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, 2012, 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 "llite_internal.h"
49 #include <lustre/ll_fiemap.h>
51 #include "cl_object.h"
53 struct ll_file_data *ll_file_data_get(void)
55 struct ll_file_data *fd;
57 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, CFS_ALLOC_IO);
58 fd->fd_write_failed = false;
62 static void ll_file_data_put(struct ll_file_data *fd)
65 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
68 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
69 struct lustre_handle *fh)
71 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
72 op_data->op_attr.ia_mode = inode->i_mode;
73 op_data->op_attr.ia_atime = inode->i_atime;
74 op_data->op_attr.ia_mtime = inode->i_mtime;
75 op_data->op_attr.ia_ctime = inode->i_ctime;
76 op_data->op_attr.ia_size = i_size_read(inode);
77 op_data->op_attr_blocks = inode->i_blocks;
78 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags =
79 ll_inode_to_ext_flags(inode->i_flags);
80 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
82 op_data->op_handle = *fh;
83 op_data->op_capa1 = ll_mdscapa_get(inode);
85 if (LLIF_DATA_MODIFIED & ll_i2info(inode)->lli_flags)
86 op_data->op_bias |= MDS_DATA_MODIFIED;
90 * Closes the IO epoch and packs all the attributes into @op_data for
93 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
94 struct obd_client_handle *och)
98 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
99 ATTR_MTIME_SET | ATTR_CTIME_SET;
101 if (!(och->och_flags & FMODE_WRITE))
104 if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
105 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
107 ll_ioepoch_close(inode, op_data, &och, 0);
110 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
111 ll_prep_md_op_data(op_data, inode, NULL, NULL,
112 0, 0, LUSTRE_OPC_ANY, NULL);
116 static int ll_close_inode_openhandle(struct obd_export *md_exp,
118 struct obd_client_handle *och)
120 struct obd_export *exp = ll_i2mdexp(inode);
121 struct md_op_data *op_data;
122 struct ptlrpc_request *req = NULL;
123 struct obd_device *obd = class_exp2obd(exp);
130 * XXX: in case of LMV, is this correct to access
133 CERROR("Invalid MDC connection handle "LPX64"\n",
134 ll_i2mdexp(inode)->exp_handle.h_cookie);
138 OBD_ALLOC_PTR(op_data);
140 GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here.
142 ll_prepare_close(inode, op_data, och);
143 epoch_close = (op_data->op_flags & MF_EPOCH_CLOSE);
144 rc = md_close(md_exp, op_data, och->och_mod, &req);
146 /* This close must have the epoch closed. */
147 LASSERT(epoch_close);
148 /* MDS has instructed us to obtain Size-on-MDS attribute from
149 * OSTs and send setattr to back to MDS. */
150 rc = ll_som_update(inode, op_data);
152 CERROR("inode %lu mdc Size-on-MDS update failed: "
153 "rc = %d\n", inode->i_ino, rc);
157 CERROR("inode %lu mdc close failed: rc = %d\n",
161 /* DATA_MODIFIED flag was successfully sent on close, cancel data
162 * modification flag. */
163 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
164 struct ll_inode_info *lli = ll_i2info(inode);
166 spin_lock(&lli->lli_lock);
167 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
168 spin_unlock(&lli->lli_lock);
171 ll_finish_md_op_data(op_data);
174 rc = ll_objects_destroy(req, inode);
176 CERROR("inode %lu ll_objects destroy: rc = %d\n",
183 if (exp_connect_som(exp) && !epoch_close &&
184 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
185 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
187 md_clear_open_replay_data(md_exp, och);
188 /* Free @och if it is not waiting for DONE_WRITING. */
189 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
192 if (req) /* This is close request */
193 ptlrpc_req_finished(req);
197 int ll_md_real_close(struct inode *inode, int flags)
199 struct ll_inode_info *lli = ll_i2info(inode);
200 struct obd_client_handle **och_p;
201 struct obd_client_handle *och;
206 if (flags & FMODE_WRITE) {
207 och_p = &lli->lli_mds_write_och;
208 och_usecount = &lli->lli_open_fd_write_count;
209 } else if (flags & FMODE_EXEC) {
210 och_p = &lli->lli_mds_exec_och;
211 och_usecount = &lli->lli_open_fd_exec_count;
213 LASSERT(flags & FMODE_READ);
214 och_p = &lli->lli_mds_read_och;
215 och_usecount = &lli->lli_open_fd_read_count;
218 mutex_lock(&lli->lli_och_mutex);
219 if (*och_usecount) { /* There are still users of this handle, so
221 mutex_unlock(&lli->lli_och_mutex);
226 mutex_unlock(&lli->lli_och_mutex);
228 if (och) { /* There might be a race and somebody have freed this och
230 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
237 int ll_md_close(struct obd_export *md_exp, struct inode *inode,
240 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
241 struct ll_inode_info *lli = ll_i2info(inode);
245 /* clear group lock, if present */
246 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
247 ll_put_grouplock(inode, file, fd->fd_grouplock.cg_gid);
249 /* Let's see if we have good enough OPEN lock on the file and if
250 we can skip talking to MDS */
251 if (file->f_dentry->d_inode) { /* Can this ever be false? */
253 int flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
254 struct lustre_handle lockh;
255 struct inode *inode = file->f_dentry->d_inode;
256 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
258 mutex_lock(&lli->lli_och_mutex);
259 if (fd->fd_omode & FMODE_WRITE) {
261 LASSERT(lli->lli_open_fd_write_count);
262 lli->lli_open_fd_write_count--;
263 } else if (fd->fd_omode & FMODE_EXEC) {
265 LASSERT(lli->lli_open_fd_exec_count);
266 lli->lli_open_fd_exec_count--;
269 LASSERT(lli->lli_open_fd_read_count);
270 lli->lli_open_fd_read_count--;
272 mutex_unlock(&lli->lli_och_mutex);
274 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
275 LDLM_IBITS, &policy, lockmode,
277 rc = ll_md_real_close(file->f_dentry->d_inode,
281 CERROR("Releasing a file %p with negative dentry %p. Name %s",
282 file, file->f_dentry, file->f_dentry->d_name.name);
285 LUSTRE_FPRIVATE(file) = NULL;
286 ll_file_data_put(fd);
287 ll_capa_close(inode);
292 /* While this returns an error code, fput() the caller does not, so we need
293 * to make every effort to clean up all of our state here. Also, applications
294 * rarely check close errors and even if an error is returned they will not
295 * re-try the close call.
297 int ll_file_release(struct inode *inode, struct file *file)
299 struct ll_file_data *fd;
300 struct ll_sb_info *sbi = ll_i2sbi(inode);
301 struct ll_inode_info *lli = ll_i2info(inode);
305 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
306 inode->i_generation, inode);
308 #ifdef CONFIG_FS_POSIX_ACL
309 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
310 inode == inode->i_sb->s_root->d_inode) {
311 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
314 if (unlikely(fd->fd_flags & LL_FILE_RMTACL)) {
315 fd->fd_flags &= ~LL_FILE_RMTACL;
316 rct_del(&sbi->ll_rct, cfs_curproc_pid());
317 et_search_free(&sbi->ll_et, cfs_curproc_pid());
322 if (inode->i_sb->s_root != file->f_dentry)
323 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
324 fd = LUSTRE_FPRIVATE(file);
327 /* The last ref on @file, maybe not the the owner pid of statahead.
328 * Different processes can open the same dir, "ll_opendir_key" means:
329 * it is me that should stop the statahead thread. */
330 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
331 lli->lli_opendir_pid != 0)
332 ll_stop_statahead(inode, lli->lli_opendir_key);
334 if (inode->i_sb->s_root == file->f_dentry) {
335 LUSTRE_FPRIVATE(file) = NULL;
336 ll_file_data_put(fd);
340 if (!S_ISDIR(inode->i_mode)) {
341 lov_read_and_clear_async_rc(lli->lli_clob);
342 lli->lli_async_rc = 0;
345 rc = ll_md_close(sbi->ll_md_exp, inode, file);
347 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
348 libcfs_debug_dumplog();
353 static int ll_intent_file_open(struct file *file, void *lmm,
354 int lmmsize, struct lookup_intent *itp)
356 struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
357 struct dentry *parent = file->f_dentry->d_parent;
358 const char *name = file->f_dentry->d_name.name;
359 const int len = file->f_dentry->d_name.len;
360 struct md_op_data *op_data;
361 struct ptlrpc_request *req;
362 __u32 opc = LUSTRE_OPC_ANY;
369 /* Usually we come here only for NFSD, and we want open lock.
370 But we can also get here with pre 2.6.15 patchless kernels, and in
371 that case that lock is also ok */
372 /* We can also get here if there was cached open handle in revalidate_it
373 * but it disappeared while we were getting from there to ll_file_open.
374 * But this means this file was closed and immediatelly opened which
375 * makes a good candidate for using OPEN lock */
376 /* If lmmsize & lmm are not 0, we are just setting stripe info
377 * parameters. No need for the open lock */
378 if (lmm == NULL && lmmsize == 0) {
379 itp->it_flags |= MDS_OPEN_LOCK;
380 if (itp->it_flags & FMODE_WRITE)
381 opc = LUSTRE_OPC_CREATE;
384 op_data = ll_prep_md_op_data(NULL, parent->d_inode,
385 file->f_dentry->d_inode, name, len,
388 RETURN(PTR_ERR(op_data));
390 itp->it_flags |= MDS_OPEN_BY_FID;
391 rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
392 0 /*unused */, &req, ll_md_blocking_ast, 0);
393 ll_finish_md_op_data(op_data);
395 /* reason for keep own exit path - don`t flood log
396 * with messages with -ESTALE errors.
