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 * Different processes can open the same dir, "ll_opendir_key" means:
388 * it is me that should stop the statahead thread. */
389 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
390 lli->lli_opendir_pid != 0)
391 ll_stop_statahead(inode, lli->lli_opendir_key);
393 if (inode->i_sb->s_root == file->f_dentry) {
394 LUSTRE_FPRIVATE(file) = NULL;
395 ll_file_data_put(fd);
399 if (!S_ISDIR(inode->i_mode)) {
400 if (lli->lli_clob != NULL)
401 lov_read_and_clear_async_rc(lli->lli_clob);
402 lli->lli_async_rc = 0;
405 rc = ll_md_close(sbi->ll_md_exp, inode, file);
407 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
408 libcfs_debug_dumplog();
413 static int ll_intent_file_open(struct file *file, void *lmm, int lmmsize,
414 struct lookup_intent *itp)
416 struct dentry *de = file->f_dentry;
417 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
418 struct dentry *parent = de->d_parent;
419 const char *name = NULL;
421 struct md_op_data *op_data;
422 struct ptlrpc_request *req = NULL;
426 LASSERT(parent != NULL);
427 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
429 /* if server supports open-by-fid, or file name is invalid, don't pack
430 * name in open request */
431 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
432 lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
433 name = de->d_name.name;
434 len = de->d_name.len;
437 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
438 name, len, 0, LUSTRE_OPC_ANY, NULL);
440 RETURN(PTR_ERR(op_data));
441 op_data->op_data = lmm;
442 op_data->op_data_size = lmmsize;
444 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
445 &ll_md_blocking_ast, 0);
446 ll_finish_md_op_data(op_data);
448 /* reason for keep own exit path - don`t flood log
449 * with messages with -ESTALE errors.
451 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
452 it_open_error(DISP_OPEN_OPEN, itp))
454 ll_release_openhandle(de, itp);
458 if (it_disposition(itp, DISP_LOOKUP_NEG))
459 GOTO(out, rc = -ENOENT);
461 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
462 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
463 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
467 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
468 if (!rc && itp->d.lustre.it_lock_mode)
469 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
472 ptlrpc_req_finished(req);
473 ll_intent_drop_lock(itp);
479 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
480 * not believe attributes if a few ioepoch holders exist. Attributes for
481 * previous ioepoch if new one is opened are also skipped by MDS.
483 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
485 if (ioepoch && lli->lli_ioepoch != ioepoch) {
486 lli->lli_ioepoch = ioepoch;
487 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
488 ioepoch, PFID(&lli->lli_fid));
492 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
493 struct obd_client_handle *och)
495 struct ptlrpc_request *req = it->d.lustre.it_data;
496 struct mdt_body *body;
498 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
499 och->och_fh = body->mbo_handle;
500 och->och_fid = body->mbo_fid1;
501 och->och_lease_handle.cookie = it->d.lustre.it_lock_handle;
502 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
503 och->och_flags = it->it_flags;
505 return md_set_open_replay_data(md_exp, och, it);
508 static int ll_local_open(struct file *file, struct lookup_intent *it,
509 struct ll_file_data *fd, struct obd_client_handle *och)
511 struct inode *inode = file->f_dentry->d_inode;
512 struct ll_inode_info *lli = ll_i2info(inode);
515 LASSERT(!LUSTRE_FPRIVATE(file));
520 struct ptlrpc_request *req = it->d.lustre.it_data;
521 struct mdt_body *body;
524 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
528 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
529 ll_ioepoch_open(lli, body->mbo_ioepoch);
532 LUSTRE_FPRIVATE(file) = fd;
533 ll_readahead_init(inode, &fd->fd_ras);
534 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
536 /* ll_cl_context initialize */
537 rwlock_init(&fd->fd_lock);
538 INIT_LIST_HEAD(&fd->fd_lccs);
543 /* Open a file, and (for the very first open) create objects on the OSTs at
544 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
545 * creation or open until ll_lov_setstripe() ioctl is called.
547 * If we already have the stripe MD locally then we don't request it in
548 * md_open(), by passing a lmm_size = 0.
550 * It is up to the application to ensure no other processes open this file
551 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
552 * used. We might be able to avoid races of that sort by getting lli_open_sem
553 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
554 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
556 int ll_file_open(struct inode *inode, struct file *file)
558 struct ll_inode_info *lli = ll_i2info(inode);
559 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
560 .it_flags = file->f_flags };
561 struct obd_client_handle **och_p = NULL;
562 __u64 *och_usecount = NULL;
563 struct ll_file_data *fd;
564 int rc = 0, opendir_set = 0;
567 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
568 PFID(ll_inode2fid(inode)), inode, file->f_flags);
570 it = file->private_data; /* XXX: compat macro */
571 file->private_data = NULL; /* prevent ll_local_open assertion */
573 fd = ll_file_data_get();
575 GOTO(out_openerr, rc = -ENOMEM);
578 if (S_ISDIR(inode->i_mode)) {
579 spin_lock(&lli->lli_sa_lock);
580 if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL &&
581 lli->lli_opendir_pid == 0) {
582 lli->lli_opendir_key = fd;
583 lli->lli_opendir_pid = current_pid();
586 spin_unlock(&lli->lli_sa_lock);
589 if (inode->i_sb->s_root == file->f_dentry) {
590 LUSTRE_FPRIVATE(file) = fd;
594 if (!it || !it->d.lustre.it_disposition) {
595 /* Convert f_flags into access mode. We cannot use file->f_mode,
596 * because everything but O_ACCMODE mask was stripped from
598 if ((oit.it_flags + 1) & O_ACCMODE)
600 if (file->f_flags & O_TRUNC)
601 oit.it_flags |= FMODE_WRITE;
603 /* kernel only call f_op->open in dentry_open. filp_open calls
604 * dentry_open after call to open_namei that checks permissions.
605 * Only nfsd_open call dentry_open directly without checking
606 * permissions and because of that this code below is safe. */
607 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
608 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
610 /* We do not want O_EXCL here, presumably we opened the file
611 * already? XXX - NFS implications? */
612 oit.it_flags &= ~O_EXCL;
614 /* bug20584, if "it_flags" contains O_CREAT, the file will be
615 * created if necessary, then "IT_CREAT" should be set to keep
616 * consistent with it */
617 if (oit.it_flags & O_CREAT)
618 oit.it_op |= IT_CREAT;
624 /* Let's see if we have file open on MDS already. */
625 if (it->it_flags & FMODE_WRITE) {
626 och_p = &lli->lli_mds_write_och;
627 och_usecount = &lli->lli_open_fd_write_count;
628 } else if (it->it_flags & FMODE_EXEC) {
629 och_p = &lli->lli_mds_exec_och;
630 och_usecount = &lli->lli_open_fd_exec_count;
632 och_p = &lli->lli_mds_read_och;
633 och_usecount = &lli->lli_open_fd_read_count;
636 mutex_lock(&lli->lli_och_mutex);
637 if (*och_p) { /* Open handle is present */
638 if (it_disposition(it, DISP_OPEN_OPEN)) {
639 /* Well, there's extra open request that we do not need,
640 let's close it somehow. This will decref request. */
641 rc = it_open_error(DISP_OPEN_OPEN, it);
643 mutex_unlock(&lli->lli_och_mutex);
644 GOTO(out_openerr, rc);
647 ll_release_openhandle(file->f_dentry, it);
651 rc = ll_local_open(file, it, fd, NULL);
654 mutex_unlock(&lli->lli_och_mutex);
655 GOTO(out_openerr, rc);
658 LASSERT(*och_usecount == 0);
659 if (!it->d.lustre.it_disposition) {
660 /* We cannot just request lock handle now, new ELC code
661 means that one of other OPEN locks for this file
662 could be cancelled, and since blocking ast handler
663 would attempt to grab och_mutex as well, that would
664 result in a deadlock */
665 mutex_unlock(&lli->lli_och_mutex);
667 * Normally called under two situations:
669 * 2. A race/condition on MDS resulting in no open
670 * handle to be returned from LOOKUP|OPEN request,
671 * for example if the target entry was a symlink.
673 * Always fetch MDS_OPEN_LOCK if this is not setstripe.
675 * Always specify MDS_OPEN_BY_FID because we don't want
676 * to get file with different fid.
678 it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
679 rc = ll_intent_file_open(file, NULL, 0, it);
681 GOTO(out_openerr, rc);
685 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
687 GOTO(out_och_free, rc = -ENOMEM);
691 /* md_intent_lock() didn't get a request ref if there was an
692 * open error, so don't do cleanup on the request here
694 /* XXX (green): Should not we bail out on any error here, not
695 * just open error? */
696 rc = it_open_error(DISP_OPEN_OPEN, it);
698 GOTO(out_och_free, rc);
700 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
701 "inode %p: disposition %x, status %d\n", inode,
702 it_disposition(it, ~0), it->d.lustre.it_status);
704 rc = ll_local_open(file, it, fd, *och_p);
706 GOTO(out_och_free, rc);
708 mutex_unlock(&lli->lli_och_mutex);
711 /* Must do this outside lli_och_mutex lock to prevent deadlock where
712 different kind of OPEN lock for this same inode gets cancelled
713 by ldlm_cancel_lru */
714 if (!S_ISREG(inode->i_mode))
715 GOTO(out_och_free, rc);
719 if (!lli->lli_has_smd &&
720 (cl_is_lov_delay_create(file->f_flags) ||
721 (file->f_mode & FMODE_WRITE) == 0)) {
722 CDEBUG(D_INODE, "object creation was delayed\n");
723 GOTO(out_och_free, rc);
725 cl_lov_delay_create_clear(&file->f_flags);
726 GOTO(out_och_free, rc);
730 if (och_p && *och_p) {
731 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
732 *och_p = NULL; /* OBD_FREE writes some magic there */
735 mutex_unlock(&lli->lli_och_mutex);
738 if (opendir_set != 0)
739 ll_stop_statahead(inode, lli->lli_opendir_key);
741 ll_file_data_put(fd);
743 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
746 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
747 ptlrpc_req_finished(it->d.lustre.it_data);
748 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
754 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
755 struct ldlm_lock_desc *desc, void *data, int flag)
758 struct lustre_handle lockh;
762 case LDLM_CB_BLOCKING:
763 ldlm_lock2handle(lock, &lockh);
764 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
766 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
770 case LDLM_CB_CANCELING:
778 * Acquire a lease and open the file.
