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 if (pathlen > PATH_MAX)
1919 outsize = sizeof(*gfout) + pathlen;
1920 OBD_ALLOC(gfout, outsize);
1924 if (copy_from_user(gfout, arg, sizeof(*gfout)))
1925 GOTO(gf_free, rc = -EFAULT);
1927 /* Call mdc_iocontrol */
1928 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1932 if (copy_to_user(arg, gfout, outsize))
1936 OBD_FREE(gfout, outsize);
1940 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1942 struct ll_user_fiemap *fiemap_s;
1943 size_t num_bytes, ret_bytes;
1944 unsigned int extent_count;
1947 /* Get the extent count so we can calculate the size of
1948 * required fiemap buffer */
1949 if (get_user(extent_count,
1950 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1952 num_bytes = sizeof(*fiemap_s) + (extent_count *
1953 sizeof(struct ll_fiemap_extent));
1955 OBD_ALLOC_LARGE(fiemap_s, num_bytes);
1956 if (fiemap_s == NULL)
1959 /* get the fiemap value */
1960 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1962 GOTO(error, rc = -EFAULT);
1964 /* If fm_extent_count is non-zero, read the first extent since
1965 * it is used to calculate end_offset and device from previous
1968 if (copy_from_user(&fiemap_s->fm_extents[0],
1969 (char __user *)arg + sizeof(*fiemap_s),
1970 sizeof(struct ll_fiemap_extent)))
1971 GOTO(error, rc = -EFAULT);
1974 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1978 ret_bytes = sizeof(struct ll_user_fiemap);
1980 if (extent_count != 0)
1981 ret_bytes += (fiemap_s->fm_mapped_extents *
1982 sizeof(struct ll_fiemap_extent));
1984 if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
1988 OBD_FREE_LARGE(fiemap_s, num_bytes);
1993 * Read the data_version for inode.
1995 * This value is computed using stripe object version on OST.
1996 * Version is computed using server side locking.
1998 * @param sync if do sync on the OST side;
2000 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2001 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2003 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2005 struct lov_stripe_md *lsm = NULL;
2006 struct ll_sb_info *sbi = ll_i2sbi(inode);
2007 struct obdo *obdo = NULL;
2011 /* If no stripe, we consider version is 0. */
2012 lsm = ccc_inode_lsm_get(inode);
2013 if (!lsm_has_objects(lsm)) {
2015 CDEBUG(D_INODE, "No object for inode\n");
2019 OBD_ALLOC_PTR(obdo);
2021 GOTO(out, rc = -ENOMEM);
2023 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, flags);
2025 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
2028 *data_version = obdo->o_data_version;
2034 ccc_inode_lsm_put(inode, lsm);
2039 * Trigger a HSM release request for the provided inode.
2041 int ll_hsm_release(struct inode *inode)
2043 struct cl_env_nest nest;
2045 struct obd_client_handle *och = NULL;
2046 __u64 data_version = 0;
2050 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2051 ll_get_fsname(inode->i_sb, NULL, 0),
2052 PFID(&ll_i2info(inode)->lli_fid));
2054 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2056 GOTO(out, rc = PTR_ERR(och));
2058 /* Grab latest data_version and [am]time values */
2059 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2063 env = cl_env_nested_get(&nest);
2065 GOTO(out, rc = PTR_ERR(env));
2067 ll_merge_lvb(env, inode);
2068 cl_env_nested_put(&nest, env);
2070 /* Release the file.
2071 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2072 * we still need it to pack l_remote_handle to MDT. */
2073 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
2079 if (och != NULL && !IS_ERR(och)) /* close the file */
2080 ll_lease_close(och, inode, NULL);
2085 struct ll_swap_stack {
2086 struct iattr ia1, ia2;
2088 struct inode *inode1, *inode2;
2089 bool check_dv1, check_dv2;
2092 static int ll_swap_layouts(struct file *file1, struct file *file2,
2093 struct lustre_swap_layouts *lsl)
2095 struct mdc_swap_layouts msl;
2096 struct md_op_data *op_data;
2099 struct ll_swap_stack *llss = NULL;
2102 OBD_ALLOC_PTR(llss);
2106 llss->inode1 = file1->f_dentry->d_inode;
2107 llss->inode2 = file2->f_dentry->d_inode;
2109 if (!S_ISREG(llss->inode2->i_mode))
2110 GOTO(free, rc = -EINVAL);
2112 if (inode_permission(llss->inode1, MAY_WRITE) ||
2113 inode_permission(llss->inode2, MAY_WRITE))
2114 GOTO(free, rc = -EPERM);
2116 if (llss->inode2->i_sb != llss->inode1->i_sb)
2117 GOTO(free, rc = -EXDEV);
2119 /* we use 2 bool because it is easier to swap than 2 bits */
2120 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2121 llss->check_dv1 = true;
2123 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2124 llss->check_dv2 = true;
2126 /* we cannot use lsl->sl_dvX directly because we may swap them */
2127 llss->dv1 = lsl->sl_dv1;
2128 llss->dv2 = lsl->sl_dv2;
2130 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2131 if (rc == 0) /* same file, done! */
2134 if (rc < 0) { /* sequentialize it */
2135 swap(llss->inode1, llss->inode2);
2137 swap(llss->dv1, llss->dv2);
2138 swap(llss->check_dv1, llss->check_dv2);
2142 if (gid != 0) { /* application asks to flush dirty cache */
2143 rc = ll_get_grouplock(llss->inode1, file1, gid);
2147 rc = ll_get_grouplock(llss->inode2, file2, gid);
2149 ll_put_grouplock(llss->inode1, file1, gid);
2154 /* to be able to restore mtime and atime after swap
2155 * we need to first save them */
2157 (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
2158 llss->ia1.ia_mtime = llss->inode1->i_mtime;
2159 llss->ia1.ia_atime = llss->inode1->i_atime;
2160 llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
2161 llss->ia2.ia_mtime = llss->inode2->i_mtime;
2162 llss->ia2.ia_atime = llss->inode2->i_atime;
2163 llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
2166 /* ultimate check, before swaping the layouts we check if
2167 * dataversion has changed (if requested) */
2168 if (llss->check_dv1) {
2169 rc = ll_data_version(llss->inode1, &dv, 0);
2172 if (dv != llss->dv1)
2173 GOTO(putgl, rc = -EAGAIN);
2176 if (llss->check_dv2) {
2177 rc = ll_data_version(llss->inode2, &dv, 0);
2180 if (dv != llss->dv2)