398 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
399 it_open_error(DISP_OPEN_OPEN, itp))
401 ll_release_openhandle(file->f_dentry, itp);
405 if (it_disposition(itp, DISP_LOOKUP_NEG))
406 GOTO(out, rc = -ENOENT);
408 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
409 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
410 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
414 rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp);
415 if (!rc && itp->d.lustre.it_lock_mode)
416 ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode,
420 ptlrpc_req_finished(itp->d.lustre.it_data);
421 it_clear_disposition(itp, DISP_ENQ_COMPLETE);
422 ll_intent_drop_lock(itp);
428 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
429 * not believe attributes if a few ioepoch holders exist. Attributes for
430 * previous ioepoch if new one is opened are also skipped by MDS.
432 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
434 if (ioepoch && lli->lli_ioepoch != ioepoch) {
435 lli->lli_ioepoch = ioepoch;
436 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
437 ioepoch, PFID(&lli->lli_fid));
441 static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli,
442 struct lookup_intent *it, struct obd_client_handle *och)
444 struct ptlrpc_request *req = it->d.lustre.it_data;
445 struct mdt_body *body;
449 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
450 LASSERT(body != NULL); /* reply already checked out */
452 memcpy(&och->och_fh, &body->handle, sizeof(body->handle));
453 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
454 och->och_fid = lli->lli_fid;
455 och->och_flags = it->it_flags;
456 ll_ioepoch_open(lli, body->ioepoch);
458 return md_set_open_replay_data(md_exp, och, req);
461 int ll_local_open(struct file *file, struct lookup_intent *it,
462 struct ll_file_data *fd, struct obd_client_handle *och)
464 struct inode *inode = file->f_dentry->d_inode;
465 struct ll_inode_info *lli = ll_i2info(inode);
468 LASSERT(!LUSTRE_FPRIVATE(file));
473 struct ptlrpc_request *req = it->d.lustre.it_data;
474 struct mdt_body *body;
477 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, lli, it, och);
481 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
482 if ((it->it_flags & FMODE_WRITE) &&
483 (body->valid & OBD_MD_FLSIZE))
484 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
485 lli->lli_ioepoch, PFID(&lli->lli_fid));
488 LUSTRE_FPRIVATE(file) = fd;
489 ll_readahead_init(inode, &fd->fd_ras);
490 fd->fd_omode = it->it_flags;
494 /* Open a file, and (for the very first open) create objects on the OSTs at
495 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
496 * creation or open until ll_lov_setstripe() ioctl is called.
498 * If we already have the stripe MD locally then we don't request it in
499 * md_open(), by passing a lmm_size = 0.
501 * It is up to the application to ensure no other processes open this file
502 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
503 * used. We might be able to avoid races of that sort by getting lli_open_sem
504 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
505 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
507 int ll_file_open(struct inode *inode, struct file *file)
509 struct ll_inode_info *lli = ll_i2info(inode);
510 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
511 .it_flags = file->f_flags };
512 struct obd_client_handle **och_p = NULL;
513 __u64 *och_usecount = NULL;
514 struct ll_file_data *fd;
515 int rc = 0, opendir_set = 0;
518 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), flags %o\n", inode->i_ino,
519 inode->i_generation, inode, file->f_flags);
521 it = file->private_data; /* XXX: compat macro */
522 file->private_data = NULL; /* prevent ll_local_open assertion */
524 fd = ll_file_data_get();
526 GOTO(out_och_free, rc = -ENOMEM);
529 if (S_ISDIR(inode->i_mode)) {
530 spin_lock(&lli->lli_sa_lock);
531 if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL &&
532 lli->lli_opendir_pid == 0) {
533 lli->lli_opendir_key = fd;
534 lli->lli_opendir_pid = cfs_curproc_pid();
537 spin_unlock(&lli->lli_sa_lock);
540 if (inode->i_sb->s_root == file->f_dentry) {
541 LUSTRE_FPRIVATE(file) = fd;
545 if (!it || !it->d.lustre.it_disposition) {
546 /* Convert f_flags into access mode. We cannot use file->f_mode,
547 * because everything but O_ACCMODE mask was stripped from
549 if ((oit.it_flags + 1) & O_ACCMODE)
551 if (file->f_flags & O_TRUNC)
552 oit.it_flags |= FMODE_WRITE;
554 /* kernel only call f_op->open in dentry_open. filp_open calls
555 * dentry_open after call to open_namei that checks permissions.
556 * Only nfsd_open call dentry_open directly without checking
557 * permissions and because of that this code below is safe. */
558 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
559 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
561 /* We do not want O_EXCL here, presumably we opened the file
562 * already? XXX - NFS implications? */
563 oit.it_flags &= ~O_EXCL;
565 /* bug20584, if "it_flags" contains O_CREAT, the file will be
566 * created if necessary, then "IT_CREAT" should be set to keep
567 * consistent with it */
568 if (oit.it_flags & O_CREAT)
569 oit.it_op |= IT_CREAT;
575 /* Let's see if we have file open on MDS already. */
576 if (it->it_flags & FMODE_WRITE) {
577 och_p = &lli->lli_mds_write_och;
578 och_usecount = &lli->lli_open_fd_write_count;
579 } else if (it->it_flags & FMODE_EXEC) {
580 och_p = &lli->lli_mds_exec_och;
581 och_usecount = &lli->lli_open_fd_exec_count;
583 och_p = &lli->lli_mds_read_och;
584 och_usecount = &lli->lli_open_fd_read_count;
587 mutex_lock(&lli->lli_och_mutex);
588 if (*och_p) { /* Open handle is present */
589 if (it_disposition(it, DISP_OPEN_OPEN)) {
590 /* Well, there's extra open request that we do not need,
591 let's close it somehow. This will decref request. */
592 rc = it_open_error(DISP_OPEN_OPEN, it);
594 mutex_unlock(&lli->lli_och_mutex);
595 GOTO(out_openerr, rc);
598 ll_release_openhandle(file->f_dentry, it);
602 rc = ll_local_open(file, it, fd, NULL);
605 mutex_unlock(&lli->lli_och_mutex);
606 GOTO(out_openerr, rc);
609 LASSERT(*och_usecount == 0);
610 if (!it->d.lustre.it_disposition) {
611 /* We cannot just request lock handle now, new ELC code
612 means that one of other OPEN locks for this file
613 could be cancelled, and since blocking ast handler
614 would attempt to grab och_mutex as well, that would
615 result in a deadlock */
616 mutex_unlock(&lli->lli_och_mutex);
617 it->it_create_mode |= M_CHECK_STALE;
618 rc = ll_intent_file_open(file, NULL, 0, it);
619 it->it_create_mode &= ~M_CHECK_STALE;
621 GOTO(out_openerr, rc);
625 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
627 GOTO(out_och_free, rc = -ENOMEM);
631 /* md_intent_lock() didn't get a request ref if there was an
632 * open error, so don't do cleanup on the request here
634 /* XXX (green): Should not we bail out on any error here, not
635 * just open error? */
636 rc = it_open_error(DISP_OPEN_OPEN, it);
638 GOTO(out_och_free, rc);
640 LASSERT(it_disposition(it, DISP_ENQ_OPEN_REF));
642 rc = ll_local_open(file, it, fd, *och_p);
644 GOTO(out_och_free, rc);
646 mutex_unlock(&lli->lli_och_mutex);
649 /* Must do this outside lli_och_mutex lock to prevent deadlock where
650 different kind of OPEN lock for this same inode gets cancelled
651 by ldlm_cancel_lru */
652 if (!S_ISREG(inode->i_mode))
653 GOTO(out_och_free, rc);
657 if (!lli->lli_has_smd) {
658 if (file->f_flags & O_LOV_DELAY_CREATE ||
659 !(file->f_mode & FMODE_WRITE)) {
660 CDEBUG(D_INODE, "object creation was delayed\n");
661 GOTO(out_och_free, rc);
664 file->f_flags &= ~O_LOV_DELAY_CREATE;
665 GOTO(out_och_free, rc);
669 if (och_p && *och_p) {
670 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
671 *och_p = NULL; /* OBD_FREE writes some magic there */
674 mutex_unlock(&lli->lli_och_mutex);
677 if (opendir_set != 0)
678 ll_stop_statahead(inode, lli->lli_opendir_key);
680 ll_file_data_put(fd);
682 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
685 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
686 ptlrpc_req_finished(it->d.lustre.it_data);
687 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
693 /* Fills the obdo with the attributes for the lsm */
694 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
695 struct obd_capa *capa, struct obdo *obdo,
696 __u64 ioepoch, int sync)
698 struct ptlrpc_request_set *set;
699 struct obd_info oinfo = { { { 0 } } };
704 LASSERT(lsm != NULL);
708 oinfo.oi_oa->o_id = lsm->lsm_object_id;
709 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
710 oinfo.oi_oa->o_mode = S_IFREG;
711 oinfo.oi_oa->o_ioepoch = ioepoch;
712 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
713 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
714 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
715 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
716 OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
717 OBD_MD_FLDATAVERSION;
718 oinfo.oi_capa = capa;
720 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
721 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
724 set = ptlrpc_prep_set();
726 CERROR("can't allocate ptlrpc set\n");
729 rc = obd_getattr_async(exp, &oinfo, set);
731 rc = ptlrpc_set_wait(set);
732 ptlrpc_set_destroy(set);
735 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
736 OBD_MD_FLATIME | OBD_MD_FLMTIME |
737 OBD_MD_FLCTIME | OBD_MD_FLSIZE |
738 OBD_MD_FLDATAVERSION);
743 * Performs the getattr on the inode and updates its fields.