780 static struct obd_client_handle *
781 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
784 struct lookup_intent it = { .it_op = IT_OPEN };
785 struct ll_sb_info *sbi = ll_i2sbi(inode);
786 struct md_op_data *op_data;
787 struct ptlrpc_request *req = NULL;
788 struct lustre_handle old_handle = { 0 };
789 struct obd_client_handle *och = NULL;
794 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
795 RETURN(ERR_PTR(-EINVAL));
798 struct ll_inode_info *lli = ll_i2info(inode);
799 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
800 struct obd_client_handle **och_p;
803 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
804 RETURN(ERR_PTR(-EPERM));
806 /* Get the openhandle of the file */
808 mutex_lock(&lli->lli_och_mutex);
809 if (fd->fd_lease_och != NULL) {
810 mutex_unlock(&lli->lli_och_mutex);
814 if (fd->fd_och == NULL) {
815 if (file->f_mode & FMODE_WRITE) {
816 LASSERT(lli->lli_mds_write_och != NULL);
817 och_p = &lli->lli_mds_write_och;
818 och_usecount = &lli->lli_open_fd_write_count;
820 LASSERT(lli->lli_mds_read_och != NULL);
821 och_p = &lli->lli_mds_read_och;
822 och_usecount = &lli->lli_open_fd_read_count;
824 if (*och_usecount == 1) {
831 mutex_unlock(&lli->lli_och_mutex);
832 if (rc < 0) /* more than 1 opener */
835 LASSERT(fd->fd_och != NULL);
836 old_handle = fd->fd_och->och_fh;
841 RETURN(ERR_PTR(-ENOMEM));
843 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
844 LUSTRE_OPC_ANY, NULL);
846 GOTO(out, rc = PTR_ERR(op_data));
848 /* To tell the MDT this openhandle is from the same owner */
849 op_data->op_handle = old_handle;
851 it.it_flags = fmode | open_flags;
852 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
853 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
854 &ll_md_blocking_lease_ast,
855 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
856 * it can be cancelled which may mislead applications that the lease is
858 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
859 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
860 * doesn't deal with openhandle, so normal openhandle will be leaked. */
861 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
862 ll_finish_md_op_data(op_data);
863 ptlrpc_req_finished(req);
865 GOTO(out_release_it, rc);
867 if (it_disposition(&it, DISP_LOOKUP_NEG))
868 GOTO(out_release_it, rc = -ENOENT);
870 rc = it_open_error(DISP_OPEN_OPEN, &it);
872 GOTO(out_release_it, rc);
874 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
875 ll_och_fill(sbi->ll_md_exp, &it, och);
877 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
878 GOTO(out_close, rc = -EOPNOTSUPP);
880 /* already get lease, handle lease lock */
881 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
882 if (it.d.lustre.it_lock_mode == 0 ||
883 it.d.lustre.it_lock_bits != MDS_INODELOCK_OPEN) {
884 /* open lock must return for lease */
885 CERROR(DFID "lease granted but no open lock, %d/"LPU64".\n",
886 PFID(ll_inode2fid(inode)), it.d.lustre.it_lock_mode,
887 it.d.lustre.it_lock_bits);
888 GOTO(out_close, rc = -EPROTO);
891 ll_intent_release(&it);
895 /* Cancel open lock */
896 if (it.d.lustre.it_lock_mode != 0) {
897 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
898 it.d.lustre.it_lock_mode);
899 it.d.lustre.it_lock_mode = 0;
900 och->och_lease_handle.cookie = 0ULL;
902 rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL);
904 CERROR("%s: error closing file "DFID": %d\n",
905 ll_get_fsname(inode->i_sb, NULL, 0),
906 PFID(&ll_i2info(inode)->lli_fid), rc2);
907 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
909 ll_intent_release(&it);
917 * Release lease and close the file.
918 * It will check if the lease has ever broken.
920 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
923 struct ldlm_lock *lock;
924 bool cancelled = true;
928 lock = ldlm_handle2lock(&och->och_lease_handle);
930 lock_res_and_lock(lock);
931 cancelled = ldlm_is_cancel(lock);
932 unlock_res_and_lock(lock);
936 CDEBUG(D_INODE, "lease for "DFID" broken? %d\n",
937 PFID(&ll_i2info(inode)->lli_fid), cancelled);
940 ldlm_cli_cancel(&och->och_lease_handle, 0);
941 if (lease_broken != NULL)
942 *lease_broken = cancelled;
944 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
949 /* Fills the obdo with the attributes for the lsm */
950 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
951 struct obd_capa *capa, struct obdo *obdo,
952 __u64 ioepoch, int dv_flags)
954 struct ptlrpc_request_set *set;
955 struct obd_info oinfo = { { { 0 } } };
960 LASSERT(lsm != NULL);
964 oinfo.oi_oa->o_oi = lsm->lsm_oi;
965 oinfo.oi_oa->o_mode = S_IFREG;
966 oinfo.oi_oa->o_ioepoch = ioepoch;
967 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
968 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
969 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
970 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
971 OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
972 OBD_MD_FLDATAVERSION;
973 oinfo.oi_capa = capa;
974 if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) {
975 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
976 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
977 if (dv_flags & LL_DV_WR_FLUSH)
978 oinfo.oi_oa->o_flags |= OBD_FL_FLUSH;
981 set = ptlrpc_prep_set();
983 CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM);
986 rc = obd_getattr_async(exp, &oinfo, set);
988 rc = ptlrpc_set_wait(set);
989 ptlrpc_set_destroy(set);
992 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
993 OBD_MD_FLATIME | OBD_MD_FLMTIME |
994 OBD_MD_FLCTIME | OBD_MD_FLSIZE |
995 OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS);
996 if (dv_flags & LL_DV_WR_FLUSH &&
997 !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
998 oinfo.oi_oa->o_flags & OBD_FL_FLUSH))
1005 * Performs the getattr on the inode and updates its fields.
1006 * If @sync != 0, perform the getattr under the server-side lock.
1008 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
1009 __u64 ioepoch, int sync)
1011 struct obd_capa *capa = ll_mdscapa_get(inode);
1012 struct lov_stripe_md *lsm;
1016 lsm = ccc_inode_lsm_get(inode);
1017 rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode),
1018 capa, obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0);
1021 struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi;
1023 obdo_refresh_inode(inode, obdo, obdo->o_valid);
1024 CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu,"
1025 " blksize %lu\n", POSTID(oi), i_size_read(inode),
1026 (unsigned long long)inode->i_blocks,
1027 (unsigned long)ll_inode_blksize(inode));
1029 ccc_inode_lsm_put(inode, lsm);
1033 int ll_merge_lvb(const struct lu_env *env, struct inode *inode)
1035 struct ll_inode_info *lli = ll_i2info(inode);
1036 struct cl_object *obj = lli->lli_clob;
1037 struct cl_attr *attr = ccc_env_thread_attr(env);
1043 ll_inode_size_lock(inode);
1044 /* merge timestamps the most recently obtained from mds with
1045 timestamps obtained from osts */
1046 LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime;
1047 LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime;
1048 LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime;
1049 inode_init_lvb(inode, &lvb);
1051 cl_object_attr_lock(obj);
1052 rc = cl_object_attr_get(env, obj, attr);
1053 cl_object_attr_unlock(obj);
1056 if (lvb.lvb_atime < attr->cat_atime)
1057 lvb.lvb_atime = attr->cat_atime;
1058 if (lvb.lvb_ctime < attr->cat_ctime)
1059 lvb.lvb_ctime = attr->cat_ctime;
1060 if (lvb.lvb_mtime < attr->cat_mtime)
1061 lvb.lvb_mtime = attr->cat_mtime;
1063 CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n",
1064 PFID(&lli->lli_fid), attr->cat_size);
1065 cl_isize_write_nolock(inode, attr->cat_size);
1067 inode->i_blocks = attr->cat_blocks;
1069 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
1070 LTIME_S(inode->i_atime) = lvb.lvb_atime;
1071 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
1073 ll_inode_size_unlock(inode);
1078 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
1081 struct obdo obdo = { 0 };
1084 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0);
1086 st->st_size = obdo.o_size;
1087 st->st_blocks = obdo.o_blocks;
1088 st->st_mtime = obdo.o_mtime;
1089 st->st_atime = obdo.o_atime;
1090 st->st_ctime = obdo.o_ctime;
1095 static bool file_is_noatime(const struct file *file)
1097 const struct vfsmount *mnt = file->f_path.mnt;
1098 const struct inode *inode = file->f_path.dentry->d_inode;
1100 /* Adapted from file_accessed() and touch_atime().*/
1101 if (file->f_flags & O_NOATIME)
1104 if (inode->i_flags & S_NOATIME)
1107 if (IS_NOATIME(inode))
1110 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1113 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1116 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1122 static void ll_io_init(struct cl_io *io, const struct file *file, int write)
1124 struct inode *inode = file->f_dentry->d_inode;
1126 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1128 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1129 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1130 file->f_flags & O_DIRECT ||
1133 io->ci_obj = ll_i2info(inode)->lli_clob;
1134 io->ci_lockreq = CILR_MAYBE;
1135 if (ll_file_nolock(file)) {
1136 io->ci_lockreq = CILR_NEVER;
1137 io->ci_no_srvlock = 1;
1138 } else if (file->f_flags & O_APPEND) {
1139 io->ci_lockreq = CILR_MANDATORY;
1142 io->ci_noatime = file_is_noatime(file);
1146 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1147 struct file *file, enum cl_io_type iot,
1148 loff_t *ppos, size_t count)
1150 struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
1151 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1156 CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: "LPU64", count: %zd\n",
1157 file->f_dentry->d_name.name, iot, *ppos, count);
1160 io = ccc_env_thread_io(env);
1161 ll_io_init(io, file, iot == CIT_WRITE);
1163 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1164 struct vvp_io *vio = vvp_env_io(env);
1165 struct ccc_io *cio = ccc_env_io(env);
1166 int write_mutex_locked = 0;
1168 cio->cui_fd = LUSTRE_FPRIVATE(file);
1169 vio->cui_io_subtype = args->via_io_subtype;
1171 switch (vio->cui_io_subtype) {
1173 cio->cui_iov = args->u.normal.via_iov;
1174 cio->cui_nrsegs = args->u.normal.via_nrsegs;
1175 cio->cui_tot_nrsegs = cio->cui_nrsegs;
1176 cio->cui_iocb = args->u.normal.via_iocb;
1177 if ((iot == CIT_WRITE) &&
1178 !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1179 if (mutex_lock_interruptible(&lli->
1181 GOTO(out, result = -ERESTARTSYS);
1182 write_mutex_locked = 1;
1184 down_read(&lli->lli_trunc_sem);
1187 vio->u.splice.cui_pipe = args->u.splice.via_pipe;
1188 vio->u.splice.cui_flags = args->u.splice.via_flags;
1191 CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
1195 ll_cl_add(file, env, io);
1196 result = cl_io_loop(env, io);
1197 ll_cl_remove(file, env);
1199 if (args->via_io_subtype == IO_NORMAL)
1200 up_read(&lli->lli_trunc_sem);
1201 if (write_mutex_locked)
1202 mutex_unlock(&lli->lli_write_mutex);
1204 /* cl_io_rw_init() handled IO */
1205 result = io->ci_result;
1208 if (io->ci_nob > 0) {
1209 result = io->ci_nob;
1210 *ppos = io->u.ci_wr.wr.crw_pos;
1214 cl_io_fini(env, io);
1215 /* If any bit been read/written (result != 0), we just return
1216 * short read/write instead of restart io. */
1217 if ((result == 0 || result == -ENODATA) && io->ci_need_restart) {
1218 CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n",
1219 iot == CIT_READ ? "read" : "write",
1220 file->f_dentry->d_name.name, *ppos, count);
1221 LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob);
1225 if (iot == CIT_READ) {
1227 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
1228 LPROC_LL_READ_BYTES, result);
1229 } else if (iot == CIT_WRITE) {
1231 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
1232 LPROC_LL_WRITE_BYTES, result);
1233 fd->fd_write_failed = false;
1234 } else if (result != -ERESTARTSYS) {
1235 fd->fd_write_failed = true;
1238 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1245 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1247 static int ll_file_get_iov_count(const struct iovec *iov,
1248 unsigned long *nr_segs, size_t *count)
1253 for (seg = 0; seg < *nr_segs; seg++) {
1254 const struct iovec *iv = &iov[seg];
1257 * If any segment has a negative length, or the cumulative
1258 * length ever wraps negative then return -EINVAL.