2181 GOTO(putgl, rc = -EAGAIN);
2184 /* struct md_op_data is used to send the swap args to the mdt
2185 * only flags is missing, so we use struct mdc_swap_layouts
2186 * through the md_op_data->op_data */
2187 /* flags from user space have to be converted before they are send to
2188 * server, no flag is sent today, they are only used on the client */
2191 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2192 0, LUSTRE_OPC_ANY, &msl);
2193 if (IS_ERR(op_data))
2194 GOTO(free, rc = PTR_ERR(op_data));
2196 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2197 sizeof(*op_data), op_data, NULL);
2198 ll_finish_md_op_data(op_data);
2202 ll_put_grouplock(llss->inode2, file2, gid);
2203 ll_put_grouplock(llss->inode1, file1, gid);
2206 /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
2210 /* clear useless flags */
2211 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
2212 llss->ia1.ia_valid &= ~ATTR_MTIME;
2213 llss->ia2.ia_valid &= ~ATTR_MTIME;
2216 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
2217 llss->ia1.ia_valid &= ~ATTR_ATIME;
2218 llss->ia2.ia_valid &= ~ATTR_ATIME;
2221 /* update time if requested */
2223 if (llss->ia2.ia_valid != 0) {
2224 mutex_lock(&llss->inode1->i_mutex);
2225 rc = ll_setattr(file1->f_dentry, &llss->ia2);
2226 mutex_unlock(&llss->inode1->i_mutex);
2229 if (llss->ia1.ia_valid != 0) {
2232 mutex_lock(&llss->inode2->i_mutex);
2233 rc1 = ll_setattr(file2->f_dentry, &llss->ia1);
2234 mutex_unlock(&llss->inode2->i_mutex);
2246 static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2248 struct md_op_data *op_data;
2251 /* Non-root users are forbidden to set or clear flags which are
2252 * NOT defined in HSM_USER_MASK. */
2253 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2254 !cfs_capable(CFS_CAP_SYS_ADMIN))
2257 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2258 LUSTRE_OPC_ANY, hss);
2259 if (IS_ERR(op_data))
2260 RETURN(PTR_ERR(op_data));
2262 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2263 sizeof(*op_data), op_data, NULL);
2265 ll_finish_md_op_data(op_data);
2270 static int ll_hsm_import(struct inode *inode, struct file *file,
2271 struct hsm_user_import *hui)
2273 struct hsm_state_set *hss = NULL;
2274 struct iattr *attr = NULL;
2278 if (!S_ISREG(inode->i_mode))
2284 GOTO(out, rc = -ENOMEM);
2286 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2287 hss->hss_archive_id = hui->hui_archive_id;
2288 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2289 rc = ll_hsm_state_set(inode, hss);
2293 OBD_ALLOC_PTR(attr);
2295 GOTO(out, rc = -ENOMEM);
2297 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2298 attr->ia_mode |= S_IFREG;
2299 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2300 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2301 attr->ia_size = hui->hui_size;
2302 attr->ia_mtime.tv_sec = hui->hui_mtime;
2303 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2304 attr->ia_atime.tv_sec = hui->hui_atime;
2305 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2307 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2308 ATTR_UID | ATTR_GID |
2309 ATTR_MTIME | ATTR_MTIME_SET |
2310 ATTR_ATIME | ATTR_ATIME_SET;
2312 mutex_lock(&inode->i_mutex);
2314 rc = ll_setattr_raw(file->f_dentry, attr, true);
2318 mutex_unlock(&inode->i_mutex);
2330 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2332 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2333 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2337 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2339 struct inode *inode = file->f_dentry->d_inode;
2340 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2344 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
2345 PFID(ll_inode2fid(inode)), inode, cmd);
2346 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2348 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2349 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2353 case LL_IOC_GETFLAGS:
2354 /* Get the current value of the file flags */
2355 return put_user(fd->fd_flags, (int *)arg);
2356 case LL_IOC_SETFLAGS:
2357 case LL_IOC_CLRFLAGS:
2358 /* Set or clear specific file flags */
2359 /* XXX This probably needs checks to ensure the flags are
2360 * not abused, and to handle any flag side effects.
2362 if (get_user(flags, (int *) arg))
2365 if (cmd == LL_IOC_SETFLAGS) {
2366 if ((flags & LL_FILE_IGNORE_LOCK) &&
2367 !(file->f_flags & O_DIRECT)) {
2368 CERROR("%s: unable to disable locking on "
2369 "non-O_DIRECT file\n", current->comm);
2373 fd->fd_flags |= flags;
2375 fd->fd_flags &= ~flags;
2378 case LL_IOC_LOV_SETSTRIPE:
2379 RETURN(ll_lov_setstripe(inode, file, arg));
2380 case LL_IOC_LOV_SETEA:
2381 RETURN(ll_lov_setea(inode, file, arg));
2382 case LL_IOC_LOV_SWAP_LAYOUTS: {
2384 struct lustre_swap_layouts lsl;
2386 if (copy_from_user(&lsl, (char *)arg,
2387 sizeof(struct lustre_swap_layouts)))
2390 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
2393 file2 = fget(lsl.sl_fd);
2398 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
2399 rc = ll_swap_layouts(file, file2, &lsl);
2403 case LL_IOC_LOV_GETSTRIPE:
2404 RETURN(ll_lov_getstripe(inode, arg));
2405 case LL_IOC_RECREATE_OBJ:
2406 RETURN(ll_lov_recreate_obj(inode, arg));
2407 case LL_IOC_RECREATE_FID:
2408 RETURN(ll_lov_recreate_fid(inode, arg));
2409 case FSFILT_IOC_FIEMAP:
2410 RETURN(ll_ioctl_fiemap(inode, arg));
2411 case FSFILT_IOC_GETFLAGS:
2412 case FSFILT_IOC_SETFLAGS:
2413 RETURN(ll_iocontrol(inode, file, cmd, arg));
2414 case FSFILT_IOC_GETVERSION_OLD:
2415 case FSFILT_IOC_GETVERSION:
2416 RETURN(put_user(inode->i_generation, (int *)arg));
2417 case LL_IOC_GROUP_LOCK:
2418 RETURN(ll_get_grouplock(inode, file, arg));
2419 case LL_IOC_GROUP_UNLOCK:
2420 RETURN(ll_put_grouplock(inode, file, arg));
2421 case IOC_OBD_STATFS:
2422 RETURN(ll_obd_statfs(inode, (void *)arg));
2424 /* We need to special case any other ioctls we want to handle,
2425 * to send them to the MDS/OST as appropriate and to properly
2426 * network encode the arg field.