744 * If @sync != 0, perform the getattr under the server-side lock.
746 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
747 __u64 ioepoch, int sync)
749 struct obd_capa *capa = ll_mdscapa_get(inode);
750 struct lov_stripe_md *lsm;
754 lsm = ccc_inode_lsm_get(inode);
755 rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode),
756 capa, obdo, ioepoch, sync);
759 obdo_refresh_inode(inode, obdo, obdo->o_valid);
761 "objid "LPX64" size %llu, blocks %llu, blksize %lu\n",
762 lsm ? lsm->lsm_object_id : 0, i_size_read(inode),
763 (unsigned long long)inode->i_blocks,
764 (unsigned long)ll_inode_blksize(inode));
766 ccc_inode_lsm_put(inode, lsm);
770 int ll_merge_lvb(struct inode *inode)
772 struct ll_inode_info *lli = ll_i2info(inode);
773 struct ll_sb_info *sbi = ll_i2sbi(inode);
774 struct lov_stripe_md *lsm;
780 lsm = ccc_inode_lsm_get(inode);
781 ll_inode_size_lock(inode);
782 inode_init_lvb(inode, &lvb);
784 /* merge timestamps the most resently obtained from mds with
785 timestamps obtained from osts */
786 lvb.lvb_atime = lli->lli_lvb.lvb_atime;
787 lvb.lvb_mtime = lli->lli_lvb.lvb_mtime;
788 lvb.lvb_ctime = lli->lli_lvb.lvb_ctime;
790 rc = obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 0);
791 cl_isize_write_nolock(inode, lvb.lvb_size);
793 CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n",
794 PFID(&lli->lli_fid), lvb.lvb_size);
795 inode->i_blocks = lvb.lvb_blocks;
797 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
798 LTIME_S(inode->i_atime) = lvb.lvb_atime;
799 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
800 ll_inode_size_unlock(inode);
801 ccc_inode_lsm_put(inode, lsm);
806 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
809 struct obdo obdo = { 0 };
812 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0);
814 st->st_size = obdo.o_size;
815 st->st_blocks = obdo.o_blocks;
816 st->st_mtime = obdo.o_mtime;
817 st->st_atime = obdo.o_atime;
818 st->st_ctime = obdo.o_ctime;
823 void ll_io_init(struct cl_io *io, const struct file *file, int write)
825 struct inode *inode = file->f_dentry->d_inode;
827 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
829 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
830 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || IS_SYNC(inode);
832 io->ci_obj = ll_i2info(inode)->lli_clob;
833 io->ci_lockreq = CILR_MAYBE;
834 if (ll_file_nolock(file)) {
835 io->ci_lockreq = CILR_NEVER;
836 io->ci_no_srvlock = 1;
837 } else if (file->f_flags & O_APPEND) {
838 io->ci_lockreq = CILR_MANDATORY;
843 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
844 struct file *file, enum cl_io_type iot,
845 loff_t *ppos, size_t count)
847 struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
848 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
853 io = ccc_env_thread_io(env);
854 ll_io_init(io, file, iot == CIT_WRITE);
856 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
857 struct vvp_io *vio = vvp_env_io(env);
858 struct ccc_io *cio = ccc_env_io(env);
859 int write_mutex_locked = 0;
861 cio->cui_fd = LUSTRE_FPRIVATE(file);
862 vio->cui_io_subtype = args->via_io_subtype;
864 switch (vio->cui_io_subtype) {
866 cio->cui_iov = args->u.normal.via_iov;
867 cio->cui_nrsegs = args->u.normal.via_nrsegs;
868 cio->cui_tot_nrsegs = cio->cui_nrsegs;
869 #ifndef HAVE_FILE_WRITEV
870 cio->cui_iocb = args->u.normal.via_iocb;
872 if ((iot == CIT_WRITE) &&
873 !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
874 if (mutex_lock_interruptible(&lli->
876 GOTO(out, result = -ERESTARTSYS);
877 write_mutex_locked = 1;
878 } else if (iot == CIT_READ) {
879 down_read(&lli->lli_trunc_sem);
883 vio->u.sendfile.cui_actor = args->u.sendfile.via_actor;
884 vio->u.sendfile.cui_target = args->u.sendfile.via_target;
887 vio->u.splice.cui_pipe = args->u.splice.via_pipe;
888 vio->u.splice.cui_flags = args->u.splice.via_flags;
891 CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
894 result = cl_io_loop(env, io);
895 if (write_mutex_locked)
896 mutex_unlock(&lli->lli_write_mutex);
897 else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ)
898 up_read(&lli->lli_trunc_sem);
900 /* cl_io_rw_init() handled IO */
901 result = io->ci_result;
904 if (io->ci_nob > 0) {
906 *ppos = io->u.ci_wr.wr.crw_pos;
912 if (iot == CIT_READ) {
914 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
915 LPROC_LL_READ_BYTES, result);
916 } else if (iot == CIT_WRITE) {
918 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
919 LPROC_LL_WRITE_BYTES, result);
920 fd->fd_write_failed = false;
922 fd->fd_write_failed = true;
931 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
933 static int ll_file_get_iov_count(const struct iovec *iov,
934 unsigned long *nr_segs, size_t *count)
939 for (seg = 0; seg < *nr_segs; seg++) {
940 const struct iovec *iv = &iov[seg];
943 * If any segment has a negative length, or the cumulative
944 * length ever wraps negative then return -EINVAL.
947 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
949 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
954 cnt -= iv->iov_len; /* This segment is no good */
961 #ifdef HAVE_FILE_READV
962 static ssize_t ll_file_readv(struct file *file, const struct iovec *iov,
963 unsigned long nr_segs, loff_t *ppos)
966 struct vvp_io_args *args;
972 result = ll_file_get_iov_count(iov, &nr_segs, &count);
976 env = cl_env_get(&refcheck);
978 RETURN(PTR_ERR(env));
980 args = vvp_env_args(env, IO_NORMAL);
981 args->u.normal.via_iov = (struct iovec *)iov;
982 args->u.normal.via_nrsegs = nr_segs;
984 result = ll_file_io_generic(env, args, file, CIT_READ, ppos, count);
985 cl_env_put(env, &refcheck);
989 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
993 struct iovec *local_iov;
998 env = cl_env_get(&refcheck);
1000 RETURN(PTR_ERR(env));
1002 local_iov = &vvp_env_info(env)->vti_local_iov;
1003 local_iov->iov_base = (void __user *)buf;
1004 local_iov->iov_len = count;
1005 result = ll_file_readv(file, local_iov, 1, ppos);
1006 cl_env_put(env, &refcheck);
1011 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1012 unsigned long nr_segs, loff_t pos)
1015 struct vvp_io_args *args;
1021 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1025 env = cl_env_get(&refcheck);
1027 RETURN(PTR_ERR(env));
1029 args = vvp_env_args(env, IO_NORMAL);
1030 args->u.normal.via_iov = (struct iovec *)iov;
1031 args->u.normal.via_nrsegs = nr_segs;
1032 args->u.normal.via_iocb = iocb;
1034 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1035 &iocb->ki_pos, count);
1036 cl_env_put(env, &refcheck);
1040 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1044 struct iovec *local_iov;
1045 struct kiocb *kiocb;
1050 env = cl_env_get(&refcheck);
1052 RETURN(PTR_ERR(env));
1054 local_iov = &vvp_env_info(env)->vti_local_iov;
1055 kiocb = &vvp_env_info(env)->vti_kiocb;
1056 local_iov->iov_base = (void __user *)buf;
1057 local_iov->iov_len = count;
1058 init_sync_kiocb(kiocb, file);
1059 kiocb->ki_pos = *ppos;
1060 kiocb->ki_left = count;
1062 result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos);
1063 *ppos = kiocb->ki_pos;
1065 cl_env_put(env, &refcheck);
1071 * Write to a file (through the page cache).