1261 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1263 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1268 cnt -= iv->iov_len; /* This segment is no good */
1275 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1276 unsigned long nr_segs, loff_t pos)
1279 struct vvp_io_args *args;
1285 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1289 env = cl_env_get(&refcheck);
1291 RETURN(PTR_ERR(env));
1293 args = vvp_env_args(env, IO_NORMAL);
1294 args->u.normal.via_iov = (struct iovec *)iov;
1295 args->u.normal.via_nrsegs = nr_segs;
1296 args->u.normal.via_iocb = iocb;
1298 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1299 &iocb->ki_pos, count);
1300 cl_env_put(env, &refcheck);
1304 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1308 struct iovec *local_iov;
1309 struct kiocb *kiocb;
1314 env = cl_env_get(&refcheck);
1316 RETURN(PTR_ERR(env));
1318 local_iov = &vvp_env_info(env)->vti_local_iov;
1319 kiocb = &vvp_env_info(env)->vti_kiocb;
1320 local_iov->iov_base = (void __user *)buf;
1321 local_iov->iov_len = count;
1322 init_sync_kiocb(kiocb, file);
1323 kiocb->ki_pos = *ppos;
1324 #ifdef HAVE_KIOCB_KI_LEFT
1325 kiocb->ki_left = count;
1327 kiocb->ki_nbytes = count;
1330 result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos);
1331 *ppos = kiocb->ki_pos;
1333 cl_env_put(env, &refcheck);
1338 * Write to a file (through the page cache).
1341 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1342 unsigned long nr_segs, loff_t pos)
1345 struct vvp_io_args *args;
1351 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1355 env = cl_env_get(&refcheck);
1357 RETURN(PTR_ERR(env));
1359 args = vvp_env_args(env, IO_NORMAL);
1360 args->u.normal.via_iov = (struct iovec *)iov;
1361 args->u.normal.via_nrsegs = nr_segs;
1362 args->u.normal.via_iocb = iocb;
1364 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1365 &iocb->ki_pos, count);
1366 cl_env_put(env, &refcheck);
1370 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1374 struct iovec *local_iov;
1375 struct kiocb *kiocb;
1380 env = cl_env_get(&refcheck);
1382 RETURN(PTR_ERR(env));
1384 local_iov = &vvp_env_info(env)->vti_local_iov;
1385 kiocb = &vvp_env_info(env)->vti_kiocb;
1386 local_iov->iov_base = (void __user *)buf;
1387 local_iov->iov_len = count;
1388 init_sync_kiocb(kiocb, file);
1389 kiocb->ki_pos = *ppos;
1390 #ifdef HAVE_KIOCB_KI_LEFT
1391 kiocb->ki_left = count;
1393 kiocb->ki_nbytes = count;
1396 result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos);
1397 *ppos = kiocb->ki_pos;
1399 cl_env_put(env, &refcheck);
1404 * Send file content (through pagecache) somewhere with helper
1406 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1407 struct pipe_inode_info *pipe, size_t count,
1411 struct vvp_io_args *args;
1416 env = cl_env_get(&refcheck);
1418 RETURN(PTR_ERR(env));
1420 args = vvp_env_args(env, IO_SPLICE);
1421 args->u.splice.via_pipe = pipe;
1422 args->u.splice.via_flags = flags;
1424 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1425 cl_env_put(env, &refcheck);
1429 static int ll_lov_recreate(struct inode *inode, struct ost_id *oi,
1432 struct obd_export *exp = ll_i2dtexp(inode);
1433 struct obd_trans_info oti = { 0 };
1434 struct obdo *oa = NULL;
1437 struct lov_stripe_md *lsm = NULL, *lsm2;
1444 lsm = ccc_inode_lsm_get(inode);
1445 if (!lsm_has_objects(lsm))
1446 GOTO(out, rc = -ENOENT);
1448 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1449 (lsm->lsm_stripe_count));
1451 OBD_ALLOC_LARGE(lsm2, lsm_size);
1453 GOTO(out, rc = -ENOMEM);
1456 oa->o_nlink = ost_idx;
1457 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1458 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1459 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1460 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1461 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1462 memcpy(lsm2, lsm, lsm_size);
1463 ll_inode_size_lock(inode);
1464 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1465 ll_inode_size_unlock(inode);
1467 OBD_FREE_LARGE(lsm2, lsm_size);
1470 ccc_inode_lsm_put(inode, lsm);
1475 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1477 struct ll_recreate_obj ucreat;
1481 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1484 if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
1488 ostid_set_seq_mdt0(&oi);
1489 ostid_set_id(&oi, ucreat.lrc_id);
1490 RETURN(ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx));
1493 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1500 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1503 if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid)))
1506 fid_to_ostid(&fid, &oi);
1507 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1508 RETURN(ll_lov_recreate(inode, &oi, ost_idx));
1511 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1512 __u64 flags, struct lov_user_md *lum,
1515 struct lov_stripe_md *lsm = NULL;
1516 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1520 lsm = ccc_inode_lsm_get(inode);
1522 ccc_inode_lsm_put(inode, lsm);
1523 CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n",
1524 PFID(ll_inode2fid(inode)));
1525 GOTO(out, rc = -EEXIST);
1528 ll_inode_size_lock(inode);
1529 oit.it_flags |= MDS_OPEN_BY_FID;
1530 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1532 GOTO(out_unlock, rc);
1533 rc = oit.d.lustre.it_status;
1535 GOTO(out_req_free, rc);
1537 ll_release_openhandle(file->f_dentry, &oit);
1540 ll_inode_size_unlock(inode);
1541 ll_intent_release(&oit);
1542 ccc_inode_lsm_put(inode, lsm);
1544 cl_lov_delay_create_clear(&file->f_flags);
1547 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1551 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1552 struct lov_mds_md **lmmp, int *lmm_size,
1553 struct ptlrpc_request **request)
1555 struct ll_sb_info *sbi = ll_i2sbi(inode);
1556 struct mdt_body *body;
1557 struct lov_mds_md *lmm = NULL;
1558 struct ptlrpc_request *req = NULL;
1559 struct md_op_data *op_data;
1562 rc = ll_get_default_mdsize(sbi, &lmmsize);
1566 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1567 strlen(filename), lmmsize,
1568 LUSTRE_OPC_ANY, NULL);
1569 if (IS_ERR(op_data))
1570 RETURN(PTR_ERR(op_data));
1572 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1573 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1574 ll_finish_md_op_data(op_data);
1576 CDEBUG(D_INFO, "md_getattr_name failed "
1577 "on %s: rc %d\n", filename, rc);
1581 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1582 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1584 lmmsize = body->mbo_eadatasize;
1586 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1588 GOTO(out, rc = -ENODATA);
1591 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1592 LASSERT(lmm != NULL);
1594 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1595 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1596 GOTO(out, rc = -EPROTO);
1600 * This is coming from the MDS, so is probably in
1601 * little endian. We convert it to host endian before
1602 * passing it to userspace.
1604 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1607 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1608 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1611 /* if function called for directory - we should
1612 * avoid swab not existent lsm objects */
1613 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1614 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1615 if (S_ISREG(body->mbo_mode))
1616 lustre_swab_lov_user_md_objects(
1617 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1619 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1620 lustre_swab_lov_user_md_v3(
1621 (struct lov_user_md_v3 *)lmm);
1622 if (S_ISREG(body->mbo_mode))
1623 lustre_swab_lov_user_md_objects(
1624 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1631 *lmm_size = lmmsize;
1636 static int ll_lov_setea(struct inode *inode, struct file *file,
1639 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1640 struct lov_user_md *lump;
1641 int lum_size = sizeof(struct lov_user_md) +
1642 sizeof(struct lov_user_ost_data);
1646 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1649 OBD_ALLOC_LARGE(lump, lum_size);
1653 if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) {
1654 OBD_FREE_LARGE(lump, lum_size);
1658 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1660 OBD_FREE_LARGE(lump, lum_size);
1664 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1667 struct lov_user_md_v3 lumv3;
1668 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1669 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1670 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1672 __u64 flags = FMODE_WRITE;
1675 /* first try with v1 which is smaller than v3 */
1676 lum_size = sizeof(struct lov_user_md_v1);
1677 if (copy_from_user(lumv1, lumv1p, lum_size))
1680 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1681 lum_size = sizeof(struct lov_user_md_v3);
1682 if (copy_from_user(&lumv3, lumv3p, lum_size))
1686 rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size);
1688 struct lov_stripe_md *lsm;
1691 put_user(0, &lumv1p->lmm_stripe_count);
1693 ll_layout_refresh(inode, &gen);
1694 lsm = ccc_inode_lsm_get(inode);
1695 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1696 0, lsm, (void *)arg);
1697 ccc_inode_lsm_put(inode, lsm);
1702 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1704 struct lov_stripe_md *lsm;
1708 lsm = ccc_inode_lsm_get(inode);
1710 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1712 ccc_inode_lsm_put(inode, lsm);
1717 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1719 struct ll_inode_info *lli = ll_i2info(inode);
1720 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1721 struct ccc_grouplock grouplock;
1725 if (ll_file_nolock(file))
1726 RETURN(-EOPNOTSUPP);
1728 spin_lock(&lli->lli_lock);
1729 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1730 CWARN("group lock already existed with gid %lu\n",
1731 fd->fd_grouplock.cg_gid);
1732 spin_unlock(&lli->lli_lock);
1735 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1736 spin_unlock(&lli->lli_lock);
1738 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1739 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1743 spin_lock(&lli->lli_lock);
1744 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1745 spin_unlock(&lli->lli_lock);
1746 CERROR("another thread just won the race\n");
1747 cl_put_grouplock(&grouplock);
1751 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1752 fd->fd_grouplock = grouplock;
1753 spin_unlock(&lli->lli_lock);
1755 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1759 int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1761 struct ll_inode_info *lli = ll_i2info(inode);
1762 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1763 struct ccc_grouplock grouplock;
1766 spin_lock(&lli->lli_lock);
1767 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1768 spin_unlock(&lli->lli_lock);
1769 CWARN("no group lock held\n");
1772 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1774 if (fd->fd_grouplock.cg_gid != arg) {
1775 CWARN("group lock %lu doesn't match current id %lu\n",
1776 arg, fd->fd_grouplock.cg_gid);
1777 spin_unlock(&lli->lli_lock);
1781 grouplock = fd->fd_grouplock;
1782 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1783 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1784 spin_unlock(&lli->lli_lock);
1786 cl_put_grouplock(&grouplock);
1787 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1792 * Close inode open handle
1794 * \param dentry [in] dentry which contains the inode
1795 * \param it [in,out] intent which contains open info and result
1798 * \retval <0 failure
1800 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1802 struct inode *inode = dentry->d_inode;
1803 struct obd_client_handle *och;
1809 /* Root ? Do nothing. */
1810 if (dentry->d_inode->i_sb->s_root == dentry)
1813 /* No open handle to close? Move away */
1814 if (!it_disposition(it, DISP_OPEN_OPEN))
1817 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1819 OBD_ALLOC(och, sizeof(*och));
1821 GOTO(out, rc = -ENOMEM);
1823 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1825 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1828 /* this one is in place of ll_file_open */
1829 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1830 ptlrpc_req_finished(it->d.lustre.it_data);
1831 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1837 * Get size for inode for which FIEMAP mapping is requested.
