi(inode);
3831 struct obd_capa *oc;
3832 struct ptlrpc_request *req;
3833 struct mdt_body *body;
3840 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3841 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3842 lock->l_lvb_data, lock->l_lvb_len);
3844 if ((lock->l_lvb_data != NULL) && ldlm_is_lvb_ready(lock))
3847 /* if layout lock was granted right away, the layout is returned
3848 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3849 * blocked and then granted via completion ast, we have to fetch
3850 * layout here. Please note that we can't use the LVB buffer in
3851 * completion AST because it doesn't have a large enough buffer */
3852 oc = ll_mdscapa_get(inode);
3853 rc = ll_get_default_mdsize(sbi, &lmmsize);
3855 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
3856 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3862 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3864 GOTO(out, rc = -EPROTO);
3866 lmmsize = body->mbo_eadatasize;
3867 if (lmmsize == 0) /* empty layout */
3870 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3872 GOTO(out, rc = -EFAULT);
3874 OBD_ALLOC_LARGE(lvbdata, lmmsize);
3875 if (lvbdata == NULL)
3876 GOTO(out, rc = -ENOMEM);
3878 memcpy(lvbdata, lmm, lmmsize);
3879 lock_res_and_lock(lock);
3880 if (lock->l_lvb_data != NULL)
3881 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
3883 lock->l_lvb_data = lvbdata;
3884 lock->l_lvb_len = lmmsize;
3885 unlock_res_and_lock(lock);
3890 ptlrpc_req_finished(req);
3895 * Apply the layout to the inode. Layout lock is held and will be released
3898 static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
3899 struct inode *inode, __u32 *gen, bool reconf)
3901 struct ll_inode_info *lli = ll_i2info(inode);
3902 struct ll_sb_info *sbi = ll_i2sbi(inode);
3903 struct ldlm_lock *lock;
3904 struct lustre_md md = { NULL };
3905 struct cl_object_conf conf;
3908 bool wait_layout = false;
3911 LASSERT(lustre_handle_is_used(lockh));
3913 lock = ldlm_handle2lock(lockh);
3914 LASSERT(lock != NULL);
3915 LASSERT(ldlm_has_layout(lock));
3917 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d",
3918 PFID(&lli->lli_fid), inode, reconf);
3920 /* in case this is a caching lock and reinstate with new inode */
3921 md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
3923 lock_res_and_lock(lock);
3924 lvb_ready = ldlm_is_lvb_ready(lock);
3925 unlock_res_and_lock(lock);
3926 /* checking lvb_ready is racy but this is okay. The worst case is
3927 * that multi processes may configure the file on the same time. */
3929 if (lvb_ready || !reconf) {
3932 /* layout_gen must be valid if layout lock is not
3933 * cancelled and stripe has already set */
3934 *gen = ll_layout_version_get(lli);
3940 rc = ll_layout_fetch(inode, lock);
3944 /* for layout lock, lmm is returned in lock's lvb.
3945 * lvb_data is immutable if the lock is held so it's safe to access it
3946 * without res lock. See the description in ldlm_lock_decref_internal()
3947 * for the condition to free lvb_data of layout lock */
3948 if (lock->l_lvb_data != NULL) {
3949 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3950 lock->l_lvb_data, lock->l_lvb_len);
3952 *gen = LL_LAYOUT_GEN_EMPTY;
3954 *gen = md.lsm->lsm_layout_gen;
3957 CERROR("%s: file "DFID" unpackmd error: %d\n",
3958 ll_get_fsname(inode->i_sb, NULL, 0),
3959 PFID(&lli->lli_fid), rc);
3965 /* set layout to file. Unlikely this will fail as old layout was
3966 * surely eliminated */
3967 memset(&conf, 0, sizeof conf);
3968 conf.coc_opc = OBJECT_CONF_SET;
3969 conf.coc_inode = inode;
3970 conf.coc_lock = lock;
3971 conf.u.coc_md = &md;
3972 rc = ll_layout_conf(inode, &conf);
3975 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3977 /* refresh layout failed, need to wait */
3978 wait_layout = rc == -EBUSY;
3982 LDLM_LOCK_PUT(lock);
3983 ldlm_lock_decref(lockh, mode);
3985 /* wait for IO to complete if it's still being used. */
3987 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3988 ll_get_fsname(inode->i_sb, NULL, 0),
3989 PFID(&lli->lli_fid), inode);
3991 memset(&conf, 0, sizeof conf);
3992 conf.coc_opc = OBJECT_CONF_WAIT;
3993 conf.coc_inode = inode;
3994 rc = ll_layout_conf(inode, &conf);
3998 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
3999 ll_get_fsname(inode->i_sb, NULL, 0),
4000 PFID(&lli->lli_fid), rc);
4006 * This function checks if there exists a LAYOUT lock on the client side,
4007 * or enqueues it if it doesn't have one in cache.