1073 #ifdef HAVE_FILE_WRITEV
1074 static ssize_t ll_file_writev(struct file *file, const struct iovec *iov,
1075 unsigned long nr_segs, loff_t *ppos)
1078 struct vvp_io_args *args;
1084 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1088 env = cl_env_get(&refcheck);
1090 RETURN(PTR_ERR(env));
1092 args = vvp_env_args(env, IO_NORMAL);
1093 args->u.normal.via_iov = (struct iovec *)iov;
1094 args->u.normal.via_nrsegs = nr_segs;
1096 result = ll_file_io_generic(env, args, file, CIT_WRITE, ppos, count);
1097 cl_env_put(env, &refcheck);
1101 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1105 struct iovec *local_iov;
1110 env = cl_env_get(&refcheck);
1112 RETURN(PTR_ERR(env));
1114 local_iov = &vvp_env_info(env)->vti_local_iov;
1115 local_iov->iov_base = (void __user *)buf;
1116 local_iov->iov_len = count;
1118 result = ll_file_writev(file, local_iov, 1, ppos);
1119 cl_env_put(env, &refcheck);
1123 #else /* AIO stuff */
1124 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1125 unsigned long nr_segs, loff_t pos)
1128 struct vvp_io_args *args;
1134 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1138 env = cl_env_get(&refcheck);
1140 RETURN(PTR_ERR(env));
1142 args = vvp_env_args(env, IO_NORMAL);
1143 args->u.normal.via_iov = (struct iovec *)iov;
1144 args->u.normal.via_nrsegs = nr_segs;
1145 args->u.normal.via_iocb = iocb;
1147 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1148 &iocb->ki_pos, count);
1149 cl_env_put(env, &refcheck);
1153 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1157 struct iovec *local_iov;
1158 struct kiocb *kiocb;
1163 env = cl_env_get(&refcheck);
1165 RETURN(PTR_ERR(env));
1167 local_iov = &vvp_env_info(env)->vti_local_iov;
1168 kiocb = &vvp_env_info(env)->vti_kiocb;
1169 local_iov->iov_base = (void __user *)buf;
1170 local_iov->iov_len = count;
1171 init_sync_kiocb(kiocb, file);
1172 kiocb->ki_pos = *ppos;
1173 kiocb->ki_left = count;
1175 result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos);
1176 *ppos = kiocb->ki_pos;
1178 cl_env_put(env, &refcheck);
1184 #ifdef HAVE_KERNEL_SENDFILE
1186 * Send file content (through pagecache) somewhere with helper
1188 static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count,
1189 read_actor_t actor, void *target)
1192 struct vvp_io_args *args;
1197 env = cl_env_get(&refcheck);
1199 RETURN(PTR_ERR(env));
1201 args = vvp_env_args(env, IO_SENDFILE);
1202 args->u.sendfile.via_target = target;
1203 args->u.sendfile.via_actor = actor;
1205 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1206 cl_env_put(env, &refcheck);
1211 #ifdef HAVE_KERNEL_SPLICE_READ
1213 * Send file content (through pagecache) somewhere with helper
1215 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1216 struct pipe_inode_info *pipe, size_t count,
1220 struct vvp_io_args *args;
1225 env = cl_env_get(&refcheck);
1227 RETURN(PTR_ERR(env));
1229 args = vvp_env_args(env, IO_SPLICE);
1230 args->u.splice.via_pipe = pipe;
1231 args->u.splice.via_flags = flags;
1233 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1234 cl_env_put(env, &refcheck);
1239 static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq,
1242 struct obd_export *exp = ll_i2dtexp(inode);
1243 struct obd_trans_info oti = { 0 };
1244 struct obdo *oa = NULL;
1247 struct lov_stripe_md *lsm = NULL, *lsm2;
1254 lsm = ccc_inode_lsm_get(inode);
1256 GOTO(out, rc = -ENOENT);
1258 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1259 (lsm->lsm_stripe_count));
1261 OBD_ALLOC_LARGE(lsm2, lsm_size);
1263 GOTO(out, rc = -ENOMEM);
1267 oa->o_nlink = ost_idx;
1268 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1269 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1270 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1271 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1272 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1273 memcpy(lsm2, lsm, lsm_size);
1274 ll_inode_size_lock(inode);
1275 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1276 ll_inode_size_unlock(inode);
1278 OBD_FREE_LARGE(lsm2, lsm_size);
1281 ccc_inode_lsm_put(inode, lsm);
1286 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1288 struct ll_recreate_obj ucreat;
1291 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1294 if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
1298 RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0,
1299 ucreat.lrc_ost_idx));
1302 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1309 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1312 if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid)))
1315 id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32);
1316 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1317 RETURN(ll_lov_recreate(inode, id, 0, ost_idx));
1320 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1321 int flags, struct lov_user_md *lum, int lum_size)
1323 struct lov_stripe_md *lsm = NULL;
1324 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1328 lsm = ccc_inode_lsm_get(inode);
1330 ccc_inode_lsm_put(inode, lsm);
1331 CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n",
1336 ll_inode_size_lock(inode);
1337 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1340 rc = oit.d.lustre.it_status;
1342 GOTO(out_req_free, rc);
1344 ll_release_openhandle(file->f_dentry, &oit);
1347 ll_inode_size_unlock(inode);
1348 ll_intent_release(&oit);
1349 ccc_inode_lsm_put(inode, lsm);
1352 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1356 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1357 struct lov_mds_md **lmmp, int *lmm_size,
1358 struct ptlrpc_request **request)
1360 struct ll_sb_info *sbi = ll_i2sbi(inode);
1361 struct mdt_body *body;
1362 struct lov_mds_md *lmm = NULL;
1363 struct ptlrpc_request *req = NULL;
1364 struct md_op_data *op_data;
1367 rc = ll_get_max_mdsize(sbi, &lmmsize);
1371 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1372 strlen(filename), lmmsize,
1373 LUSTRE_OPC_ANY, NULL);
1374 if (IS_ERR(op_data))
1375 RETURN(PTR_ERR(op_data));
1377 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1378 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1379 ll_finish_md_op_data(op_data);
1381 CDEBUG(D_INFO, "md_getattr_name failed "
1382 "on %s: rc %d\n", filename, rc);
1386 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1387 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1389 lmmsize = body->eadatasize;
1391 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1393 GOTO(out, rc = -ENODATA);
1396 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1397 LASSERT(lmm != NULL);
1399 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1400 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1401 GOTO(out, rc = -EPROTO);
1405 * This is coming from the MDS, so is probably in
1406 * little endian. We convert it to host endian before
1407 * passing it to userspace.
1409 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1410 /* if function called for directory - we should
1411 * avoid swab not existent lsm objects */
1412 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1413 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1414 if (S_ISREG(body->mode))
1415 lustre_swab_lov_user_md_objects(
1416 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1417 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1418 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1419 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1420 if (S_ISREG(body->mode))
1421 lustre_swab_lov_user_md_objects(
1422 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1423 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1429 *lmm_size = lmmsize;
1434 static int ll_lov_setea(struct inode *inode, struct file *file,
1437 int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1438 struct lov_user_md *lump;
1439 int lum_size = sizeof(struct lov_user_md) +
1440 sizeof(struct lov_user_ost_data);
1444 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1447 OBD_ALLOC_LARGE(lump, lum_size);
1451 if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) {
1452 OBD_FREE_LARGE(lump, lum_size);
1456 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1458 OBD_FREE_LARGE(lump, lum_size);
1462 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1465 struct lov_user_md_v3 lumv3;
1466 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1467 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1468 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1470 int flags = FMODE_WRITE;
1473 /* first try with v1 which is smaller than v3 */
1474 lum_size = sizeof(struct lov_user_md_v1);
1475 if (copy_from_user(lumv1, lumv1p, lum_size))
1478 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1479 lum_size = sizeof(struct lov_user_md_v3);
1480 if (copy_from_user(&lumv3, lumv3p, lum_size))
1484 rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size);
1486 struct lov_stripe_md *lsm;
1489 put_user(0, &lumv1p->lmm_stripe_count);
1491 ll_layout_refresh(inode, &gen);
1492 lsm = ccc_inode_lsm_get(inode);
1493 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1494 0, lsm, (void *)arg);
1495 ccc_inode_lsm_put(inode, lsm);
1500 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1502 struct lov_stripe_md *lsm;
1506 lsm = ccc_inode_lsm_get(inode);
1508 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1510 ccc_inode_lsm_put(inode, lsm);
1514 int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1516 struct ll_inode_info *lli = ll_i2info(inode);
1517 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1518 struct ccc_grouplock grouplock;
1522 if (ll_file_nolock(file))
1523 RETURN(-EOPNOTSUPP);
1525 spin_lock(&lli->lli_lock);
1526 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1527 CWARN("group lock already existed with gid %lu\n",
1528 fd->fd_grouplock.cg_gid);
1529 spin_unlock(&lli->lli_lock);
1532 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1533 spin_unlock(&lli->lli_lock);
1535 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1536 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1540 spin_lock(&lli->lli_lock);
1541 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1542 spin_unlock(&lli->lli_lock);
1543 CERROR("another thread just won the race\n");
1544 cl_put_grouplock(&grouplock);
1548 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1549 fd->fd_grouplock = grouplock;
1550 spin_unlock(&lli->lli_lock);
1552 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1556 int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1558 struct ll_inode_info *lli = ll_i2info(inode);
1559 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1560 struct ccc_grouplock grouplock;
1563 spin_lock(&lli->lli_lock);
1564 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1565 spin_unlock(&lli->lli_lock);
1566 CWARN("no group lock held\n");
1569 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1571 if (fd->fd_grouplock.cg_gid != arg) {
1572 CWARN("group lock %lu doesn't match current id %lu\n",
1573 arg, fd->fd_grouplock.cg_gid);
1574 spin_unlock(&lli->lli_lock);
1578 grouplock = fd->fd_grouplock;
1579 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1580 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1581 spin_unlock(&lli->lli_lock);
1583 cl_put_grouplock(&grouplock);
1584 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1589 * Close inode open handle
1591 * \param dentry [in] dentry which contains the inode
1592 * \param it [in,out] intent which contains open info and result
1595 * \retval <0 failure
1597 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1599 struct inode *inode = dentry->d_inode;
1600 struct obd_client_handle *och;
1606 /* Root ? Do nothing. */
1607 if (dentry->d_inode->i_sb->s_root == dentry)
1610 /* No open handle to close? Move away */
1611 if (!it_disposition(it, DISP_OPEN_OPEN))
1614 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1616 OBD_ALLOC(och, sizeof(*och));
1618 GOTO(out, rc = -ENOMEM);
1620 ll_och_fill(ll_i2sbi(inode)->ll_md_exp,
1621 ll_i2info(inode), it, och);
1623 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1626 /* this one is in place of ll_file_open */
1627 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1628 ptlrpc_req_finished(it->d.lustre.it_data);
1629 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1635 * Get size for inode for which FIEMAP mapping is requested.