1838 * Make the FIEMAP get_info call and returns the result.
1840 static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1843 struct obd_export *exp = ll_i2dtexp(inode);
1844 struct lov_stripe_md *lsm = NULL;
1845 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1846 __u32 vallen = num_bytes;
1850 /* Checks for fiemap flags */
1851 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1852 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1856 /* Check for FIEMAP_FLAG_SYNC */
1857 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1858 rc = filemap_fdatawrite(inode->i_mapping);
1863 lsm = ccc_inode_lsm_get(inode);
1867 /* If the stripe_count > 1 and the application does not understand
1868 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1870 if (lsm->lsm_stripe_count > 1 &&
1871 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER))
1872 GOTO(out, rc = -EOPNOTSUPP);
1874 fm_key.oa.o_oi = lsm->lsm_oi;
1875 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1877 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1878 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1879 /* If filesize is 0, then there would be no objects for mapping */
1880 if (fm_key.oa.o_size == 0) {
1881 fiemap->fm_mapped_extents = 0;
1885 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1887 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1890 CERROR("obd_get_info failed: rc = %d\n", rc);
1893 ccc_inode_lsm_put(inode, lsm);
1897 int ll_fid2path(struct inode *inode, void __user *arg)
1899 struct obd_export *exp = ll_i2mdexp(inode);
1900 const struct getinfo_fid2path __user *gfin = arg;
1902 struct getinfo_fid2path *gfout;
1908 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1909 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1912 /* Only need to get the buflen */
1913 if (get_user(pathlen, &gfin->gf_pathlen))
1916 outsize = sizeof(*gfout) + pathlen;
1917 OBD_ALLOC(gfout, outsize);
1921 if (copy_from_user(gfout, arg, sizeof(*gfout)))
1922 GOTO(gf_free, rc = -EFAULT);
1924 /* Call mdc_iocontrol */
1925 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1929 if (copy_to_user(arg, gfout, outsize))
1933 OBD_FREE(gfout, outsize);
1937 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1939 struct ll_user_fiemap *fiemap_s;
1940 size_t num_bytes, ret_bytes;
1941 unsigned int extent_count;
1944 /* Get the extent count so we can calculate the size of
1945 * required fiemap buffer */
1946 if (get_user(extent_count,
1947 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1949 num_bytes = sizeof(*fiemap_s) + (extent_count *
1950 sizeof(struct ll_fiemap_extent));
1952 OBD_ALLOC_LARGE(fiemap_s, num_bytes);
1953 if (fiemap_s == NULL)
1956 /* get the fiemap value */
1957 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1959 GOTO(error, rc = -EFAULT);
1961 /* If fm_extent_count is non-zero, read the first extent since
1962 * it is used to calculate end_offset and device from previous
1965 if (copy_from_user(&fiemap_s->fm_extents[0],
1966 (char __user *)arg + sizeof(*fiemap_s),
1967 sizeof(struct ll_fiemap_extent)))
1968 GOTO(error, rc = -EFAULT);
1971 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1975 ret_bytes = sizeof(struct ll_user_fiemap);
1977 if (extent_count != 0)
1978 ret_bytes += (fiemap_s->fm_mapped_extents *
1979 sizeof(struct ll_fiemap_extent));
1981 if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
1985 OBD_FREE_LARGE(fiemap_s, num_bytes);
1990 * Read the data_version for inode.
1992 * This value is computed using stripe object version on OST.
1993 * Version is computed using server side locking.
1995 * @param sync if do sync on the OST side;
1997 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1998 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2000 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2002 struct lov_stripe_md *lsm = NULL;
2003 struct ll_sb_info *sbi = ll_i2sbi(inode);
2004 struct obdo *obdo = NULL;
2008 /* If no stripe, we consider version is 0. */
2009 lsm = ccc_inode_lsm_get(inode);
2010 if (!lsm_has_objects(lsm)) {
2012 CDEBUG(D_INODE, "No object for inode\n");
2016 OBD_ALLOC_PTR(obdo);
2018 GOTO(out, rc = -ENOMEM);
2020 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, flags);
2022 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
2025 *data_version = obdo->o_data_version;
2031 ccc_inode_lsm_put(inode, lsm);
2036 * Trigger a HSM release request for the provided inode.
2038 int ll_hsm_release(struct inode *inode)
2040 struct cl_env_nest nest;
2042 struct obd_client_handle *och = NULL;
2043 __u64 data_version = 0;
2047 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2048 ll_get_fsname(inode->i_sb, NULL, 0),
2049 PFID(&ll_i2info(inode)->lli_fid));
2051 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2053 GOTO(out, rc = PTR_ERR(och));
2055 /* Grab latest data_version and [am]time values */
2056 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2060 env = cl_env_nested_get(&nest);
2062 GOTO(out, rc = PTR_ERR(env));
2064 ll_merge_lvb(env, inode);
2065 cl_env_nested_put(&nest, env);
2067 /* Release the file.
2068 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2069 * we still need it to pack l_remote_handle to MDT. */
2070 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
2076 if (och != NULL && !IS_ERR(och)) /* close the file */
2077 ll_lease_close(och, inode, NULL);
2082 struct ll_swap_stack {
2083 struct iattr ia1, ia2;
2085 struct inode *inode1, *inode2;
2086 bool check_dv1, check_dv2;
2089 static int ll_swap_layouts(struct file *file1, struct file *file2,
2090 struct lustre_swap_layouts *lsl)
2092 struct mdc_swap_layouts msl;
2093 struct md_op_data *op_data;
2096 struct ll_swap_stack *llss = NULL;
2099 OBD_ALLOC_PTR(llss);
2103 llss->inode1 = file1->f_dentry->d_inode;
2104 llss->inode2 = file2->f_dentry->d_inode;
2106 if (!S_ISREG(llss->inode2->i_mode))
2107 GOTO(free, rc = -EINVAL);
2109 if (inode_permission(llss->inode1, MAY_WRITE) ||
2110 inode_permission(llss->inode2, MAY_WRITE))
2111 GOTO(free, rc = -EPERM);
2113 if (llss->inode2->i_sb != llss->inode1->i_sb)
2114 GOTO(free, rc = -EXDEV);
2116 /* we use 2 bool because it is easier to swap than 2 bits */
2117 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2118 llss->check_dv1 = true;
2120 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2121 llss->check_dv2 = true;
2123 /* we cannot use lsl->sl_dvX directly because we may swap them */
2124 llss->dv1 = lsl->sl_dv1;
2125 llss->dv2 = lsl->sl_dv2;
2127 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2128 if (rc == 0) /* same file, done! */
2131 if (rc < 0) { /* sequentialize it */
2132 swap(llss->inode1, llss->inode2);
2134 swap(llss->dv1, llss->dv2);
2135 swap(llss->check_dv1, llss->check_dv2);
2139 if (gid != 0) { /* application asks to flush dirty cache */
2140 rc = ll_get_grouplock(llss->inode1, file1, gid);
2144 rc = ll_get_grouplock(llss->inode2, file2, gid);
2146 ll_put_grouplock(llss->inode1, file1, gid);
2151 /* to be able to restore mtime and atime after swap
2152 * we need to first save them */
2154 (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
2155 llss->ia1.ia_mtime = llss->inode1->i_mtime;
2156 llss->ia1.ia_atime = llss->inode1->i_atime;
2157 llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
2158 llss->ia2.ia_mtime = llss->inode2->i_mtime;
2159 llss->ia2.ia_atime = llss->inode2->i_atime;
2160 llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
2163 /* ultimate check, before swaping the layouts we check if
2164 * dataversion has changed (if requested) */
2165 if (llss->check_dv1) {
2166 rc = ll_data_version(llss->inode1, &dv, 0);
2169 if (dv != llss->dv1)
2170 GOTO(putgl, rc = -EAGAIN);
2173 if (llss->check_dv2) {
2174 rc = ll_data_version(llss->inode2, &dv, 0);
2177 if (dv != llss->dv2)
2178 GOTO(putgl, rc = -EAGAIN);
2181 /* struct md_op_data is used to send the swap args to the mdt
2182 * only flags is missing, so we use struct mdc_swap_layouts
2183 * through the md_op_data->op_data */
2184 /* flags from user space have to be converted before they are send to
2185 * server, no flag is sent today, they are only used on the client */
2188 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2189 0, LUSTRE_OPC_ANY, &msl);
2190 if (IS_ERR(op_data))
2191 GOTO(free, rc = PTR_ERR(op_data));
2193 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2194 sizeof(*op_data), op_data, NULL);
2195 ll_finish_md_op_data(op_data);
2199 ll_put_grouplock(llss->inode2, file2, gid);
2200 ll_put_grouplock(llss->inode1, file1, gid);
2203 /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
2207 /* clear useless flags */
2208 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
2209 llss->ia1.ia_valid &= ~ATTR_MTIME;
2210 llss->ia2.ia_valid &= ~ATTR_MTIME;
2213 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
2214 llss->ia1.ia_valid &= ~ATTR_ATIME;
2215 llss->ia2.ia_valid &= ~ATTR_ATIME;
2218 /* update time if requested */
2220 if (llss->ia2.ia_valid != 0) {
2221 mutex_lock(&llss->inode1->i_mutex);
2222 rc = ll_setattr(file1->f_dentry, &llss->ia2);
2223 mutex_unlock(&llss->inode1->i_mutex);
2226 if (llss->ia1.ia_valid != 0) {
2229 mutex_lock(&llss->inode2->i_mutex);
2230 rc1 = ll_setattr(file2->f_dentry, &llss->ia1);
2231 mutex_unlock(&llss->inode2->i_mutex);
2243 static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2245 struct md_op_data *op_data;
2248 /* Non-root users are forbidden to set or clear flags which are
2249 * NOT defined in HSM_USER_MASK. */
2250 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2251 !cfs_capable(CFS_CAP_SYS_ADMIN))
2254 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2255 LUSTRE_OPC_ANY, hss);
2256 if (IS_ERR(op_data))
2257 RETURN(PTR_ERR(op_data));
2259 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2260 sizeof(*op_data), op_data, NULL);
2262 ll_finish_md_op_data(op_data);
2267 static int ll_hsm_import(struct inode *inode, struct file *file,
2268 struct hsm_user_import *hui)
2270 struct hsm_state_set *hss = NULL;
2271 struct iattr *attr = NULL;
2275 if (!S_ISREG(inode->i_mode))
2281 GOTO(out, rc = -ENOMEM);
2283 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2284 hss->hss_archive_id = hui->hui_archive_id;
2285 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2286 rc = ll_hsm_state_set(inode, hss);
2290 OBD_ALLOC_PTR(attr);
2292 GOTO(out, rc = -ENOMEM);
2294 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2295 attr->ia_mode |= S_IFREG;
2296 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2297 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2298 attr->ia_size = hui->hui_size;
2299 attr->ia_mtime.tv_sec = hui->hui_mtime;
2300 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2301 attr->ia_atime.tv_sec = hui->hui_atime;
2302 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2304 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2305 ATTR_UID | ATTR_GID |
2306 ATTR_MTIME | ATTR_MTIME_SET |
2307 ATTR_ATIME | ATTR_ATIME_SET;
2309 rc = ll_setattr_raw(file->f_dentry, attr, true);
2324 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2326 struct inode *inode = file->f_dentry->d_inode;
2327 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2331 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
2332 PFID(ll_inode2fid(inode)), inode, cmd);
2333 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2335 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2336 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2340 case LL_IOC_GETFLAGS:
2341 /* Get the current value of the file flags */
2342 return put_user(fd->fd_flags, (int *)arg);
2343 case LL_IOC_SETFLAGS:
2344 case LL_IOC_CLRFLAGS:
2345 /* Set or clear specific file flags */
2346 /* XXX This probably needs checks to ensure the flags are
2347 * not abused, and to handle any flag side effects.