2427 case FSFILT_IOC_SETVERSION_OLD:
2428 case FSFILT_IOC_SETVERSION:
2430 case LL_IOC_FLUSHCTX:
2431 RETURN(ll_flush_ctx(inode));
2432 case LL_IOC_PATH2FID: {
2433 if (copy_to_user((void *)arg, ll_inode2fid(inode),
2434 sizeof(struct lu_fid)))
2439 case OBD_IOC_FID2PATH:
2440 RETURN(ll_fid2path(inode, (void *)arg));
2441 case LL_IOC_DATA_VERSION: {
2442 struct ioc_data_version idv;
2445 if (copy_from_user(&idv, (char *)arg, sizeof(idv)))
2448 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2449 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2451 if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv)))
2457 case LL_IOC_GET_MDTIDX: {
2460 mdtidx = ll_get_mdt_idx(inode);
2464 if (put_user((int)mdtidx, (int*)arg))
2469 case OBD_IOC_GETDTNAME:
2470 case OBD_IOC_GETMDNAME:
2471 RETURN(ll_get_obd_name(inode, cmd, arg));
2472 case LL_IOC_HSM_STATE_GET: {
2473 struct md_op_data *op_data;
2474 struct hsm_user_state *hus;
2481 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2482 LUSTRE_OPC_ANY, hus);
2483 if (IS_ERR(op_data)) {
2485 RETURN(PTR_ERR(op_data));
2488 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2491 if (copy_to_user((void *)arg, hus, sizeof(*hus)))
2494 ll_finish_md_op_data(op_data);
2498 case LL_IOC_HSM_STATE_SET: {
2499 struct hsm_state_set *hss;
2506 if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
2511 rc = ll_hsm_state_set(inode, hss);
2516 case LL_IOC_HSM_ACTION: {
2517 struct md_op_data *op_data;
2518 struct hsm_current_action *hca;
2525 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2526 LUSTRE_OPC_ANY, hca);
2527 if (IS_ERR(op_data)) {
2529 RETURN(PTR_ERR(op_data));
2532 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2535 if (copy_to_user((char *)arg, hca, sizeof(*hca)))
2538 ll_finish_md_op_data(op_data);
2542 case LL_IOC_SET_LEASE: {
2543 struct ll_inode_info *lli = ll_i2info(inode);
2544 struct obd_client_handle *och = NULL;
2549 case LL_LEASE_WRLCK:
2550 if (!(file->f_mode & FMODE_WRITE))
2552 fmode = FMODE_WRITE;
2554 case LL_LEASE_RDLCK:
2555 if (!(file->f_mode & FMODE_READ))
2559 case LL_LEASE_UNLCK:
2560 mutex_lock(&lli->lli_och_mutex);
2561 if (fd->fd_lease_och != NULL) {
2562 och = fd->fd_lease_och;
2563 fd->fd_lease_och = NULL;
2565 mutex_unlock(&lli->lli_och_mutex);
2570 fmode = och->och_flags;
2571 rc = ll_lease_close(och, inode, &lease_broken);
2578 RETURN(ll_lease_type_from_fmode(fmode));
2583 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
2585 /* apply for lease */
2586 och = ll_lease_open(inode, file, fmode, 0);
2588 RETURN(PTR_ERR(och));
2591 mutex_lock(&lli->lli_och_mutex);
2592 if (fd->fd_lease_och == NULL) {
2593 fd->fd_lease_och = och;
2596 mutex_unlock(&lli->lli_och_mutex);
2598 /* impossible now that only excl is supported for now */
2599 ll_lease_close(och, inode, &lease_broken);
2604 case LL_IOC_GET_LEASE: {
2605 struct ll_inode_info *lli = ll_i2info(inode);
2606 struct ldlm_lock *lock = NULL;
2609 mutex_lock(&lli->lli_och_mutex);
2610 if (fd->fd_lease_och != NULL) {
2611 struct obd_client_handle *och = fd->fd_lease_och;
2613 lock = ldlm_handle2lock(&och->och_lease_handle);
2615 lock_res_and_lock(lock);
2616 if (!ldlm_is_cancel(lock))
2617 fmode = och->och_flags;
2619 unlock_res_and_lock(lock);
2620 LDLM_LOCK_PUT(lock);
2623 mutex_unlock(&lli->lli_och_mutex);
2625 RETURN(ll_lease_type_from_fmode(fmode));
2627 case LL_IOC_HSM_IMPORT: {
2628 struct hsm_user_import *hui;
2634 if (copy_from_user(hui, (void *)arg, sizeof(*hui))) {
2639 rc = ll_hsm_import(inode, file, hui);
2649 ll_iocontrol_call(inode, file, cmd, arg, &err))
2652 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2658 #ifndef HAVE_FILE_LLSEEK_SIZE
2659 static inline loff_t
2660 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2662 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2664 if (offset > maxsize)
2667 if (offset != file->f_pos) {
2668 file->f_pos = offset;
2669 file->f_version = 0;
2675 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2676 loff_t maxsize, loff_t eof)
2678 struct inode *inode = file->f_dentry->d_inode;
2686 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2687 * position-querying operation. Avoid rewriting the "same"
2688 * f_pos value back to the file because a concurrent read(),
2689 * write() or lseek() might have altered it
2694 * f_lock protects against read/modify/write race with other
2695 * SEEK_CURs. Note that parallel writes and reads behave
2698 mutex_lock(&inode->i_mutex);
2699 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2700 mutex_unlock(&inode->i_mutex);
2704 * In the generic case the entire file is data, so as long as
2705 * offset isn't at the end of the file then the offset is data.
2712 * There is a virtual hole at the end of the file, so as long as
2713 * offset isn't i_size or larger, return i_size.
2721 return llseek_execute(file, offset, maxsize);
2725 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2727 struct inode *inode = file->f_dentry->d_inode;
2728 loff_t retval, eof = 0;
2731 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2732 (origin == SEEK_CUR) ? file->f_pos : 0);
2733 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2734 PFID(ll_inode2fid(inode)), inode, retval, retval,
2736 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2738 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2739 retval = ll_glimpse_size(inode);
2742 eof = i_size_read(inode);
2745 retval = ll_generic_file_llseek_size(file, offset, origin,
2746 ll_file_maxbytes(inode), eof);
2750 static int ll_flush(struct file *file, fl_owner_t id)
2752 struct inode *inode = file->f_dentry->d_inode;
2753 struct ll_inode_info *lli = ll_i2info(inode);
2754 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2757 LASSERT(!S_ISDIR(inode->i_mode));
2759 /* catch async errors that were recorded back when async writeback
2760 * failed for pages in this mapping. */
2761 rc = lli->lli_async_rc;
2762 lli->lli_async_rc = 0;
2763 if (lli->lli_clob != NULL) {
2764 err = lov_read_and_clear_async_rc(lli->lli_clob);
2769 /* The application has been told write failure already.
2770 * Do not report failure again. */
2771 if (fd->fd_write_failed)
2773 return rc ? -EIO : 0;
2777 * Called to make sure a portion of file has been written out.
2778 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2780 * Return how many pages have been written.