1636 * Make the FIEMAP get_info call and returns the result.
1638 int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1641 struct obd_export *exp = ll_i2dtexp(inode);
1642 struct lov_stripe_md *lsm = NULL;
1643 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1644 int vallen = num_bytes;
1648 /* Checks for fiemap flags */
1649 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1650 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1654 /* Check for FIEMAP_FLAG_SYNC */
1655 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1656 rc = filemap_fdatawrite(inode->i_mapping);
1661 lsm = ccc_inode_lsm_get(inode);
1665 /* If the stripe_count > 1 and the application does not understand
1666 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1668 if (lsm->lsm_stripe_count > 1 &&
1669 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER))
1670 GOTO(out, rc = -EOPNOTSUPP);
1672 fm_key.oa.o_id = lsm->lsm_object_id;
1673 fm_key.oa.o_seq = lsm->lsm_object_seq;
1674 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1676 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1677 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1678 /* If filesize is 0, then there would be no objects for mapping */
1679 if (fm_key.oa.o_size == 0) {
1680 fiemap->fm_mapped_extents = 0;
1684 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1686 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1689 CERROR("obd_get_info failed: rc = %d\n", rc);
1692 ccc_inode_lsm_put(inode, lsm);
1696 int ll_fid2path(struct inode *inode, void *arg)
1698 struct obd_export *exp = ll_i2mdexp(inode);
1699 struct getinfo_fid2path *gfout, *gfin;
1703 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1704 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1707 /* Need to get the buflen */
1708 OBD_ALLOC_PTR(gfin);
1711 if (copy_from_user(gfin, arg, sizeof(*gfin))) {
1716 outsize = sizeof(*gfout) + gfin->gf_pathlen;
1717 OBD_ALLOC(gfout, outsize);
1718 if (gfout == NULL) {
1722 memcpy(gfout, gfin, sizeof(*gfout));
1725 /* Call mdc_iocontrol */
1726 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1729 if (copy_to_user(arg, gfout, outsize))
1733 OBD_FREE(gfout, outsize);
1737 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1739 struct ll_user_fiemap *fiemap_s;
1740 size_t num_bytes, ret_bytes;
1741 unsigned int extent_count;
1744 /* Get the extent count so we can calculate the size of
1745 * required fiemap buffer */
1746 if (get_user(extent_count,
1747 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1749 num_bytes = sizeof(*fiemap_s) + (extent_count *
1750 sizeof(struct ll_fiemap_extent));
1752 OBD_ALLOC_LARGE(fiemap_s, num_bytes);
1753 if (fiemap_s == NULL)
1756 /* get the fiemap value */
1757 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1759 GOTO(error, rc = -EFAULT);
1761 /* If fm_extent_count is non-zero, read the first extent since
1762 * it is used to calculate end_offset and device from previous
1765 if (copy_from_user(&fiemap_s->fm_extents[0],
1766 (char __user *)arg + sizeof(*fiemap_s),
1767 sizeof(struct ll_fiemap_extent)))
1768 GOTO(error, rc = -EFAULT);
1771 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1775 ret_bytes = sizeof(struct ll_user_fiemap);
1777 if (extent_count != 0)
1778 ret_bytes += (fiemap_s->fm_mapped_extents *
1779 sizeof(struct ll_fiemap_extent));
1781 if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
1785 OBD_FREE_LARGE(fiemap_s, num_bytes);
1790 * Read the data_version for inode.
1792 * This value is computed using stripe object version on OST.
1793 * Version is computed using server side locking.
1795 * @param extent_lock Take extent lock. Not needed if a process is already
1796 * holding the OST object group locks.
1798 int ll_data_version(struct inode *inode, __u64 *data_version,
1801 struct lov_stripe_md *lsm = NULL;
1802 struct ll_sb_info *sbi = ll_i2sbi(inode);
1803 struct obdo *obdo = NULL;
1807 /* If no stripe, we consider version is 0. */
1808 lsm = ccc_inode_lsm_get(inode);
1811 CDEBUG(D_INODE, "No object for inode\n");
1815 OBD_ALLOC_PTR(obdo);
1817 ccc_inode_lsm_put(inode, lsm);
1821 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock);
1823 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
1826 *data_version = obdo->o_data_version;
1830 ccc_inode_lsm_put(inode, lsm);
1835 long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1837 struct inode *inode = file->f_dentry->d_inode;
1838 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1843 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino,
1844 inode->i_generation, inode, cmd);
1845 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1847 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1848 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1852 case LL_IOC_GETFLAGS:
1853 /* Get the current value of the file flags */
1854 return put_user(fd->fd_flags, (int *)arg);
1855 case LL_IOC_SETFLAGS:
1856 case LL_IOC_CLRFLAGS:
1857 /* Set or clear specific file flags */
1858 /* XXX This probably needs checks to ensure the flags are
1859 * not abused, and to handle any flag side effects.
1861 if (get_user(flags, (int *) arg))
1864 if (cmd == LL_IOC_SETFLAGS) {
1865 if ((flags & LL_FILE_IGNORE_LOCK) &&
1866 !(file->f_flags & O_DIRECT)) {
1867 CERROR("%s: unable to disable locking on "
1868 "non-O_DIRECT file\n", current->comm);
1872 fd->fd_flags |= flags;
1874 fd->fd_flags &= ~flags;
1877 case LL_IOC_LOV_SETSTRIPE:
1878 RETURN(ll_lov_setstripe(inode, file, arg));
1879 case LL_IOC_LOV_SETEA:
1880 RETURN(ll_lov_setea(inode, file, arg));
1881 case LL_IOC_LOV_GETSTRIPE:
1882 RETURN(ll_lov_getstripe(inode, arg));
1883 case LL_IOC_RECREATE_OBJ:
1884 RETURN(ll_lov_recreate_obj(inode, arg));
1885 case LL_IOC_RECREATE_FID:
1886 RETURN(ll_lov_recreate_fid(inode, arg));
1887 case FSFILT_IOC_FIEMAP:
1888 RETURN(ll_ioctl_fiemap(inode, arg));
1889 case FSFILT_IOC_GETFLAGS:
1890 case FSFILT_IOC_SETFLAGS:
1891 RETURN(ll_iocontrol(inode, file, cmd, arg));
1892 case FSFILT_IOC_GETVERSION_OLD:
1893 case FSFILT_IOC_GETVERSION:
1894 RETURN(put_user(inode->i_generation, (int *)arg));
1895 case LL_IOC_GROUP_LOCK:
1896 RETURN(ll_get_grouplock(inode, file, arg));
1897 case LL_IOC_GROUP_UNLOCK:
1898 RETURN(ll_put_grouplock(inode, file, arg));
1899 case IOC_OBD_STATFS:
1900 RETURN(ll_obd_statfs(inode, (void *)arg));
1902 /* We need to special case any other ioctls we want to handle,
1903 * to send them to the MDS/OST as appropriate and to properly
1904 * network encode the arg field.
1905 case FSFILT_IOC_SETVERSION_OLD:
1906 case FSFILT_IOC_SETVERSION:
1908 case LL_IOC_FLUSHCTX:
1909 RETURN(ll_flush_ctx(inode));
1910 case LL_IOC_PATH2FID: {
1911 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1912 sizeof(struct lu_fid)))
1917 case OBD_IOC_FID2PATH:
1918 RETURN(ll_fid2path(inode, (void *)arg));
1919 case LL_IOC_DATA_VERSION: {
1920 struct ioc_data_version idv;
1923 if (copy_from_user(&idv, (char *)arg, sizeof(idv)))
1926 rc = ll_data_version(inode, &idv.idv_version,
1927 !(idv.idv_flags & LL_DV_NOFLUSH));
1929 if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv)))
1935 case LL_IOC_GET_MDTIDX: {
1938 mdtidx = ll_get_mdt_idx(inode);
1942 if (put_user((int)mdtidx, (int*)arg))
1947 case OBD_IOC_GETDTNAME:
1948 case OBD_IOC_GETMDNAME:
1949 RETURN(ll_get_obd_name(inode, cmd, arg));
1950 case LL_IOC_HSM_STATE_GET: {
1951 struct md_op_data *op_data;
1952 struct hsm_user_state *hus;
1959 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1960 LUSTRE_OPC_ANY, hus);
1961 if (op_data == NULL) {
1966 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
1969 if (copy_to_user((void *)arg, hus, sizeof(*hus)))
1972 ll_finish_md_op_data(op_data);
1976 case LL_IOC_HSM_STATE_SET: {
1977 struct md_op_data *op_data;
1978 struct hsm_state_set *hss;
1984 if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
1989 /* Non-root users are forbidden to set or clear flags which are
1990 * NOT defined in HSM_USER_MASK. */
1991 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK)
1992 && !cfs_capable(CFS_CAP_SYS_ADMIN)) {
1997 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1998 LUSTRE_OPC_ANY, hss);
1999 if (op_data == NULL) {
2004 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2007 ll_finish_md_op_data(op_data);
2017 ll_iocontrol_call(inode, file, cmd, arg, &err))
2020 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2026 #ifndef HAVE_FILE_LLSEEK_SIZE
2027 static inline loff_t
2028 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2030 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2032 if (offset > maxsize)
2035 if (offset != file->f_pos) {
2036 file->f_pos = offset;
2037 file->f_version = 0;
2043 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2044 loff_t maxsize, loff_t eof)
2046 struct inode *inode = file->f_dentry->d_inode;
2054 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2055 * position-querying operation. Avoid rewriting the "same"
2056 * f_pos value back to the file because a concurrent read(),
2057 * write() or lseek() might have altered it
2062 * f_lock protects against read/modify/write race with other
2063 * SEEK_CURs. Note that parallel writes and reads behave
2066 mutex_lock(&inode->i_mutex);
2067 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2068 mutex_unlock(&inode->i_mutex);
2072 * In the generic case the entire file is data, so as long as
2073 * offset isn't at the end of the file then the offset is data.