2349 if (get_user(flags, (int *) arg))
2352 if (cmd == LL_IOC_SETFLAGS) {
2353 if ((flags & LL_FILE_IGNORE_LOCK) &&
2354 !(file->f_flags & O_DIRECT)) {
2355 CERROR("%s: unable to disable locking on "
2356 "non-O_DIRECT file\n", current->comm);
2360 fd->fd_flags |= flags;
2362 fd->fd_flags &= ~flags;
2365 case LL_IOC_LOV_SETSTRIPE:
2366 RETURN(ll_lov_setstripe(inode, file, arg));
2367 case LL_IOC_LOV_SETEA:
2368 RETURN(ll_lov_setea(inode, file, arg));
2369 case LL_IOC_LOV_SWAP_LAYOUTS: {
2371 struct lustre_swap_layouts lsl;
2373 if (copy_from_user(&lsl, (char *)arg,
2374 sizeof(struct lustre_swap_layouts)))
2377 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
2380 file2 = fget(lsl.sl_fd);
2385 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
2386 rc = ll_swap_layouts(file, file2, &lsl);
2390 case LL_IOC_LOV_GETSTRIPE:
2391 RETURN(ll_lov_getstripe(inode, arg));
2392 case LL_IOC_RECREATE_OBJ:
2393 RETURN(ll_lov_recreate_obj(inode, arg));
2394 case LL_IOC_RECREATE_FID:
2395 RETURN(ll_lov_recreate_fid(inode, arg));
2396 case FSFILT_IOC_FIEMAP:
2397 RETURN(ll_ioctl_fiemap(inode, arg));
2398 case FSFILT_IOC_GETFLAGS:
2399 case FSFILT_IOC_SETFLAGS:
2400 RETURN(ll_iocontrol(inode, file, cmd, arg));
2401 case FSFILT_IOC_GETVERSION_OLD:
2402 case FSFILT_IOC_GETVERSION:
2403 RETURN(put_user(inode->i_generation, (int *)arg));
2404 case LL_IOC_GROUP_LOCK:
2405 RETURN(ll_get_grouplock(inode, file, arg));
2406 case LL_IOC_GROUP_UNLOCK:
2407 RETURN(ll_put_grouplock(inode, file, arg));
2408 case IOC_OBD_STATFS:
2409 RETURN(ll_obd_statfs(inode, (void *)arg));
2411 /* We need to special case any other ioctls we want to handle,
2412 * to send them to the MDS/OST as appropriate and to properly
2413 * network encode the arg field.
2414 case FSFILT_IOC_SETVERSION_OLD:
2415 case FSFILT_IOC_SETVERSION:
2417 case LL_IOC_FLUSHCTX:
2418 RETURN(ll_flush_ctx(inode));
2419 case LL_IOC_PATH2FID: {
2420 if (copy_to_user((void *)arg, ll_inode2fid(inode),
2421 sizeof(struct lu_fid)))
2426 case OBD_IOC_FID2PATH:
2427 RETURN(ll_fid2path(inode, (void *)arg));
2428 case LL_IOC_DATA_VERSION: {
2429 struct ioc_data_version idv;
2432 if (copy_from_user(&idv, (char *)arg, sizeof(idv)))
2435 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2436 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2438 if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv)))
2444 case LL_IOC_GET_MDTIDX: {
2447 mdtidx = ll_get_mdt_idx(inode);
2451 if (put_user((int)mdtidx, (int*)arg))
2456 case OBD_IOC_GETDTNAME:
2457 case OBD_IOC_GETMDNAME:
2458 RETURN(ll_get_obd_name(inode, cmd, arg));
2459 case LL_IOC_HSM_STATE_GET: {
2460 struct md_op_data *op_data;
2461 struct hsm_user_state *hus;
2468 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2469 LUSTRE_OPC_ANY, hus);
2470 if (IS_ERR(op_data)) {
2472 RETURN(PTR_ERR(op_data));
2475 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2478 if (copy_to_user((void *)arg, hus, sizeof(*hus)))
2481 ll_finish_md_op_data(op_data);
2485 case LL_IOC_HSM_STATE_SET: {
2486 struct hsm_state_set *hss;
2493 if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
2498 rc = ll_hsm_state_set(inode, hss);
2503 case LL_IOC_HSM_ACTION: {
2504 struct md_op_data *op_data;
2505 struct hsm_current_action *hca;
2512 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2513 LUSTRE_OPC_ANY, hca);
2514 if (IS_ERR(op_data)) {
2516 RETURN(PTR_ERR(op_data));
2519 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2522 if (copy_to_user((char *)arg, hca, sizeof(*hca)))
2525 ll_finish_md_op_data(op_data);
2529 case LL_IOC_SET_LEASE: {
2530 struct ll_inode_info *lli = ll_i2info(inode);
2531 struct obd_client_handle *och = NULL;
2537 if (!(file->f_mode & FMODE_WRITE))
2542 if (!(file->f_mode & FMODE_READ))
2547 mutex_lock(&lli->lli_och_mutex);
2548 if (fd->fd_lease_och != NULL) {
2549 och = fd->fd_lease_och;
2550 fd->fd_lease_och = NULL;
2552 mutex_unlock(&lli->lli_och_mutex);
2555 mode = och->och_flags &(FMODE_READ|FMODE_WRITE);
2556 rc = ll_lease_close(och, inode, &lease_broken);
2557 if (rc == 0 && lease_broken)
2563 /* return the type of lease or error */
2564 RETURN(rc < 0 ? rc : (int)mode);
2569 CDEBUG(D_INODE, "Set lease with mode %d\n", mode);
2571 /* apply for lease */
2572 och = ll_lease_open(inode, file, mode, 0);
2574 RETURN(PTR_ERR(och));
2577 mutex_lock(&lli->lli_och_mutex);
2578 if (fd->fd_lease_och == NULL) {
2579 fd->fd_lease_och = och;
2582 mutex_unlock(&lli->lli_och_mutex);
2584 /* impossible now that only excl is supported for now */
2585 ll_lease_close(och, inode, &lease_broken);
2590 case LL_IOC_GET_LEASE: {
2591 struct ll_inode_info *lli = ll_i2info(inode);
2592 struct ldlm_lock *lock = NULL;
2595 mutex_lock(&lli->lli_och_mutex);
2596 if (fd->fd_lease_och != NULL) {
2597 struct obd_client_handle *och = fd->fd_lease_och;
2599 lock = ldlm_handle2lock(&och->och_lease_handle);
2601 lock_res_and_lock(lock);
2602 if (!ldlm_is_cancel(lock))
2603 rc = och->och_flags &
2604 (FMODE_READ | FMODE_WRITE);
2605 unlock_res_and_lock(lock);
2606 LDLM_LOCK_PUT(lock);
2609 mutex_unlock(&lli->lli_och_mutex);
2612 case LL_IOC_HSM_IMPORT: {
2613 struct hsm_user_import *hui;
2619 if (copy_from_user(hui, (void *)arg, sizeof(*hui))) {
2624 rc = ll_hsm_import(inode, file, hui);
2634 ll_iocontrol_call(inode, file, cmd, arg, &err))
2637 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2643 #ifndef HAVE_FILE_LLSEEK_SIZE
2644 static inline loff_t
2645 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2647 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2649 if (offset > maxsize)
2652 if (offset != file->f_pos) {
2653 file->f_pos = offset;
2654 file->f_version = 0;
2660 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2661 loff_t maxsize, loff_t eof)
2663 struct inode *inode = file->f_dentry->d_inode;
2671 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2672 * position-querying operation. Avoid rewriting the "same"
2673 * f_pos value back to the file because a concurrent read(),
2674 * write() or lseek() might have altered it
2679 * f_lock protects against read/modify/write race with other
2680 * SEEK_CURs. Note that parallel writes and reads behave
2683 mutex_lock(&inode->i_mutex);
2684 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2685 mutex_unlock(&inode->i_mutex);
2689 * In the generic case the entire file is data, so as long as
2690 * offset isn't at the end of the file then the offset is data.
2697 * There is a virtual hole at the end of the file, so as long as
2698 * offset isn't i_size or larger, return i_size.
2706 return llseek_execute(file, offset, maxsize);
2710 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2712 struct inode *inode = file->f_dentry->d_inode;
2713 loff_t retval, eof = 0;
2716 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2717 (origin == SEEK_CUR) ? file->f_pos : 0);
2718 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2719 PFID(ll_inode2fid(inode)), inode, retval, retval,
2721 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2723 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2724 retval = ll_glimpse_size(inode);
2727 eof = i_size_read(inode);
2730 retval = ll_generic_file_llseek_size(file, offset, origin,
2731 ll_file_maxbytes(inode), eof);
2735 static int ll_flush(struct file *file, fl_owner_t id)
2737 struct inode *inode = file->f_dentry->d_inode;
2738 struct ll_inode_info *lli = ll_i2info(inode);
2739 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2742 LASSERT(!S_ISDIR(inode->i_mode));
2744 /* catch async errors that were recorded back when async writeback
2745 * failed for pages in this mapping. */
2746 rc = lli->lli_async_rc;
2747 lli->lli_async_rc = 0;
2748 if (lli->lli_clob != NULL) {
2749 err = lov_read_and_clear_async_rc(lli->lli_clob);
2754 /* The application has been told write failure already.
2755 * Do not report failure again. */
2756 if (fd->fd_write_failed)
2758 return rc ? -EIO : 0;
2762 * Called to make sure a portion of file has been written out.
2763 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2765 * Return how many pages have been written.