2782 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2783 enum cl_fsync_mode mode, int ignore_layout)
2785 struct cl_env_nest nest;
2788 struct obd_capa *capa = NULL;
2789 struct cl_fsync_io *fio;
2793 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2794 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2797 env = cl_env_nested_get(&nest);
2799 RETURN(PTR_ERR(env));
2801 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
2803 io = ccc_env_thread_io(env);
2804 io->ci_obj = cl_i2info(inode)->lli_clob;
2805 io->ci_ignore_layout = ignore_layout;
2807 /* initialize parameters for sync */
2808 fio = &io->u.ci_fsync;
2809 fio->fi_capa = capa;
2810 fio->fi_start = start;
2812 fio->fi_fid = ll_inode2fid(inode);
2813 fio->fi_mode = mode;
2814 fio->fi_nr_written = 0;
2816 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2817 result = cl_io_loop(env, io);
2819 result = io->ci_result;
2821 result = fio->fi_nr_written;
2822 cl_io_fini(env, io);
2823 cl_env_nested_put(&nest, env);
2831 * When dentry is provided (the 'else' case), *file->f_dentry may be
2832 * null and dentry must be used directly rather than pulled from
2833 * *file->f_dentry as is done otherwise.
2836 #ifdef HAVE_FILE_FSYNC_4ARGS
2837 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2839 struct dentry *dentry = file->f_dentry;
2840 #elif defined(HAVE_FILE_FSYNC_2ARGS)
2841 int ll_fsync(struct file *file, int datasync)
2843 struct dentry *dentry = file->f_dentry;
2845 loff_t end = LLONG_MAX;
2847 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
2850 loff_t end = LLONG_MAX;
2852 struct inode *inode = dentry->d_inode;
2853 struct ll_inode_info *lli = ll_i2info(inode);
2854 struct ptlrpc_request *req;
2855 struct obd_capa *oc;
2859 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2860 PFID(ll_inode2fid(inode)), inode);
2861 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2863 #ifdef HAVE_FILE_FSYNC_4ARGS
2864 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2865 mutex_lock(&inode->i_mutex);
2867 /* fsync's caller has already called _fdata{sync,write}, we want
2868 * that IO to finish before calling the osc and mdc sync methods */
2869 rc = filemap_fdatawait(inode->i_mapping);
2872 /* catch async errors that were recorded back when async writeback
2873 * failed for pages in this mapping. */
2874 if (!S_ISDIR(inode->i_mode)) {
2875 err = lli->lli_async_rc;
2876 lli->lli_async_rc = 0;
2879 err = lov_read_and_clear_async_rc(lli->lli_clob);
2884 oc = ll_mdscapa_get(inode);
2885 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2891 ptlrpc_req_finished(req);
2893 if (S_ISREG(inode->i_mode)) {
2894 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2896 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2897 if (rc == 0 && err < 0)
2900 fd->fd_write_failed = true;
2902 fd->fd_write_failed = false;
2905 #ifdef HAVE_FILE_FSYNC_4ARGS
2906 mutex_unlock(&inode->i_mutex);
2912 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2914 struct inode *inode = file->f_dentry->d_inode;
2915 struct ll_sb_info *sbi = ll_i2sbi(inode);
2916 struct ldlm_enqueue_info einfo = {
2917 .ei_type = LDLM_FLOCK,
2918 .ei_cb_cp = ldlm_flock_completion_ast,
2919 .ei_cbdata = file_lock,
2921 struct md_op_data *op_data;
2922 struct lustre_handle lockh = {0};
2923 ldlm_policy_data_t flock = {{0}};
2924 int fl_type = file_lock->fl_type;
2930 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2931 PFID(ll_inode2fid(inode)), file_lock);
2933 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2935 if (file_lock->fl_flags & FL_FLOCK) {
2936 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2937 /* flocks are whole-file locks */
2938 flock.l_flock.end = OFFSET_MAX;
2939 /* For flocks owner is determined by the local file desctiptor*/
2940 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
2941 } else if (file_lock->fl_flags & FL_POSIX) {
2942 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2943 flock.l_flock.start = file_lock->fl_start;
2944 flock.l_flock.end = file_lock->fl_end;
2948 flock.l_flock.pid = file_lock->fl_pid;
2950 /* Somewhat ugly workaround for svc lockd.
2951 * lockd installs custom fl_lmops->lm_compare_owner that checks
2952 * for the fl_owner to be the same (which it always is on local node
2953 * I guess between lockd processes) and then compares pid.
2954 * As such we assign pid to the owner field to make it all work,
2955 * conflict with normal locks is unlikely since pid space and
2956 * pointer space for current->files are not intersecting */
2957 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2958 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2962 einfo.ei_mode = LCK_PR;
2965 /* An unlock request may or may not have any relation to
2966 * existing locks so we may not be able to pass a lock handle
2967 * via a normal ldlm_lock_cancel() request. The request may even
2968 * unlock a byte range in the middle of an existing lock. In
2969 * order to process an unlock request we need all of the same
2970 * information that is given with a normal read or write record
2971 * lock request. To avoid creating another ldlm unlock (cancel)
2972 * message we'll treat a LCK_NL flock request as an unlock. */
2973 einfo.ei_mode = LCK_NL;
2976 einfo.ei_mode = LCK_PW;
2979 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
2994 flags = LDLM_FL_BLOCK_NOWAIT;
3000 flags = LDLM_FL_TEST_LOCK;
3003 CERROR("unknown fcntl lock command: %d\n", cmd);
3007 /* Save the old mode so that if the mode in the lock changes we
3008 * can decrement the appropriate reader or writer refcount. */
3009 file_lock->fl_type = einfo.ei_mode;
3011 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3012 LUSTRE_OPC_ANY, NULL);
3013 if (IS_ERR(op_data))
3014 RETURN(PTR_ERR(op_data));
3016 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags="LPX64", mode=%u, "
3017 "start="LPU64", end="LPU64"\n", PFID(ll_inode2fid(inode)),
3018 flock.l_flock.pid, flags, einfo.ei_mode,
3019 flock.l_flock.start, flock.l_flock.end);
3021 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
3024 /* Restore the file lock type if not TEST lock. */
3025 if (!(flags & LDLM_FL_TEST_LOCK))
3026 file_lock->fl_type = fl_type;
3028 if ((file_lock->fl_flags & FL_FLOCK) &&
3029 (rc == 0 || file_lock->fl_type == F_UNLCK))