2080 * There is a virtual hole at the end of the file, so as long as
2081 * offset isn't i_size or larger, return i_size.
2089 return llseek_execute(file, offset, maxsize);
2093 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2095 struct inode *inode = file->f_dentry->d_inode;
2096 loff_t retval, eof = 0;
2099 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2100 (origin == SEEK_CUR) ? file->f_pos : 0);
2101 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%d)\n",
2102 inode->i_ino, inode->i_generation, inode, retval, retval,
2104 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2106 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2107 retval = ll_glimpse_size(inode);
2110 eof = i_size_read(inode);
2113 retval = generic_file_llseek_size(file, offset, origin,
2114 ll_file_maxbytes(inode), eof);
2118 int ll_flush(struct file *file, fl_owner_t id)
2120 struct inode *inode = file->f_dentry->d_inode;
2121 struct ll_inode_info *lli = ll_i2info(inode);
2122 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2125 LASSERT(!S_ISDIR(inode->i_mode));
2127 /* catch async errors that were recorded back when async writeback
2128 * failed for pages in this mapping. */
2129 rc = lli->lli_async_rc;
2130 lli->lli_async_rc = 0;
2131 err = lov_read_and_clear_async_rc(lli->lli_clob);
2135 /* The application has been told write failure already.
2136 * Do not report failure again. */
2137 if (fd->fd_write_failed)
2139 return rc ? -EIO : 0;
2143 * Called to make sure a portion of file has been written out.
2144 * if @local_only is not true, it will send OST_SYNC RPCs to ost.
2146 * Return how many pages have been written.
2148 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2149 enum cl_fsync_mode mode)
2151 struct cl_env_nest nest;
2154 struct obd_capa *capa = NULL;
2155 struct cl_fsync_io *fio;
2159 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2160 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2163 env = cl_env_nested_get(&nest);
2165 RETURN(PTR_ERR(env));
2167 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
2169 io = ccc_env_thread_io(env);
2170 io->ci_obj = cl_i2info(inode)->lli_clob;
2171 io->ci_ignore_layout = 1;
2173 /* initialize parameters for sync */
2174 fio = &io->u.ci_fsync;
2175 fio->fi_capa = capa;
2176 fio->fi_start = start;
2178 fio->fi_fid = ll_inode2fid(inode);
2179 fio->fi_mode = mode;
2180 fio->fi_nr_written = 0;
2182 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2183 result = cl_io_loop(env, io);
2185 result = io->ci_result;
2187 result = fio->fi_nr_written;
2188 cl_io_fini(env, io);
2189 cl_env_nested_put(&nest, env);
2196 #ifdef HAVE_FILE_FSYNC_4ARGS
2197 int ll_fsync(struct file *file, loff_t start, loff_t end, int data)
2198 #elif defined(HAVE_FILE_FSYNC_2ARGS)
2199 int ll_fsync(struct file *file, int data)
2201 int ll_fsync(struct file *file, struct dentry *dentry, int data)
2204 struct inode *inode = file->f_dentry->d_inode;
2205 struct ll_inode_info *lli = ll_i2info(inode);
2206 struct ptlrpc_request *req;
2207 struct obd_capa *oc;
2211 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2212 inode->i_generation, inode);
2213 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2215 #ifdef HAVE_FILE_FSYNC_4ARGS
2216 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2217 mutex_lock(&inode->i_mutex);
2219 /* fsync's caller has already called _fdata{sync,write}, we want
2220 * that IO to finish before calling the osc and mdc sync methods */
2221 rc = filemap_fdatawait(inode->i_mapping);
2224 /* catch async errors that were recorded back when async writeback
2225 * failed for pages in this mapping. */
2226 if (!S_ISDIR(inode->i_mode)) {
2227 err = lli->lli_async_rc;
2228 lli->lli_async_rc = 0;
2231 err = lov_read_and_clear_async_rc(lli->lli_clob);
2236 oc = ll_mdscapa_get(inode);
2237 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2243 ptlrpc_req_finished(req);
2246 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2248 err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2250 if (rc == 0 && err < 0)
2253 fd->fd_write_failed = true;
2255 fd->fd_write_failed = false;
2258 #ifdef HAVE_FILE_FSYNC_4ARGS
2259 mutex_unlock(&inode->i_mutex);
2264 int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2266 struct inode *inode = file->f_dentry->d_inode;
2267 struct ll_sb_info *sbi = ll_i2sbi(inode);
2268 struct ldlm_enqueue_info einfo = { .ei_type = LDLM_FLOCK,
2269 .ei_cb_cp =ldlm_flock_completion_ast,
2270 .ei_cbdata = file_lock };
2271 struct md_op_data *op_data;
2272 struct lustre_handle lockh = {0};
2273 ldlm_policy_data_t flock = {{0}};
2279 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n",
2280 inode->i_ino, file_lock);
2282 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2284 if (file_lock->fl_flags & FL_FLOCK) {
2285 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2286 /* flocks are whole-file locks */
2287 flock.l_flock.end = OFFSET_MAX;
2288 /* For flocks owner is determined by the local file desctiptor*/
2289 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
2290 } else if (file_lock->fl_flags & FL_POSIX) {
2291 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2292 flock.l_flock.start = file_lock->fl_start;
2293 flock.l_flock.end = file_lock->fl_end;
2297 flock.l_flock.pid = file_lock->fl_pid;
2299 /* Somewhat ugly workaround for svc lockd.
2300 * lockd installs custom fl_lmops->lm_compare_owner that checks
2301 * for the fl_owner to be the same (which it always is on local node
2302 * I guess between lockd processes) and then compares pid.