2767 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2768 enum cl_fsync_mode mode, int ignore_layout)
2770 struct cl_env_nest nest;
2773 struct obd_capa *capa = NULL;
2774 struct cl_fsync_io *fio;
2778 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2779 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2782 env = cl_env_nested_get(&nest);
2784 RETURN(PTR_ERR(env));
2786 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
2788 io = ccc_env_thread_io(env);
2789 io->ci_obj = cl_i2info(inode)->lli_clob;
2790 io->ci_ignore_layout = ignore_layout;
2792 /* initialize parameters for sync */
2793 fio = &io->u.ci_fsync;
2794 fio->fi_capa = capa;
2795 fio->fi_start = start;
2797 fio->fi_fid = ll_inode2fid(inode);
2798 fio->fi_mode = mode;
2799 fio->fi_nr_written = 0;
2801 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2802 result = cl_io_loop(env, io);
2804 result = io->ci_result;
2806 result = fio->fi_nr_written;
2807 cl_io_fini(env, io);
2808 cl_env_nested_put(&nest, env);
2816 * When dentry is provided (the 'else' case), *file->f_dentry may be
2817 * null and dentry must be used directly rather than pulled from
2818 * *file->f_dentry as is done otherwise.
2821 #ifdef HAVE_FILE_FSYNC_4ARGS
2822 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2824 struct dentry *dentry = file->f_dentry;
2825 #elif defined(HAVE_FILE_FSYNC_2ARGS)
2826 int ll_fsync(struct file *file, int datasync)
2828 struct dentry *dentry = file->f_dentry;
2830 loff_t end = LLONG_MAX;
2832 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
2835 loff_t end = LLONG_MAX;
2837 struct inode *inode = dentry->d_inode;
2838 struct ll_inode_info *lli = ll_i2info(inode);
2839 struct ptlrpc_request *req;
2840 struct obd_capa *oc;
2844 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2845 PFID(ll_inode2fid(inode)), inode);
2846 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2848 #ifdef HAVE_FILE_FSYNC_4ARGS
2849 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2850 mutex_lock(&inode->i_mutex);
2852 /* fsync's caller has already called _fdata{sync,write}, we want
2853 * that IO to finish before calling the osc and mdc sync methods */
2854 rc = filemap_fdatawait(inode->i_mapping);
2857 /* catch async errors that were recorded back when async writeback
2858 * failed for pages in this mapping. */
2859 if (!S_ISDIR(inode->i_mode)) {
2860 err = lli->lli_async_rc;
2861 lli->lli_async_rc = 0;
2864 err = lov_read_and_clear_async_rc(lli->lli_clob);
2869 oc = ll_mdscapa_get(inode);
2870 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2876 ptlrpc_req_finished(req);
2878 if (S_ISREG(inode->i_mode)) {
2879 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2881 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2882 if (rc == 0 && err < 0)
2885 fd->fd_write_failed = true;
2887 fd->fd_write_failed = false;
2890 #ifdef HAVE_FILE_FSYNC_4ARGS
2891 mutex_unlock(&inode->i_mutex);
2897 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2899 struct inode *inode = file->f_dentry->d_inode;
2900 struct ll_sb_info *sbi = ll_i2sbi(inode);
2901 struct ldlm_enqueue_info einfo = {
2902 .ei_type = LDLM_FLOCK,
2903 .ei_cb_cp = ldlm_flock_completion_ast,
2904 .ei_cbdata = file_lock,
2906 struct md_op_data *op_data;
2907 struct lustre_handle lockh = {0};
2908 ldlm_policy_data_t flock = {{0}};
2909 int fl_type = file_lock->fl_type;
2915 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2916 PFID(ll_inode2fid(inode)), file_lock);
2918 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2920 if (file_lock->fl_flags & FL_FLOCK) {
2921 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2922 /* flocks are whole-file locks */
2923 flock.l_flock.end = OFFSET_MAX;
2924 /* For flocks owner is determined by the local file desctiptor*/
2925 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
2926 } else if (file_lock->fl_flags & FL_POSIX) {
2927 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2928 flock.l_flock.start = file_lock->fl_start;
2929 flock.l_flock.end = file_lock->fl_end;
2933 flock.l_flock.pid = file_lock->fl_pid;
2935 /* Somewhat ugly workaround for svc lockd.
2936 * lockd installs custom fl_lmops->lm_compare_owner that checks
2937 * for the fl_owner to be the same (which it always is on local node
2938 * I guess between lockd processes) and then compares pid.
2939 * As such we assign pid to the owner field to make it all work,
2940 * conflict with normal locks is unlikely since pid space and
2941 * pointer space for current->files are not intersecting */
2942 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2943 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2947 einfo.ei_mode = LCK_PR;
2950 /* An unlock request may or may not have any relation to
2951 * existing locks so we may not be able to pass a lock handle
2952 * via a normal ldlm_lock_cancel() request. The request may even
2953 * unlock a byte range in the middle of an existing lock. In
2954 * order to process an unlock request we need all of the same
2955 * information that is given with a normal read or write record
2956 * lock request. To avoid creating another ldlm unlock (cancel)
2957 * message we'll treat a LCK_NL flock request as an unlock. */
2958 einfo.ei_mode = LCK_NL;
2961 einfo.ei_mode = LCK_PW;
2964 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
2979 flags = LDLM_FL_BLOCK_NOWAIT;
2985 flags = LDLM_FL_TEST_LOCK;
2988 CERROR("unknown fcntl lock command: %d\n", cmd);
2992 /* Save the old mode so that if the mode in the lock changes we
2993 * can decrement the appropriate reader or writer refcount. */
2994 file_lock->fl_type = einfo.ei_mode;
2996 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2997 LUSTRE_OPC_ANY, NULL);
2998 if (IS_ERR(op_data))
2999 RETURN(PTR_ERR(op_data));
3001 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags="LPX64", mode=%u, "
3002 "start="LPU64", end="LPU64"\n", PFID(ll_inode2fid(inode)),
3003 flock.l_flock.pid, flags, einfo.ei_mode,
3004 flock.l_flock.start, flock.l_flock.end);
3006 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
3009 /* Restore the file lock type if not TEST lock. */
3010 if (!(flags & LDLM_FL_TEST_LOCK))
3011 file_lock->fl_type = fl_type;
3013 if ((file_lock->fl_flags & FL_FLOCK) &&
3014 (rc == 0 || file_lock->fl_type == F_UNLCK))
3015 rc2 = flock_lock_file_wait(file, file_lock);
3016 if ((file_lock->fl_flags & FL_POSIX) &&
3017 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
3018 !(flags & LDLM_FL_TEST_LOCK))
3019 rc2 = posix_lock_file_wait(file, file_lock);
3021 if (rc2 && file_lock->fl_type != F_UNLCK) {
3022 einfo.ei_mode = LCK_NL;
3023 md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
3028 ll_finish_md_op_data(op_data);
3033 int ll_get_fid_by_name(struct inode *parent, const char *name,
3034 int namelen, struct lu_fid *fid)
3036 struct md_op_data *op_data = NULL;
3037 struct mdt_body *body;
3038 struct ptlrpc_request *req;
3042 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
3043 LUSTRE_OPC_ANY, NULL);
3044 if (IS_ERR(op_data))
3045 RETURN(PTR_ERR(op_data));
3047 op_data->op_valid = OBD_MD_FLID;
3048 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
3049 ll_finish_md_op_data(op_data);
3053 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3055 GOTO(out_req, rc = -EFAULT);
3057 *fid = body->mbo_fid1;
3059 ptlrpc_req_finished(req);
3063 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
3064 const char *name, int namelen)
3066 struct dentry *dchild = NULL;
3067 struct inode *child_inode = NULL;
3068 struct md_op_data *op_data;
3069 struct ptlrpc_request *request = NULL;
3074 CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%04x\n",
3075 name, PFID(ll_inode2fid(parent)), mdtidx);
3077 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
3078 0, LUSTRE_OPC_ANY, NULL);
3079 if (IS_ERR(op_data))
3080 RETURN(PTR_ERR(op_data));
3082 /* Get child FID first */
3083 qstr.hash = full_name_hash(name, namelen);
3086 dchild = d_lookup(file->f_dentry, &qstr);
3087 if (dchild != NULL && dchild->d_inode != NULL) {
3088 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
3089 if (dchild->d_inode != NULL) {
3090 child_inode = igrab(dchild->d_inode);
3091 ll_invalidate_aliases(child_inode);
3095 rc = ll_get_fid_by_name(parent, name, namelen,
3101 if (!fid_is_sane(&op_data->op_fid3)) {
3102 CERROR("%s: migrate %s , but fid "DFID" is insane\n",
3103 ll_get_fsname(parent->i_sb, NULL, 0), name,
3104 PFID(&op_data->op_fid3));
3105 GOTO(out_free, rc = -EINVAL);
3108 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
3113 CDEBUG(D_INFO, "%s:"DFID" is already on MDT%d.\n", name,
3114 PFID(&op_data->op_fid3), mdtidx);
3115 GOTO(out_free, rc = 0);
3118 op_data->op_mds = mdtidx;
3119 op_data->op_cli_flags = CLI_MIGRATE;
3120 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
3121 namelen, name, namelen, &request);
3123 ll_update_times(request, parent);
3125 ptlrpc_req_finished(request);
3130 if (child_inode != NULL) {
3131 clear_nlink(child_inode);
3135 ll_finish_md_op_data(op_data);
3140 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
3148 * test if some locks matching bits and l_req_mode are acquired
3149 * - bits can be in different locks
3150 * - if found clear the common lock bits in *bits
3151 * - the bits not found, are kept in *bits
3153 * \param bits [IN] searched lock bits [IN]
3154 * \param l_req_mode [IN] searched lock mode
3155 * \retval boolean, true iff all bits are found
3157 int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode)
3159 struct lustre_handle lockh;
3160 ldlm_policy_data_t policy;
3161 ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
3162 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
3171 fid = &ll_i2info(inode)->lli_fid;
3172 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
3173 ldlm_lockname[mode]);
3175 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
3176 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
3177 policy.l_inodebits.bits = *bits & (1 << i);
3178 if (policy.l_inodebits.bits == 0)
3181 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
3182 &policy, mode, &lockh)) {
3183 struct ldlm_lock *lock;
3185 lock = ldlm_handle2lock(&lockh);
3188 ~(lock->l_policy_data.l_inodebits.bits);
3189 LDLM_LOCK_PUT(lock);
3191 *bits &= ~policy.l_inodebits.bits;
3198 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
3199 struct lustre_handle *lockh, __u64 flags,
3202 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
3207 fid = &ll_i2info(inode)->lli_fid;
3208 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
3210 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
3211 fid, LDLM_IBITS, &policy, mode, lockh);
3216 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
3218 /* Already unlinked. Just update nlink and return success */
3219 if (rc == -ENOENT) {
3221 /* This path cannot be hit for regular files unless in
3222 * case of obscure races, so no need to to validate
3224 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
3226 } else if (rc != 0) {
3227 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
3228 "%s: revalidate FID "DFID" error: rc = %d\n",
3229 ll_get_fsname(inode->i_sb, NULL, 0),
3230 PFID(ll_inode2fid(inode)), rc);
3236 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3238 struct inode *inode = dentry->d_inode;
3239 struct ptlrpc_request *req = NULL;
3240 struct obd_export *exp;
3244 LASSERT(inode != NULL);
3246 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
3247 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
3249 exp = ll_i2mdexp(inode);
3251 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
3252 * But under CMD case, it caused some lock issues, should be fixed
3253 * with new CMD ibits lock. See bug 12718 */
3254 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
3255 struct lookup_intent oit = { .it_op = IT_GETATTR };
3256 struct md_op_data *op_data;
3258 if (ibits == MDS_INODELOCK_LOOKUP)
3259 oit.it_op = IT_LOOKUP;
3261 /* Call getattr by fid, so do not provide name at all. */
3262 op_data = ll_prep_md_op_data(NULL, dentry->d_inode,
3263 dentry->d_inode, NULL, 0, 0,
3264 LUSTRE_OPC_ANY, NULL);
3265 if (IS_ERR(op_data))
3266 RETURN(PTR_ERR(op_data));
3268 rc = md_intent_lock(exp, op_data, &oit, &req,
3269 &ll_md_blocking_ast, 0);
3270 ll_finish_md_op_data(op_data);
3272 rc = ll_inode_revalidate_fini(inode, rc);
3276 rc = ll_revalidate_it_finish(req, &oit, dentry);
3278 ll_intent_release(&oit);
3282 /* Unlinked? Unhash dentry, so it is not picked up later by
3283 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
3284 here to preserve get_cwd functionality on 2.6.