3030 rc2 = flock_lock_file_wait(file, file_lock);
3031 if ((file_lock->fl_flags & FL_POSIX) &&
3032 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
3033 !(flags & LDLM_FL_TEST_LOCK))
3034 rc2 = posix_lock_file_wait(file, file_lock);
3036 if (rc2 && file_lock->fl_type != F_UNLCK) {
3037 einfo.ei_mode = LCK_NL;
3038 md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
3043 ll_finish_md_op_data(op_data);
3048 int ll_get_fid_by_name(struct inode *parent, const char *name,
3049 int namelen, struct lu_fid *fid)
3051 struct md_op_data *op_data = NULL;
3052 struct mdt_body *body;
3053 struct ptlrpc_request *req;
3057 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
3058 LUSTRE_OPC_ANY, NULL);
3059 if (IS_ERR(op_data))
3060 RETURN(PTR_ERR(op_data));
3062 op_data->op_valid = OBD_MD_FLID;
3063 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
3064 ll_finish_md_op_data(op_data);
3068 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3070 GOTO(out_req, rc = -EFAULT);
3072 *fid = body->mbo_fid1;
3074 ptlrpc_req_finished(req);
3078 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
3079 const char *name, int namelen)
3081 struct dentry *dchild = NULL;
3082 struct inode *child_inode = NULL;
3083 struct md_op_data *op_data;
3084 struct ptlrpc_request *request = NULL;
3089 CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%04x\n",
3090 name, PFID(ll_inode2fid(parent)), mdtidx);
3092 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
3093 0, LUSTRE_OPC_ANY, NULL);
3094 if (IS_ERR(op_data))
3095 RETURN(PTR_ERR(op_data));
3097 /* Get child FID first */
3098 qstr.hash = full_name_hash(name, namelen);
3101 dchild = d_lookup(file->f_dentry, &qstr);
3102 if (dchild != NULL && dchild->d_inode != NULL) {
3103 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
3104 if (dchild->d_inode != NULL) {
3105 child_inode = igrab(dchild->d_inode);
3106 ll_invalidate_aliases(child_inode);
3110 rc = ll_get_fid_by_name(parent, name, namelen,
3116 if (!fid_is_sane(&op_data->op_fid3)) {
3117 CERROR("%s: migrate %s , but fid "DFID" is insane\n",
3118 ll_get_fsname(parent->i_sb, NULL, 0), name,
3119 PFID(&op_data->op_fid3));
3120 GOTO(out_free, rc = -EINVAL);
3123 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
3128 CDEBUG(D_INFO, "%s:"DFID" is already on MDT%d.\n", name,
3129 PFID(&op_data->op_fid3), mdtidx);
3130 GOTO(out_free, rc = 0);
3133 op_data->op_mds = mdtidx;
3134 op_data->op_cli_flags = CLI_MIGRATE;
3135 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
3136 namelen, name, namelen, &request);
3138 ll_update_times(request, parent);
3140 ptlrpc_req_finished(request);
3145 if (child_inode != NULL) {
3146 clear_nlink(child_inode);
3150 ll_finish_md_op_data(op_data);
3155 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
3163 * test if some locks matching bits and l_req_mode are acquired
3164 * - bits can be in different locks
3165 * - if found clear the common lock bits in *bits
3166 * - the bits not found, are kept in *bits
3168 * \param bits [IN] searched lock bits [IN]
3169 * \param l_req_mode [IN] searched lock mode
3170 * \retval boolean, true iff all bits are found
3172 int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode)
3174 struct lustre_handle lockh;
3175 ldlm_policy_data_t policy;
3176 ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
3177 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
3186 fid = &ll_i2info(inode)->lli_fid;
3187 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
3188 ldlm_lockname[mode]);
3190 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
3191 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
3192 policy.l_inodebits.bits = *bits & (1 << i);
3193 if (policy.l_inodebits.bits == 0)
3196 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
3197 &policy, mode, &lockh)) {
3198 struct ldlm_lock *lock;
3200 lock = ldlm_handle2lock(&lockh);
3203 ~(lock->l_policy_data.l_inodebits.bits);
3204 LDLM_LOCK_PUT(lock);
3206 *bits &= ~policy.l_inodebits.bits;
3213 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
3214 struct lustre_handle *lockh, __u64 flags,
3217 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
3222 fid = &ll_i2info(inode)->lli_fid;
3223 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
3225 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
3226 fid, LDLM_IBITS, &policy, mode, lockh);
3231 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
3233 /* Already unlinked. Just update nlink and return success */
3234 if (rc == -ENOENT) {
3236 /* This path cannot be hit for regular files unless in
3237 * case of obscure races, so no need to to validate
3239 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
3241 } else if (rc != 0) {
3242 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
3243 "%s: revalidate FID "DFID" error: rc = %d\n",
3244 ll_get_fsname(inode->i_sb, NULL, 0),
3245 PFID(ll_inode2fid(inode)), rc);
3251 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3253 struct inode *inode = dentry->d_inode;
3254 struct ptlrpc_request *req = NULL;
3255 struct obd_export *exp;
3259 LASSERT(inode != NULL);
3261 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
3262 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
3264 exp = ll_i2mdexp(inode);
3266 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
3267 * But under CMD case, it caused some lock issues, should be fixed
3268 * with new CMD ibits lock. See bug 12718 */
3269 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
3270 struct lookup_intent oit = { .it_op = IT_GETATTR };
3271 struct md_op_data *op_data;
3273 if (ibits == MDS_INODELOCK_LOOKUP)
3274 oit.it_op = IT_LOOKUP;
3276 /* Call getattr by fid, so do not provide name at all. */
3277 op_data = ll_prep_md_op_data(NULL, dentry->d_inode,
3278 dentry->d_inode, NULL, 0, 0,
3279 LUSTRE_OPC_ANY, NULL);
3280 if (IS_ERR(op_data))
3281 RETURN(PTR_ERR(op_data));
3283 rc = md_intent_lock(exp, op_data, &oit, &req,
3284 &ll_md_blocking_ast, 0);
3285 ll_finish_md_op_data(op_data);
3287 rc = ll_inode_revalidate_fini(inode, rc);
3291 rc = ll_revalidate_it_finish(req, &oit, dentry);
3293 ll_intent_release(&oit);
3297 /* Unlinked? Unhash dentry, so it is not picked up later by
3298 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
3299 here to preserve get_cwd functionality on 2.6.