2303 * As such we assign pid to the owner field to make it all work,
2304 * conflict with normal locks is unlikely since pid space and
2305 * pointer space for current->files are not intersecting */
2306 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2307 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2309 switch (file_lock->fl_type) {
2311 einfo.ei_mode = LCK_PR;
2314 /* An unlock request may or may not have any relation to
2315 * existing locks so we may not be able to pass a lock handle
2316 * via a normal ldlm_lock_cancel() request. The request may even
2317 * unlock a byte range in the middle of an existing lock. In
2318 * order to process an unlock request we need all of the same
2319 * information that is given with a normal read or write record
2320 * lock request. To avoid creating another ldlm unlock (cancel)
2321 * message we'll treat a LCK_NL flock request as an unlock. */
2322 einfo.ei_mode = LCK_NL;
2325 einfo.ei_mode = LCK_PW;
2328 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
2329 file_lock->fl_type);
2344 flags = LDLM_FL_BLOCK_NOWAIT;
2350 flags = LDLM_FL_TEST_LOCK;
2351 /* Save the old mode so that if the mode in the lock changes we
2352 * can decrement the appropriate reader or writer refcount. */
2353 file_lock->fl_type = einfo.ei_mode;
2356 CERROR("unknown fcntl lock command: %d\n", cmd);
2360 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2361 LUSTRE_OPC_ANY, NULL);
2362 if (IS_ERR(op_data))
2363 RETURN(PTR_ERR(op_data));
2365 CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#x, mode=%u, "
2366 "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid,
2367 flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2369 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2370 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2372 if ((file_lock->fl_flags & FL_FLOCK) &&
2373 (rc == 0 || file_lock->fl_type == F_UNLCK))
2374 rc2 = flock_lock_file_wait(file, file_lock);
2375 if ((file_lock->fl_flags & FL_POSIX) &&
2376 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
2377 !(flags & LDLM_FL_TEST_LOCK))
2378 rc2 = posix_lock_file_wait(file, file_lock);
2380 if (rc2 && file_lock->fl_type != F_UNLCK) {
2381 einfo.ei_mode = LCK_NL;
2382 md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2383 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2387 ll_finish_md_op_data(op_data);
2392 int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2400 * test if some locks matching bits and l_req_mode are acquired
2401 * - bits can be in different locks
2402 * - if found clear the common lock bits in *bits
2403 * - the bits not found, are kept in *bits
2405 * \param bits [IN] searched lock bits [IN]
2406 * \param l_req_mode [IN] searched lock mode
2407 * \retval boolean, true iff all bits are found
2409 int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode)
2411 struct lustre_handle lockh;
2412 ldlm_policy_data_t policy;
2413 ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
2414 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
2423 fid = &ll_i2info(inode)->lli_fid;
2424 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
2425 ldlm_lockname[mode]);
2427 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2428 for (i = 0; i < MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2429 policy.l_inodebits.bits = *bits & (1 << i);
2430 if (policy.l_inodebits.bits == 0)
2433 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2434 &policy, mode, &lockh)) {
2435 struct ldlm_lock *lock;
2437 lock = ldlm_handle2lock(&lockh);
2440 ~(lock->l_policy_data.l_inodebits.bits);
2441 LDLM_LOCK_PUT(lock);
2443 *bits &= ~policy.l_inodebits.bits;
2450 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
2451 struct lustre_handle *lockh, __u64 flags)
2453 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
2458 fid = &ll_i2info(inode)->lli_fid;
2459 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2461 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
2462 fid, LDLM_IBITS, &policy,
2463 LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
2467 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2469 /* Already unlinked. Just update nlink and return success */
2470 if (rc == -ENOENT) {
2472 /* This path cannot be hit for regular files unless in
2473 * case of obscure races, so no need to to validate
2475 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2477 } else if (rc != 0) {
2478 CERROR("%s: revalidate FID "DFID" error: rc = %d\n",
2479 ll_get_fsname(inode->i_sb, NULL, 0),
2480 PFID(ll_inode2fid(inode)), rc);
2486 int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
2489 struct inode *inode = dentry->d_inode;
2490 struct ptlrpc_request *req = NULL;
2491 struct obd_export *exp;
2495 LASSERT(inode != NULL);
2497 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n",
2498 inode->i_ino, inode->i_generation, inode, dentry->d_name.name);
2500 exp = ll_i2mdexp(inode);
2502 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2503 * But under CMD case, it caused some lock issues, should be fixed
2504 * with new CMD ibits lock. See bug 12718 */
2505 if (exp->exp_connect_flags & OBD_CONNECT_ATTRFID) {
2506 struct lookup_intent oit = { .it_op = IT_GETATTR };
2507 struct md_op_data *op_data;
2509 if (ibits == MDS_INODELOCK_LOOKUP)
2510 oit.it_op = IT_LOOKUP;
2512 /* Call getattr by fid, so do not provide name at all. */
2513 op_data = ll_prep_md_op_data(NULL, dentry->d_parent->d_inode,
2514 dentry->d_inode, NULL, 0, 0,
2515 LUSTRE_OPC_ANY, NULL);
2516 if (IS_ERR(op_data))
2517 RETURN(PTR_ERR(op_data));
2519 oit.it_create_mode |= M_CHECK_STALE;
2520 rc = md_intent_lock(exp, op_data, NULL, 0,
2521 /* we are not interested in name
2524 ll_md_blocking_ast, 0);
2525 ll_finish_md_op_data(op_data);
2526 oit.it_create_mode &= ~M_CHECK_STALE;
2528 rc = ll_inode_revalidate_fini(inode, rc);
2532 rc = ll_revalidate_it_finish(req, &oit, dentry);
2534 ll_intent_release(&oit);
2538 /* Unlinked? Unhash dentry, so it is not picked up later by
2539 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2540 here to preserve get_cwd functionality on 2.6.
2542 if (!dentry->d_inode->i_nlink)
2543 d_lustre_invalidate(dentry);
2545 ll_lookup_finish_locks(&oit, dentry);
2546 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
2547 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
2548 obd_valid valid = OBD_MD_FLGETATTR;
2549 struct md_op_data *op_data;
2552 if (S_ISREG(inode->i_mode)) {
2553 rc = ll_get_max_mdsize(sbi, &ealen);
2556 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2559 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2560 0, ealen, LUSTRE_OPC_ANY,
2562 if (IS_ERR(op_data))
2563 RETURN(PTR_ERR(op_data));
2565 op_data->op_valid = valid;
2566 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
2567 * capa for this inode. Because we only keep capas of dirs
2569 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2570 ll_finish_md_op_data(op_data);
2572 rc = ll_inode_revalidate_fini(inode, rc);
2576 rc = ll_prep_inode(&inode, req, NULL, NULL);
2579 ptlrpc_req_finished(req);
2583 int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
2586 struct inode *inode = dentry->d_inode;
2590 rc = __ll_inode_revalidate_it(dentry, it, ibits);
2594 /* if object isn't regular file, don't validate size */
2595 if (!S_ISREG(inode->i_mode)) {
2596 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
2597 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
2598 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
2600 rc = ll_glimpse_size(inode);
2605 int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
2606 struct lookup_intent *it, struct kstat *stat)
2608 struct inode *inode = de->d_inode;
2609 struct ll_sb_info *sbi = ll_i2sbi(inode);
2610 struct ll_inode_info *lli = ll_i2info(inode);
2613 res = ll_inode_revalidate_it(de, it, MDS_INODELOCK_UPDATE |
2614 MDS_INODELOCK_LOOKUP);
2615 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
2620 stat->dev = inode->i_sb->s_dev;
2621 if (ll_need_32bit_api(sbi))
2622 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
2624 stat->ino = inode->i_ino;
2625 stat->mode = inode->i_mode;
2626 stat->nlink = inode->i_nlink;
2627 stat->uid = inode->i_uid;
2628 stat->gid = inode->i_gid;
2629 stat->rdev = inode->i_rdev;
2630 stat->atime = inode->i_atime;
2631 stat->mtime = inode->i_mtime;
2632 stat->ctime = inode->i_ctime;
2633 stat->blksize = 1 << inode->i_blkbits;
2635 stat->size = i_size_read(inode);
2636 stat->blocks = inode->i_blocks;
2640 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
2642 struct lookup_intent it = { .it_op = IT_GETATTR };
2644 return ll_getattr_it(mnt, de, &it, stat);
2647 #ifdef HAVE_LINUX_FIEMAP_H
2648 int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2649 __u64 start, __u64 len)
2653 struct ll_user_fiemap *fiemap;
2654 unsigned int extent_count = fieinfo->fi_extents_max;
2656 num_bytes = sizeof(*fiemap) + (extent_count *
2657 sizeof(struct ll_fiemap_extent));
2658 OBD_ALLOC_LARGE(fiemap, num_bytes);
2663 fiemap->fm_flags = fieinfo->fi_flags;
2664 fiemap->fm_extent_count = fieinfo->fi_extents_max;
2665 fiemap->fm_start = start;
2666 fiemap->fm_length = len;
2667 memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
2668 sizeof(struct ll_fiemap_extent));
2670 rc = ll_do_fiemap(inode, fiemap, num_bytes);
2672 fieinfo->fi_flags = fiemap->fm_flags;
2673 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
2674 memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
2675 fiemap->fm_mapped_extents * sizeof(struct ll_fiemap_extent));
2677 OBD_FREE_LARGE(fiemap, num_bytes);
2682 struct posix_acl * ll_get_acl(struct inode *inode, int type)
2684 struct ll_inode_info *lli = ll_i2info(inode);