3286 if (!dentry->d_inode->i_nlink)
3287 d_lustre_invalidate(dentry, 0);
3289 ll_lookup_finish_locks(&oit, dentry);
3290 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
3291 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
3292 obd_valid valid = OBD_MD_FLGETATTR;
3293 struct md_op_data *op_data;
3296 if (S_ISREG(inode->i_mode)) {
3297 rc = ll_get_default_mdsize(sbi, &ealen);
3300 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
3303 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
3304 0, ealen, LUSTRE_OPC_ANY,
3306 if (IS_ERR(op_data))
3307 RETURN(PTR_ERR(op_data));
3309 op_data->op_valid = valid;
3310 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
3311 * capa for this inode. Because we only keep capas of dirs
3313 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3314 ll_finish_md_op_data(op_data);
3316 rc = ll_inode_revalidate_fini(inode, rc);
3320 rc = ll_prep_inode(&inode, req, NULL, NULL);
3323 ptlrpc_req_finished(req);
3327 static int ll_merge_md_attr(struct inode *inode)
3329 struct cl_attr attr = { 0 };
3332 LASSERT(ll_i2info(inode)->lli_lsm_md != NULL);
3333 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
3338 ll_i2info(inode)->lli_stripe_dir_size = attr.cat_size;
3339 ll_i2info(inode)->lli_stripe_dir_nlink = attr.cat_nlink;
3341 ll_i2info(inode)->lli_lvb.lvb_atime = attr.cat_atime;
3342 ll_i2info(inode)->lli_lvb.lvb_mtime = attr.cat_mtime;
3343 ll_i2info(inode)->lli_lvb.lvb_ctime = attr.cat_ctime;
3349 ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3351 struct inode *inode = dentry->d_inode;
3355 rc = __ll_inode_revalidate(dentry, ibits);
3359 /* if object isn't regular file, don't validate size */
3360 if (!S_ISREG(inode->i_mode)) {
3361 if (S_ISDIR(inode->i_mode) &&
3362 ll_i2info(inode)->lli_lsm_md != NULL) {
3363 rc = ll_merge_md_attr(inode);
3368 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
3369 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
3370 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
3372 /* In case of restore, the MDT has the right size and has
3373 * already send it back without granting the layout lock,
3374 * inode is up-to-date so glimpse is useless.
3375 * Also to glimpse we need the layout, in case of a running
3376 * restore the MDT holds the layout lock so the glimpse will
3377 * block up to the end of restore (getattr will block)
3379 if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING))
3380 rc = ll_glimpse_size(inode);
3385 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3387 struct inode *inode = de->d_inode;
3388 struct ll_sb_info *sbi = ll_i2sbi(inode);
3389 struct ll_inode_info *lli = ll_i2info(inode);
3392 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3393 MDS_INODELOCK_LOOKUP);
3394 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3399 stat->dev = inode->i_sb->s_dev;
3400 if (ll_need_32bit_api(sbi))
3401 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3403 stat->ino = inode->i_ino;
3404 stat->mode = inode->i_mode;
3405 stat->uid = inode->i_uid;
3406 stat->gid = inode->i_gid;
3407 stat->rdev = inode->i_rdev;
3408 stat->atime = inode->i_atime;
3409 stat->mtime = inode->i_mtime;
3410 stat->ctime = inode->i_ctime;
3411 stat->blksize = 1 << inode->i_blkbits;
3412 stat->blocks = inode->i_blocks;
3414 if (S_ISDIR(inode->i_mode) &&
3415 ll_i2info(inode)->lli_lsm_md != NULL) {
3416 stat->nlink = lli->lli_stripe_dir_nlink;
3417 stat->size = lli->lli_stripe_dir_size;
3419 stat->nlink = inode->i_nlink;
3420 stat->size = i_size_read(inode);
3426 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3427 __u64 start, __u64 len)
3431 struct ll_user_fiemap *fiemap;
3432 unsigned int extent_count = fieinfo->fi_extents_max;
3434 num_bytes = sizeof(*fiemap) + (extent_count *
3435 sizeof(struct ll_fiemap_extent));
3436 OBD_ALLOC_LARGE(fiemap, num_bytes);
3441 fiemap->fm_flags = fieinfo->fi_flags;
3442 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3443 fiemap->fm_start = start;
3444 fiemap->fm_length = len;
3445 if (extent_count > 0)
3446 memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3447 sizeof(struct ll_fiemap_extent));
3449 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3451 fieinfo->fi_flags = fiemap->fm_flags;
3452 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3453 if (extent_count > 0)
3454 memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3455 fiemap->fm_mapped_extents *
3456 sizeof(struct ll_fiemap_extent));
3458 OBD_FREE_LARGE(fiemap, num_bytes);
3462 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3464 struct ll_inode_info *lli = ll_i2info(inode);
3465 struct posix_acl *acl = NULL;
3468 spin_lock(&lli->lli_lock);
3469 /* VFS' acl_permission_check->check_acl will release the refcount */
3470 acl = posix_acl_dup(lli->lli_posix_acl);
3471 spin_unlock(&lli->lli_lock);
3476 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
3478 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3479 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
3481 ll_check_acl(struct inode *inode, int mask)
3484 # ifdef CONFIG_FS_POSIX_ACL
3485 struct posix_acl *acl;
3489 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3490 if (flags & IPERM_FLAG_RCU)
3493 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
3498 rc = posix_acl_permission(inode, acl, mask);
3499 posix_acl_release(acl);
3502 # else /* !CONFIG_FS_POSIX_ACL */
3504 # endif /* CONFIG_FS_POSIX_ACL */
3506 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
3508 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
3509 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
3511 # ifdef HAVE_INODE_PERMISION_2ARGS
3512 int ll_inode_permission(struct inode *inode, int mask)
3514 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
3519 struct ll_sb_info *sbi;
3520 struct root_squash_info *squash;
3521 struct cred *cred = NULL;
3522 const struct cred *old_cred = NULL;
3524 bool squash_id = false;
3527 #ifdef MAY_NOT_BLOCK
3528 if (mask & MAY_NOT_BLOCK)
3530 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
3531 if (flags & IPERM_FLAG_RCU)
3535 /* as root inode are NOT getting validated in lookup operation,
3536 * need to do it before permission check. */
3538 if (inode == inode->i_sb->s_root->d_inode) {
3539 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3540 MDS_INODELOCK_LOOKUP);
3545 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3546 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3548 /* squash fsuid/fsgid if needed */
3549 sbi = ll_i2sbi(inode);
3550 squash = &sbi->ll_squash;
3551 if (unlikely(squash->rsi_uid != 0 &&
3552 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3553 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3557 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3558 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3559 squash->rsi_uid, squash->rsi_gid);
3561 /* update current process's credentials
3562 * and FS capability */
3563 cred = prepare_creds();
3567 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3568 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3569 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3570 if ((1 << cap) & CFS_CAP_FS_MASK)
3571 cap_lower(cred->cap_effective, cap);
3573 old_cred = override_creds(cred);
3576 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
3578 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
3579 rc = lustre_check_remote_perm(inode, mask);
3581 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
3583 /* restore current process's credentials and FS capability */
3585 revert_creds(old_cred);
3592 /* -o localflock - only provides locally consistent flock locks */
3593 struct file_operations ll_file_operations = {
3594 .read = ll_file_read,
3595 .aio_read = ll_file_aio_read,
3596 .write = ll_file_write,
3597 .aio_write = ll_file_aio_write,
3598 .unlocked_ioctl = ll_file_ioctl,
3599 .open = ll_file_open,
3600 .release = ll_file_release,
3601 .mmap = ll_file_mmap,
3602 .llseek = ll_file_seek,
3603 .splice_read = ll_file_splice_read,
3608 struct file_operations ll_file_operations_flock = {
3609 .read = ll_file_read,
3610 .aio_read = ll_file_aio_read,
3611 .write = ll_file_write,
3612 .aio_write = ll_file_aio_write,
3613 .unlocked_ioctl = ll_file_ioctl,
3614 .open = ll_file_open,
3615 .release = ll_file_release,
3616 .mmap = ll_file_mmap,
3617 .llseek = ll_file_seek,
3618 .splice_read = ll_file_splice_read,
3621 .flock = ll_file_flock,
3622 .lock = ll_file_flock
3625 /* These are for -o noflock - to return ENOSYS on flock calls */
3626 struct file_operations ll_file_operations_noflock = {
3627 .read = ll_file_read,
3628 .aio_read = ll_file_aio_read,
3629 .write = ll_file_write,
3630 .aio_write = ll_file_aio_write,
3631 .unlocked_ioctl = ll_file_ioctl,
3632 .open = ll_file_open,
3633 .release = ll_file_release,
3634 .mmap = ll_file_mmap,
3635 .llseek = ll_file_seek,
3636 .splice_read = ll_file_splice_read,
3639 .flock = ll_file_noflock,
3640 .lock = ll_file_noflock
3643 struct inode_operations ll_file_inode_operations = {
3644 .setattr = ll_setattr,
3645 .getattr = ll_getattr,
3646 .permission = ll_inode_permission,
3647 .setxattr = ll_setxattr,
3648 .getxattr = ll_getxattr,
3649 .listxattr = ll_listxattr,
3650 .removexattr = ll_removexattr,
3651 .fiemap = ll_fiemap,
3652 #ifdef HAVE_IOP_GET_ACL
3653 .get_acl = ll_get_acl,
3657 /* dynamic ioctl number support routins */
3658 static struct llioc_ctl_data {
3659 struct rw_semaphore ioc_sem;
3660 struct list_head ioc_head;
3662 __RWSEM_INITIALIZER(llioc.ioc_sem),
3663 LIST_HEAD_INIT(llioc.ioc_head)
3668 struct list_head iocd_list;
3669 unsigned int iocd_size;
3670 llioc_callback_t iocd_cb;
3671 unsigned int iocd_count;
3672 unsigned int iocd_cmd[0];
3675 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3678 struct llioc_data *in_data = NULL;
3681 if (cb == NULL || cmd == NULL ||
3682 count > LLIOC_MAX_CMD || count < 0)
3685 size = sizeof(*in_data) + count * sizeof(unsigned int);
3686 OBD_ALLOC(in_data, size);
3687 if (in_data == NULL)
3690 memset(in_data, 0, sizeof(*in_data));
3691 in_data->iocd_size = size;