3301 if (!dentry->d_inode->i_nlink)
3302 d_lustre_invalidate(dentry, 0);
3304 ll_lookup_finish_locks(&oit, dentry);
3305 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
3306 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
3307 obd_valid valid = OBD_MD_FLGETATTR;
3308 struct md_op_data *op_data;
3311 if (S_ISREG(inode->i_mode)) {
3312 rc = ll_get_default_mdsize(sbi, &ealen);
3315 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
3318 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
3319 0, ealen, LUSTRE_OPC_ANY,
3321 if (IS_ERR(op_data))
3322 RETURN(PTR_ERR(op_data));
3324 op_data->op_valid = valid;
3325 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
3326 * capa for this inode. Because we only keep capas of dirs
3328 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3329 ll_finish_md_op_data(op_data);
3331 rc = ll_inode_revalidate_fini(inode, rc);
3335 rc = ll_prep_inode(&inode, req, NULL, NULL);
3338 ptlrpc_req_finished(req);
3342 static int ll_merge_md_attr(struct inode *inode)
3344 struct cl_attr attr = { 0 };
3347 LASSERT(ll_i2info(inode)->lli_lsm_md != NULL);
3348 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
3353 ll_i2info(inode)->lli_stripe_dir_size = attr.cat_size;
3354 ll_i2info(inode)->lli_stripe_dir_nlink = attr.cat_nlink;
3356 ll_i2info(inode)->lli_lvb.lvb_atime = attr.cat_atime;
3357 ll_i2info(inode)->lli_lvb.lvb_mtime = attr.cat_mtime;
3358 ll_i2info(inode)->lli_lvb.lvb_ctime = attr.cat_ctime;
3364 ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3366 struct inode *inode = dentry->d_inode;
3370 rc = __ll_inode_revalidate(dentry, ibits);
3374 /* if object isn't regular file, don't validate size */
3375 if (!S_ISREG(inode->i_mode)) {
3376 if (S_ISDIR(inode->i_mode) &&
3377 ll_i2info(inode)->lli_lsm_md != NULL) {
3378 rc = ll_merge_md_attr(inode);
3383 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
3384 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
3385 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
3387 /* In case of restore, the MDT has the right size and has
3388 * already send it back without granting the layout lock,
3389 * inode is up-to-date so glimpse is useless.
3390 * Also to glimpse we need the layout, in case of a running
3391 * restore the MDT holds the layout lock so the glimpse will
3392 * block up to the end of restore (getattr will block)
3394 if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING))
3395 rc = ll_glimpse_size(inode);
3400 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3402 struct inode *inode = de->d_inode;
3403 struct ll_sb_info *sbi = ll_i2sbi(inode);
3404 struct ll_inode_info *lli = ll_i2info(inode);
3407 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3408 MDS_INODELOCK_LOOKUP);
3409 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3414 stat->dev = inode->i_sb->s_dev;
3415 if (ll_need_32bit_api(sbi))
3416 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3418 stat->ino = inode->i_ino;
3419 stat->mode = inode->i_mode;
3420 stat->uid = inode->i_uid;
3421 stat->gid = inode->i_gid;
3422 stat->rdev = inode->i_rdev;
3423 stat->atime = inode->i_atime;
3424 stat->mtime = inode->i_mtime;
3425 stat->ctime = inode->i_ctime;
3426 stat->blksize = 1 << inode->i_blkbits;
3427 stat->blocks = inode->i_blocks;
3429 if (S_ISDIR(inode->i_mode) &&
3430 ll_i2info(inode)->lli_lsm_md != NULL) {
3431 stat->nlink = lli->lli_stripe_dir_nlink;
3432 stat->size = lli->lli_stripe_dir_size;
3434 stat->nlink = inode->i_nlink;
3435 stat->size = i_size_read(inode);
3441 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3442 __u64 start, __u64 len)
3446 struct ll_user_fiemap *fiemap;
3447 unsigned int extent_count = fieinfo->fi_extents_max;
3449 num_bytes = sizeof(*fiemap) + (extent_count *
3450 sizeof(struct ll_fiemap_extent));
3451 OBD_ALLOC_LARGE(fiemap, num_bytes);
3456 fiemap->fm_flags = fieinfo->fi_flags;
3457 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3458 fiemap->fm_start = start;
3459 fiemap->fm_length = len;
3460 if (extent_count > 0)
3461 memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3462 sizeof(struct ll_fiemap_extent));
3464 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3466 fieinfo->fi_flags = fiemap->fm_flags;
3467 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3468 if (extent_count > 0)
3469 memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3470 fiemap->fm_mapped_extents *
3471 sizeof(struct ll_fiemap_extent));
3473 OBD_FREE_LARGE(fiemap, num_bytes);
3477 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3479 struct ll_inode_info *lli = ll_i2info(inode);
3480 struct posix_acl *acl = NULL;
3483 spin_lock(&lli->lli_lock);
3484 /* VFS' acl_permission_check->check_acl will release the refcount */
3485 acl = posix_acl_dup(lli->lli_posix_acl);
3486 spin_unlock(&lli->lli_lock);
3491 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
3493 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3494 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
3496 ll_check_acl(struct inode *inode, int mask)
3499 # ifdef CONFIG_FS_POSIX_ACL
3500 struct posix_acl *acl;
3504 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3505 if (flags & IPERM_FLAG_RCU)
3508 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
3513 rc = posix_acl_permission(inode, acl, mask);
3514 posix_acl_release(acl);
3517 # else /* !CONFIG_FS_POSIX_ACL */
3519 # endif /* CONFIG_FS_POSIX_ACL */
3521 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
3523 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
3524 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
3526 # ifdef HAVE_INODE_PERMISION_2ARGS
3527 int ll_inode_permission(struct inode *inode, int mask)
3529 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
3534 struct ll_sb_info *sbi;
3535 struct root_squash_info *squash;
3536 struct cred *cred = NULL;
3537 const struct cred *old_cred = NULL;
3539 bool squash_id = false;
3542 #ifdef MAY_NOT_BLOCK
3543 if (mask & MAY_NOT_BLOCK)
3545 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
3546 if (flags & IPERM_FLAG_RCU)
3550 /* as root inode are NOT getting validated in lookup operation,
3551 * need to do it before permission check. */
3553 if (inode == inode->i_sb->s_root->d_inode) {
3554 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3555 MDS_INODELOCK_LOOKUP);
3560 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3561 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3563 /* squash fsuid/fsgid if needed */
3564 sbi = ll_i2sbi(inode);
3565 squash = &sbi->ll_squash;
3566 if (unlikely(squash->rsi_uid != 0 &&
3567 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3568 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3572 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3573 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3574 squash->rsi_uid, squash->rsi_gid);
3576 /* update current process's credentials
3577 * and FS capability */
3578 cred = prepare_creds();
3582 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3583 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3584 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3585 if ((1 << cap) & CFS_CAP_FS_MASK)
3586 cap_lower(cred->cap_effective, cap);
3588 old_cred = override_creds(cred);
3591 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
3593 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
3594 rc = lustre_check_remote_perm(inode, mask);
3596 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
3598 /* restore current process's credentials and FS capability */