4009 * This function will not hold layout lock so it may be revoked any time after
4010 * this function returns. Any operations depend on layout should be redone
4013 * This function should be called before lov_io_init() to get an uptodate
4014 * layout version, the caller should save the version number and after IO
4015 * is finished, this function should be called again to verify that layout
4016 * is not changed during IO time.
4018 int ll_layout_refresh(struct inode *inode, __u32 *gen)
4020 struct ll_inode_info *lli = ll_i2info(inode);
4021 struct ll_sb_info *sbi = ll_i2sbi(inode);
4022 struct md_op_data *op_data;
4023 struct lookup_intent it;
4024 struct lustre_handle lockh;
4026 struct ldlm_enqueue_info einfo = {
4027 .ei_type = LDLM_IBITS,
4029 .ei_cb_bl = &ll_md_blocking_ast,
4030 .ei_cb_cp = &ldlm_completion_ast,
4035 *gen = ll_layout_version_get(lli);
4036 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
4040 LASSERT(fid_is_sane(ll_inode2fid(inode)));
4041 LASSERT(S_ISREG(inode->i_mode));
4043 /* take layout lock mutex to enqueue layout lock exclusively. */
4044 mutex_lock(&lli->lli_layout_mutex);
4047 /* mostly layout lock is caching on the local side, so try to match
4048 * it before grabbing layout lock mutex. */
4049 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
4050 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
4051 if (mode != 0) { /* hit cached lock */
4052 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4056 mutex_unlock(&lli->lli_layout_mutex);
4060 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
4061 0, 0, LUSTRE_OPC_ANY, NULL);
4062 if (IS_ERR(op_data)) {
4063 mutex_unlock(&lli->lli_layout_mutex);
4064 RETURN(PTR_ERR(op_data));
4067 /* have to enqueue one */
4068 memset(&it, 0, sizeof(it));
4069 it.it_op = IT_LAYOUT;
4070 lockh.cookie = 0ULL;
4072 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
4073 ll_get_fsname(inode->i_sb, NULL, 0),
4074 PFID(&lli->lli_fid), inode);
4076 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
4077 if (it.d.lustre.it_data != NULL)
4078 ptlrpc_req_finished(it.d.lustre.it_data);
4079 it.d.lustre.it_data = NULL;
4081 ll_finish_md_op_data(op_data);
4083 mode = it.d.lustre.it_lock_mode;
4084 it.d.lustre.it_lock_mode = 0;
4085 ll_intent_drop_lock(&it);
4088 /* set lock data in case this is a new lock */
4089 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
4090 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4094 mutex_unlock(&lli->lli_layout_mutex);
4100 * This function send a restore request to the MDT
4102 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
4104 struct hsm_user_request *hur;
4108 len = sizeof(struct hsm_user_request) +
4109 sizeof(struct hsm_user_item);
4110 OBD_ALLOC(hur, len);
4114 hur->hur_request.hr_action = HUA_RESTORE;
4115 hur->hur_request.hr_archive_id = 0;
4116 hur->hur_request.hr_flags = 0;
4117 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
4118 sizeof(hur->hur_user_item[0].hui_fid));
4119 hur->hur_user_item[0].hui_extent.offset = offset;
4120 hur->hur_user_item[0].hui_extent.length = length;
4121 hur->hur_request.hr_itemcount = 1;
4122 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, cl_i2sbi(inode)->ll_md_exp,