3692 in_data->iocd_cb = cb;
3693 in_data->iocd_count = count;
3694 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3696 down_write(&llioc.ioc_sem);
3697 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3698 up_write(&llioc.ioc_sem);
3703 void ll_iocontrol_unregister(void *magic)
3705 struct llioc_data *tmp;
3710 down_write(&llioc.ioc_sem);
3711 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3713 unsigned int size = tmp->iocd_size;
3715 list_del(&tmp->iocd_list);
3716 up_write(&llioc.ioc_sem);
3718 OBD_FREE(tmp, size);
3722 up_write(&llioc.ioc_sem);
3724 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3727 EXPORT_SYMBOL(ll_iocontrol_register);
3728 EXPORT_SYMBOL(ll_iocontrol_unregister);
3730 static enum llioc_iter
3731 ll_iocontrol_call(struct inode *inode, struct file *file,
3732 unsigned int cmd, unsigned long arg, int *rcp)
3734 enum llioc_iter ret = LLIOC_CONT;
3735 struct llioc_data *data;
3736 int rc = -EINVAL, i;
3738 down_read(&llioc.ioc_sem);
3739 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3740 for (i = 0; i < data->iocd_count; i++) {
3741 if (cmd != data->iocd_cmd[i])
3744 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3748 if (ret == LLIOC_STOP)
3751 up_read(&llioc.ioc_sem);
3758 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3760 struct ll_inode_info *lli = ll_i2info(inode);
3761 struct cl_env_nest nest;
3766 if (lli->lli_clob == NULL)
3769 env = cl_env_nested_get(&nest);
3771 RETURN(PTR_ERR(env));
3773 result = cl_conf_set(env, lli->lli_clob, conf);
3774 cl_env_nested_put(&nest, env);
3776 if (conf->coc_opc == OBJECT_CONF_SET) {
3777 struct ldlm_lock *lock = conf->coc_lock;
3779 LASSERT(lock != NULL);
3780 LASSERT(ldlm_has_layout(lock));
3782 struct lustre_md *md = conf->u.coc_md;
3783 __u32 gen = LL_LAYOUT_GEN_EMPTY;
3785 /* it can only be allowed to match after layout is
3786 * applied to inode otherwise false layout would be
3787 * seen. Applying layout shoud happen before dropping
3788 * the intent lock. */
3789 ldlm_lock_allow_match(lock);
3791 lli->lli_has_smd = lsm_has_objects(md->lsm);
3792 if (md->lsm != NULL)
3793 gen = md->lsm->lsm_layout_gen;
3796 DFID ": layout version change: %u -> %u\n",
3797 PFID(&lli->lli_fid), ll_layout_version_get(lli),
3799 ll_layout_version_set(lli, gen);
3805 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3806 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3809 struct ll_sb_info *sbi = ll_i2sbi(inode);
3810 struct obd_capa *oc;
3811 struct ptlrpc_request *req;
3812 struct mdt_body *body;
3819 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3820 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3821 lock->l_lvb_data, lock->l_lvb_len);
3823 if ((lock->l_lvb_data != NULL) && ldlm_is_lvb_ready(lock))
3826 /* if layout lock was granted right away, the layout is returned
3827 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3828 * blocked and then granted via completion ast, we have to fetch
3829 * layout here. Please note that we can't use the LVB buffer in
3830 * completion AST because it doesn't have a large enough buffer */
3831 oc = ll_mdscapa_get(inode);
3832 rc = ll_get_default_mdsize(sbi, &lmmsize);
3834 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
3835 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3841 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3843 GOTO(out, rc = -EPROTO);
3845 lmmsize = body->mbo_eadatasize;
3846 if (lmmsize == 0) /* empty layout */
3849 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3851 GOTO(out, rc = -EFAULT);
3853 OBD_ALLOC_LARGE(lvbdata, lmmsize);
3854 if (lvbdata == NULL)
3855 GOTO(out, rc = -ENOMEM);
3857 memcpy(lvbdata, lmm, lmmsize);
3858 lock_res_and_lock(lock);
3859 if (lock->l_lvb_data != NULL)
3860 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
3862 lock->l_lvb_data = lvbdata;
3863 lock->l_lvb_len = lmmsize;
3864 unlock_res_and_lock(lock);
3869 ptlrpc_req_finished(req);
3874 * Apply the layout to the inode. Layout lock is held and will be released
3877 static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
3878 struct inode *inode, __u32 *gen, bool reconf)
3880 struct ll_inode_info *lli = ll_i2info(inode);
3881 struct ll_sb_info *sbi = ll_i2sbi(inode);
3882 struct ldlm_lock *lock;
3883 struct lustre_md md = { NULL };
3884 struct cl_object_conf conf;
3887 bool wait_layout = false;
3890 LASSERT(lustre_handle_is_used(lockh));
3892 lock = ldlm_handle2lock(lockh);
3893 LASSERT(lock != NULL);
3894 LASSERT(ldlm_has_layout(lock));
3896 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d",
3897 PFID(&lli->lli_fid), inode, reconf);
3899 /* in case this is a caching lock and reinstate with new inode */
3900 md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
3902 lock_res_and_lock(lock);
3903 lvb_ready = ldlm_is_lvb_ready(lock);
3904 unlock_res_and_lock(lock);
3905 /* checking lvb_ready is racy but this is okay. The worst case is
3906 * that multi processes may configure the file on the same time. */
3908 if (lvb_ready || !reconf) {
3911 /* layout_gen must be valid if layout lock is not
3912 * cancelled and stripe has already set */
3913 *gen = ll_layout_version_get(lli);
3919 rc = ll_layout_fetch(inode, lock);
3923 /* for layout lock, lmm is returned in lock's lvb.
3924 * lvb_data is immutable if the lock is held so it's safe to access it
3925 * without res lock. See the description in ldlm_lock_decref_internal()
3926 * for the condition to free lvb_data of layout lock */
3927 if (lock->l_lvb_data != NULL) {
3928 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3929 lock->l_lvb_data, lock->l_lvb_len);
3931 *gen = LL_LAYOUT_GEN_EMPTY;
3933 *gen = md.lsm->lsm_layout_gen;
3936 CERROR("%s: file "DFID" unpackmd error: %d\n",
3937 ll_get_fsname(inode->i_sb, NULL, 0),
3938 PFID(&lli->lli_fid), rc);
3944 /* set layout to file. Unlikely this will fail as old layout was
3945 * surely eliminated */
3946 memset(&conf, 0, sizeof conf);
3947 conf.coc_opc = OBJECT_CONF_SET;
3948 conf.coc_inode = inode;
3949 conf.coc_lock = lock;
3950 conf.u.coc_md = &md;
3951 rc = ll_layout_conf(inode, &conf);
3954 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3956 /* refresh layout failed, need to wait */
3957 wait_layout = rc == -EBUSY;
3961 LDLM_LOCK_PUT(lock);
3962 ldlm_lock_decref(lockh, mode);
3964 /* wait for IO to complete if it's still being used. */
3966 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3967 ll_get_fsname(inode->i_sb, NULL, 0),
3968 PFID(&lli->lli_fid), inode);
3970 memset(&conf, 0, sizeof conf);
3971 conf.coc_opc = OBJECT_CONF_WAIT;
3972 conf.coc_inode = inode;
3973 rc = ll_layout_conf(inode, &conf);
3977 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
3978 ll_get_fsname(inode->i_sb, NULL, 0),
3979 PFID(&lli->lli_fid), rc);
3985 * This function checks if there exists a LAYOUT lock on the client side,
3986 * or enqueues it if it doesn't have one in cache.
3988 * This function will not hold layout lock so it may be revoked any time after
3989 * this function returns. Any operations depend on layout should be redone
3992 * This function should be called before lov_io_init() to get an uptodate
3993 * layout version, the caller should save the version number and after IO
3994 * is finished, this function should be called again to verify that layout
3995 * is not changed during IO time.
3997 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3999 struct ll_inode_info *lli = ll_i2info(inode);
4000 struct ll_sb_info *sbi = ll_i2sbi(inode);
4001 struct md_op_data *op_data;
4002 struct lookup_intent it;
4003 struct lustre_handle lockh;
4005 struct ldlm_enqueue_info einfo = {
4006 .ei_type = LDLM_IBITS,
4008 .ei_cb_bl = &ll_md_blocking_ast,
4009 .ei_cb_cp = &ldlm_completion_ast,
4014 *gen = ll_layout_version_get(lli);
4015 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
4019 LASSERT(fid_is_sane(ll_inode2fid(inode)));
4020 LASSERT(S_ISREG(inode->i_mode));
4022 /* take layout lock mutex to enqueue layout lock exclusively. */
4023 mutex_lock(&lli->lli_layout_mutex);
4026 /* mostly layout lock is caching on the local side, so try to match
4027 * it before grabbing layout lock mutex. */
4028 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
4029 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
4030 if (mode != 0) { /* hit cached lock */
4031 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4035 mutex_unlock(&lli->lli_layout_mutex);
4039 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
4040 0, 0, LUSTRE_OPC_ANY, NULL);
4041 if (IS_ERR(op_data)) {
4042 mutex_unlock(&lli->lli_layout_mutex);
4043 RETURN(PTR_ERR(op_data));
4046 /* have to enqueue one */
4047 memset(&it, 0, sizeof(it));
4048 it.it_op = IT_LAYOUT;
4049 lockh.cookie = 0ULL;
4051 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
4052 ll_get_fsname(inode->i_sb, NULL, 0),
4053 PFID(&lli->lli_fid), inode);
4055 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
4056 if (it.d.lustre.it_data != NULL)
4057 ptlrpc_req_finished(it.d.lustre.it_data);
4058 it.d.lustre.it_data = NULL;
4060 ll_finish_md_op_data(op_data);
4062 mode = it.d.lustre.it_lock_mode;
4063 it.d.lustre.it_lock_mode = 0;
4064 ll_intent_drop_lock(&it);
4067 /* set lock data in case this is a new lock */
4068 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
4069 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4073 mutex_unlock(&lli->lli_layout_mutex);
4079 * This function send a restore request to the MDT
4081 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
4083 struct hsm_user_request *hur;
4087 len = sizeof(struct hsm_user_request) +
4088 sizeof(struct hsm_user_item);
4089 OBD_ALLOC(hur, len);
4093 hur->hur_request.hr_action = HUA_RESTORE;
4094 hur->hur_request.hr_archive_id = 0;
4095 hur->hur_request.hr_flags = 0;
4096 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
4097 sizeof(hur->hur_user_item[0].hui_fid));
4098 hur->hur_user_item[0].hui_extent.offset = offset;
4099 hur->hur_user_item[0].hui_extent.length = length;
4100 hur->hur_request.hr_itemcount = 1;
4101 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, cl_i2sbi(inode)->ll_md_exp,