3600 revert_creds(old_cred);
3607 /* -o localflock - only provides locally consistent flock locks */
3608 struct file_operations ll_file_operations = {
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,
3623 struct file_operations ll_file_operations_flock = {
3624 .read = ll_file_read,
3625 .aio_read = ll_file_aio_read,
3626 .write = ll_file_write,
3627 .aio_write = ll_file_aio_write,
3628 .unlocked_ioctl = ll_file_ioctl,
3629 .open = ll_file_open,
3630 .release = ll_file_release,
3631 .mmap = ll_file_mmap,
3632 .llseek = ll_file_seek,
3633 .splice_read = ll_file_splice_read,
3636 .flock = ll_file_flock,
3637 .lock = ll_file_flock
3640 /* These are for -o noflock - to return ENOSYS on flock calls */
3641 struct file_operations ll_file_operations_noflock = {
3642 .read = ll_file_read,
3643 .aio_read = ll_file_aio_read,
3644 .write = ll_file_write,
3645 .aio_write = ll_file_aio_write,
3646 .unlocked_ioctl = ll_file_ioctl,
3647 .open = ll_file_open,
3648 .release = ll_file_release,
3649 .mmap = ll_file_mmap,
3650 .llseek = ll_file_seek,
3651 .splice_read = ll_file_splice_read,
3654 .flock = ll_file_noflock,
3655 .lock = ll_file_noflock
3658 struct inode_operations ll_file_inode_operations = {
3659 .setattr = ll_setattr,
3660 .getattr = ll_getattr,
3661 .permission = ll_inode_permission,
3662 .setxattr = ll_setxattr,
3663 .getxattr = ll_getxattr,
3664 .listxattr = ll_listxattr,
3665 .removexattr = ll_removexattr,
3666 .fiemap = ll_fiemap,
3667 #ifdef HAVE_IOP_GET_ACL
3668 .get_acl = ll_get_acl,
3672 /* dynamic ioctl number support routins */
3673 static struct llioc_ctl_data {
3674 struct rw_semaphore ioc_sem;
3675 struct list_head ioc_head;
3677 __RWSEM_INITIALIZER(llioc.ioc_sem),
3678 LIST_HEAD_INIT(llioc.ioc_head)
3683 struct list_head iocd_list;
3684 unsigned int iocd_size;
3685 llioc_callback_t iocd_cb;
3686 unsigned int iocd_count;
3687 unsigned int iocd_cmd[0];
3690 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3693 struct llioc_data *in_data = NULL;
3696 if (cb == NULL || cmd == NULL ||
3697 count > LLIOC_MAX_CMD || count < 0)
3700 size = sizeof(*in_data) + count * sizeof(unsigned int);
3701 OBD_ALLOC(in_data, size);
3702 if (in_data == NULL)
3705 memset(in_data, 0, sizeof(*in_data));
3706 in_data->iocd_size = size;
3707 in_data->iocd_cb = cb;
3708 in_data->iocd_count = count;
3709 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3711 down_write(&llioc.ioc_sem);
3712 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3713 up_write(&llioc.ioc_sem);
3718 void ll_iocontrol_unregister(void *magic)
3720 struct llioc_data *tmp;
3725 down_write(&llioc.ioc_sem);
3726 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3728 unsigned int size = tmp->iocd_size;
3730 list_del(&tmp->iocd_list);
3731 up_write(&llioc.ioc_sem);
3733 OBD_FREE(tmp, size);
3737 up_write(&llioc.ioc_sem);
3739 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3742 EXPORT_SYMBOL(ll_iocontrol_register);
3743 EXPORT_SYMBOL(ll_iocontrol_unregister);
3745 static enum llioc_iter
3746 ll_iocontrol_call(struct inode *inode, struct file *file,
3747 unsigned int cmd, unsigned long arg, int *rcp)
3749 enum llioc_iter ret = LLIOC_CONT;
3750 struct llioc_data *data;
3751 int rc = -EINVAL, i;
3753 down_read(&llioc.ioc_sem);
3754 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3755 for (i = 0; i < data->iocd_count; i++) {
3756 if (cmd != data->iocd_cmd[i])
3759 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3763 if (ret == LLIOC_STOP)
3766 up_read(&llioc.ioc_sem);
3773 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3775 struct ll_inode_info *lli = ll_i2info(inode);
3776 struct cl_env_nest nest;
3781 if (lli->lli_clob == NULL)
3784 env = cl_env_nested_get(&nest);
3786 RETURN(PTR_ERR(env));
3788 result = cl_conf_set(env, lli->lli_clob, conf);
3789 cl_env_nested_put(&nest, env);
3791 if (conf->coc_opc == OBJECT_CONF_SET) {
3792 struct ldlm_lock *lock = conf->coc_lock;
3794 LASSERT(lock != NULL);
3795 LASSERT(ldlm_has_layout(lock));
3797 struct lustre_md *md = conf->u.coc_md;
3798 __u32 gen = LL_LAYOUT_GEN_EMPTY;
3800 /* it can only be allowed to match after layout is
3801 * applied to inode otherwise false layout would be
3802 * seen. Applying layout shoud happen before dropping
3803 * the intent lock. */
3804 ldlm_lock_allow_match(lock);
3806 lli->lli_has_smd = lsm_has_objects(md->lsm);
3807 if (md->lsm != NULL)
3808 gen = md->lsm->lsm_layout_gen;
3811 DFID ": layout version change: %u -> %u\n",
3812 PFID(&lli->lli_fid), ll_layout_version_get(lli),
3814 ll_layout_version_set(lli, gen);
3820 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3821 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3824 struct ll_sb_info *sbi = ll_i2sbi(inode);
3825 struct obd_capa *oc;
3826 struct ptlrpc_request *req;
3827 struct mdt_body *body;
3834 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3835 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3836 lock->l_lvb_data, lock->l_lvb_len);
3838 if ((lock->l_lvb_data != NULL) && ldlm_is_lvb_ready(lock))
3841 /* if layout lock was granted right away, the layout is returned
3842 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3843 * blocked and then granted via completion ast, we have to fetch
3844 * layout here. Please note that we can't use the LVB buffer in
3845 * completion AST because it doesn't have a large enough buffer */
3846 oc = ll_mdscapa_get(inode);
3847 rc = ll_get_default_mdsize(sbi, &lmmsize);
3849 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
3850 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3856 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3858 GOTO(out, rc = -EPROTO);
3860 lmmsize = body->mbo_eadatasize;
3861 if (lmmsize == 0) /* empty layout */
3864 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3866 GOTO(out, rc = -EFAULT);
3868 OBD_ALLOC_LARGE(lvbdata, lmmsize);
3869 if (lvbdata == NULL)
3870 GOTO(out, rc = -ENOMEM);
3872 memcpy(lvbdata, lmm, lmmsize);
3873 lock_res_and_lock(lock);
3874 if (lock->l_lvb_data != NULL)
3875 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
3877 lock->l_lvb_data = lvbdata;
3878 lock->l_lvb_len = lmmsize;
3879 unlock_res_and_lock(lock);
3884 ptlrpc_req_finished(req);
3889 * Apply the layout to the inode. Layout lock is held and will be released
3892 static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
3893 struct inode *inode, __u32 *gen, bool reconf)
3895 struct ll_inode_info *lli = ll_i2info(inode);
3896 struct ll_sb_info *sbi = ll_i2sbi(inode);
3897 struct ldlm_lock *lock;
3898 struct lustre_md md = { NULL };
3899 struct cl_object_conf conf;
3902 bool wait_layout = false;
3905 LASSERT(lustre_handle_is_used(lockh));
3907 lock = ldlm_handle2lock(lockh);
3908 LASSERT(lock != NULL);
3909 LASSERT(ldlm_has_layout(lock));
3911 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d",
3912 PFID(&lli->lli_fid), inode, reconf);
3914 /* in case this is a caching lock and reinstate with new inode */
3915 md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
3917 lock_res_and_lock(lock);
3918 lvb_ready = ldlm_is_lvb_ready(lock);
3919 unlock_res_and_lock(lock);
3920 /* checking lvb_ready is racy but this is okay. The worst case is
3921 * that multi processes may configure the file on the same time. */
3923 if (lvb_ready || !reconf) {
3926 /* layout_gen must be valid if layout lock is not
3927 * cancelled and stripe has already set */
3928 *gen = ll_layout_version_get(lli);
3934 rc = ll_layout_fetch(inode, lock);
3938 /* for layout lock, lmm is returned in lock's lvb.
3939 * lvb_data is immutable if the lock is held so it's safe to access it
3940 * without res lock. See the description in ldlm_lock_decref_internal()
3941 * for the condition to free lvb_data of layout lock */
3942 if (lock->l_lvb_data != NULL) {
3943 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3944 lock->l_lvb_data, lock->l_lvb_len);
3946 *gen = LL_LAYOUT_GEN_EMPTY;
3948 *gen = md.lsm->lsm_layout_gen;
3951 CERROR("%s: file "DFID" unpackmd error: %d\n",
3952 ll_get_fsname(inode->i_sb, NULL, 0),
3953 PFID(&lli->lli_fid), rc);
3959 /* set layout to file. Unlikely this will fail as old layout was
3960 * surely eliminated */
3961 memset(&conf, 0, sizeof conf);
3962 conf.coc_opc = OBJECT_CONF_SET;
3963 conf.coc_inode = inode;
3964 conf.coc_lock = lock;
3965 conf.u.coc_md = &md;
3966 rc = ll_layout_conf(inode, &conf);
3969 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3971 /* refresh layout failed, need to wait */
3972 wait_layout = rc == -EBUSY;
3976 LDLM_LOCK_PUT(lock);
3977 ldlm_lock_decref(lockh, mode);
3979 /* wait for IO to complete if it's still being used. */
3981 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3982 ll_get_fsname(inode->i_sb, NULL, 0),
3983 PFID(&lli->lli_fid), inode);
3985 memset(&conf, 0, sizeof conf);
3986 conf.coc_opc = OBJECT_CONF_WAIT;
3987 conf.coc_inode = inode;
3988 rc = ll_layout_conf(inode, &conf);
3992 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
3993 ll_get_fsname(inode->i_sb, NULL, 0),
3994 PFID(&lli->lli_fid), rc);
4000 * This function checks if there exists a LAYOUT lock on the client side,
4001 * or enqueues it if it doesn't have one in cache.
4003 * This function will not hold layout lock so it may be revoked any time after
4004 * this function returns. Any operations depend on layout should be redone
4007 * This function should be called before lov_io_init() to get an uptodate
4008 * layout version, the caller should save the version number and after IO
4009 * is finished, this function should be called again to verify that layout
4010 * is not changed during IO time.
4012 int ll_layout_refresh(struct inode *inode, __u32 *gen)
4014 struct ll_inode_info *lli = ll_i2info(inode);
4015 struct ll_sb_info *sbi = ll_i2sbi(inode);
4016 struct md_op_data *op_data;
4017 struct lookup_intent it;
4018 struct lustre_handle lockh;
4020 struct ldlm_enqueue_info einfo = {
4021 .ei_type = LDLM_IBITS,
4023 .ei_cb_bl = &ll_md_blocking_ast,
4024 .ei_cb_cp = &ldlm_completion_ast,
4029 *gen = ll_layout_version_get(lli);
4030 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
4034 LASSERT(fid_is_sane(ll_inode2fid(inode)));
4035 LASSERT(S_ISREG(inode->i_mode));
4037 /* take layout lock mutex to enqueue layout lock exclusively. */
4038 mutex_lock(&lli->lli_layout_mutex);
4041 /* mostly layout lock is caching on the local side, so try to match
4042 * it before grabbing layout lock mutex. */
4043 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
4044 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
4045 if (mode != 0) { /* hit cached lock */
4046 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4050 mutex_unlock(&lli->lli_layout_mutex);
4054 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
4055 0, 0, LUSTRE_OPC_ANY, NULL);
4056 if (IS_ERR(op_data)) {
4057 mutex_unlock(&lli->lli_layout_mutex);
4058 RETURN(PTR_ERR(op_data));
4061 /* have to enqueue one */
4062 memset(&it, 0, sizeof(it));
4063 it.it_op = IT_LAYOUT;
4064 lockh.cookie = 0ULL;
4066 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
4067 ll_get_fsname(inode->i_sb, NULL, 0),
4068 PFID(&lli->lli_fid), inode);
4070 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
4071 if (it.d.lustre.it_data != NULL)
4072 ptlrpc_req_finished(it.d.lustre.it_data);
4073 it.d.lustre.it_data = NULL;
4075 ll_finish_md_op_data(op_data);
4077 mode = it.d.lustre.it_lock_mode;
4078 it.d.lustre.it_lock_mode = 0;
4079 ll_intent_drop_lock(&it);
4082 /* set lock data in case this is a new lock */
4083 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
4084 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4088 mutex_unlock(&lli->lli_layout_mutex);
4094 * This function send a restore request to the MDT
4096 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
4098 struct hsm_user_request *hur;
4102 len = sizeof(struct hsm_user_request) +
4103 sizeof(struct hsm_user_item);
4104 OBD_ALLOC(hur, len);
4108 hur->hur_request.hr_action = HUA_RESTORE;
4109 hur->hur_request.hr_archive_id = 0;
4110 hur->hur_request.hr_flags = 0;
4111 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
4112 sizeof(hur->hur_user_item[0].hui_fid));
4113 hur->hur_user_item[0].hui_extent.offset = offset;
4114 hur->hur_user_item[0].hui_extent.length = length;
4115 hur->hur_request.hr_itemcount = 1;
4116 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, cl_i2sbi(inode)->ll_md_exp,