2685 struct posix_acl *acl = NULL;
2688 spin_lock(&lli->lli_lock);
2689 /* VFS' acl_permission_check->check_acl will release the refcount */
2690 acl = posix_acl_dup(lli->lli_posix_acl);
2691 spin_unlock(&lli->lli_lock);
2696 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
2698 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
2699 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
2701 ll_check_acl(struct inode *inode, int mask)
2704 # ifdef CONFIG_FS_POSIX_ACL
2705 struct posix_acl *acl;
2709 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
2710 if (flags & IPERM_FLAG_RCU)
2713 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
2718 rc = posix_acl_permission(inode, acl, mask);
2719 posix_acl_release(acl);
2722 # else /* !CONFIG_FS_POSIX_ACL */
2724 # endif /* CONFIG_FS_POSIX_ACL */
2726 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
2728 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
2729 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
2731 # ifdef HAVE_INODE_PERMISION_2ARGS
2732 int ll_inode_permission(struct inode *inode, int mask)
2734 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
2741 #ifdef MAY_NOT_BLOCK
2742 if (mask & MAY_NOT_BLOCK)
2744 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
2745 if (flags & IPERM_FLAG_RCU)
2749 /* as root inode are NOT getting validated in lookup operation,
2750 * need to do it before permission check. */
2752 if (inode == inode->i_sb->s_root->d_inode) {
2753 struct lookup_intent it = { .it_op = IT_LOOKUP };
2755 rc = __ll_inode_revalidate_it(inode->i_sb->s_root, &it,
2756 MDS_INODELOCK_LOOKUP);
2761 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), inode mode %x mask %o\n",
2762 inode->i_ino, inode->i_generation, inode, inode->i_mode, mask);
2764 if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
2765 return lustre_check_remote_perm(inode, mask);
2767 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
2768 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
2773 #ifdef HAVE_FILE_READV
2774 #define READ_METHOD readv
2775 #define READ_FUNCTION ll_file_readv
2776 #define WRITE_METHOD writev
2777 #define WRITE_FUNCTION ll_file_writev
2779 #define READ_METHOD aio_read
2780 #define READ_FUNCTION ll_file_aio_read
2781 #define WRITE_METHOD aio_write
2782 #define WRITE_FUNCTION ll_file_aio_write
2785 /* -o localflock - only provides locally consistent flock locks */
2786 struct file_operations ll_file_operations = {
2787 .read = ll_file_read,
2788 .READ_METHOD = READ_FUNCTION,
2789 .write = ll_file_write,
2790 .WRITE_METHOD = WRITE_FUNCTION,
2791 .unlocked_ioctl = ll_file_ioctl,
2792 .open = ll_file_open,
2793 .release = ll_file_release,
2794 .mmap = ll_file_mmap,
2795 .llseek = ll_file_seek,
2796 #ifdef HAVE_KERNEL_SENDFILE
2797 .sendfile = ll_file_sendfile,
2799 #ifdef HAVE_KERNEL_SPLICE_READ
2800 .splice_read = ll_file_splice_read,
2806 struct file_operations ll_file_operations_flock = {
2807 .read = ll_file_read,
2808 .READ_METHOD = READ_FUNCTION,
2809 .write = ll_file_write,
2810 .WRITE_METHOD = WRITE_FUNCTION,
2811 .unlocked_ioctl = ll_file_ioctl,
2812 .open = ll_file_open,
2813 .release = ll_file_release,
2814 .mmap = ll_file_mmap,
2815 .llseek = ll_file_seek,
2816 #ifdef HAVE_KERNEL_SENDFILE
2817 .sendfile = ll_file_sendfile,
2819 #ifdef HAVE_KERNEL_SPLICE_READ
2820 .splice_read = ll_file_splice_read,
2824 .flock = ll_file_flock,
2825 .lock = ll_file_flock
2828 /* These are for -o noflock - to return ENOSYS on flock calls */
2829 struct file_operations ll_file_operations_noflock = {
2830 .read = ll_file_read,
2831 .READ_METHOD = READ_FUNCTION,
2832 .write = ll_file_write,
2833 .WRITE_METHOD = WRITE_FUNCTION,
2834 .unlocked_ioctl = ll_file_ioctl,
2835 .open = ll_file_open,
2836 .release = ll_file_release,
2837 .mmap = ll_file_mmap,
2838 .llseek = ll_file_seek,
2839 #ifdef HAVE_KERNEL_SENDFILE
2840 .sendfile = ll_file_sendfile,
2842 #ifdef HAVE_KERNEL_SPLICE_READ
2843 .splice_read = ll_file_splice_read,
2847 .flock = ll_file_noflock,
2848 .lock = ll_file_noflock
2851 struct inode_operations ll_file_inode_operations = {
2852 .setattr = ll_setattr,
2853 .getattr = ll_getattr,
2854 .permission = ll_inode_permission,
2855 .setxattr = ll_setxattr,
2856 .getxattr = ll_getxattr,
2857 .listxattr = ll_listxattr,
2858 .removexattr = ll_removexattr,
2859 #ifdef HAVE_LINUX_FIEMAP_H
2860 .fiemap = ll_fiemap,
2862 #ifdef HAVE_IOP_GET_ACL
2863 .get_acl = ll_get_acl,
2867 /* dynamic ioctl number support routins */
2868 static struct llioc_ctl_data {
2869 struct rw_semaphore ioc_sem;
2870 cfs_list_t ioc_head;
2872 __RWSEM_INITIALIZER(llioc.ioc_sem),
2873 CFS_LIST_HEAD_INIT(llioc.ioc_head)
2878 cfs_list_t iocd_list;
2879 unsigned int iocd_size;
2880 llioc_callback_t iocd_cb;
2881 unsigned int iocd_count;
2882 unsigned int iocd_cmd[0];
2885 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
2888 struct llioc_data *in_data = NULL;
2891 if (cb == NULL || cmd == NULL ||
2892 count > LLIOC_MAX_CMD || count < 0)
2895 size = sizeof(*in_data) + count * sizeof(unsigned int);
2896 OBD_ALLOC(in_data, size);
2897 if (in_data == NULL)
2900 memset(in_data, 0, sizeof(*in_data));
2901 in_data->iocd_size = size;
2902 in_data->iocd_cb = cb;
2903 in_data->iocd_count = count;
2904 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
2906 down_write(&llioc.ioc_sem);
2907 cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
2908 up_write(&llioc.ioc_sem);
2913 void ll_iocontrol_unregister(void *magic)
2915 struct llioc_data *tmp;
2920 down_write(&llioc.ioc_sem);
2921 cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
2923 unsigned int size = tmp->iocd_size;
2925 cfs_list_del(&tmp->iocd_list);
2926 up_write(&llioc.ioc_sem);
2928 OBD_FREE(tmp, size);
2932 up_write(&llioc.ioc_sem);
2934 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
2937 EXPORT_SYMBOL(ll_iocontrol_register);
2938 EXPORT_SYMBOL(ll_iocontrol_unregister);
2940 enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
2941 unsigned int cmd, unsigned long arg, int *rcp)
2943 enum llioc_iter ret = LLIOC_CONT;
2944 struct llioc_data *data;
2945 int rc = -EINVAL, i;
2947 down_read(&llioc.ioc_sem);
2948 cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
2949 for (i = 0; i < data->iocd_count; i++) {
2950 if (cmd != data->iocd_cmd[i])
2953 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
2957 if (ret == LLIOC_STOP)
2960 up_read(&llioc.ioc_sem);
2967 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
2969 struct ll_inode_info *lli = ll_i2info(inode);
2970 struct cl_env_nest nest;
2975 if (lli->lli_clob == NULL)
2978 env = cl_env_nested_get(&nest);
2980 RETURN(PTR_ERR(env));
2982 result = cl_conf_set(env, lli->lli_clob, conf);
2983 cl_env_nested_put(&nest, env);
2985 if (conf->coc_opc == OBJECT_CONF_SET) {
2986 struct ldlm_lock *lock = conf->coc_lock;
2988 LASSERT(lock != NULL);
2989 LASSERT(ldlm_has_layout(lock));
2991 /* it can only be allowed to match after layout is
2992 * applied to inode otherwise false layout would be
2993 * seen. Applying layout shoud happen before dropping
2994 * the intent lock. */
2995 ldlm_lock_allow_match(lock);
3002 * Apply the layout to the inode. Layout lock is held and will be released
3005 static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
3006 struct inode *inode, __u32 *gen, bool reconf)
3008 struct ll_inode_info *lli = ll_i2info(inode);
3009 struct ll_sb_info *sbi = ll_i2sbi(inode);
3010 struct ldlm_lock *lock;
3011 struct lustre_md md = { NULL };
3012 struct cl_object_conf conf;
3017 LASSERT(lustre_handle_is_used(lockh));
3019 lock = ldlm_handle2lock(lockh);
3020 LASSERT(lock != NULL);
3021 LASSERT(ldlm_has_layout(lock));
3023 LDLM_DEBUG(lock, "File %p/"DFID" being reconfigured: %d.\n",
3024 inode, PFID(&lli->lli_fid), reconf);
3026 lock_res_and_lock(lock);
3027 lvb_ready = !!(lock->l_flags & LDLM_FL_LVB_READY);
3028 unlock_res_and_lock(lock);
3029 /* checking lvb_ready is racy but this is okay. The worst case is
3030 * that multi processes may configure the file on the same time. */
3031 if (lvb_ready || !reconf) {
3032 LDLM_LOCK_PUT(lock);
3036 /* layout_gen must be valid if layout lock is not
3037 * cancelled and stripe has already set */
3038 *gen = lli->lli_layout_gen;
3041 ldlm_lock_decref(lockh, mode);
3045 /* for layout lock, lmm is returned in lock's lvb.
3046 * lvb_data is immutable if the lock is held so it's safe to access it
3047 * without res lock. See the description in ldlm_lock_decref_internal()
3048 * for the condition to free lvb_data of layout lock */
3049 if (lock->l_lvb_data != NULL) {
3050 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3051 lock->l_lvb_data, lock->l_lvb_len);
3054 *gen = md.lsm->lsm_layout_gen;
3057 CERROR("%s: file "DFID" unpackmd error: %d\n",
3058 ll_get_fsname(inode->i_sb, NULL, 0),
3059 PFID(&lli->lli_fid), rc);
3063 LDLM_LOCK_PUT(lock);
3064 ldlm_lock_decref(lockh, mode);
3068 /* set layout to file. Unlikely this will fail as old layout was
3069 * surely eliminated */
3070 memset(&conf, 0, sizeof conf);
3071 conf.coc_opc = OBJECT_CONF_SET;
3072 conf.coc_inode = inode;
3073 conf.coc_lock = lock;
3074 conf.u.coc_md = &md;
3075 rc = ll_layout_conf(inode, &conf);
3076 LDLM_LOCK_PUT(lock);
3078 ldlm_lock_decref(lockh, mode);
3081 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3083 /* wait for IO to complete if it's still being used. */
3085 CDEBUG(D_INODE, "%s: %p/"DFID" wait for layout reconf.\n",
3086 ll_get_fsname(inode->i_sb, NULL, 0),
3087 inode, PFID(&lli->lli_fid));
3089 memset(&conf, 0, sizeof conf);
3090 conf.coc_opc = OBJECT_CONF_WAIT;
3091 conf.coc_inode = inode;
3092 rc = ll_layout_conf(inode, &conf);
3096 CDEBUG(D_INODE, "file: "DFID" waiting layout return: %d.\n",
3097 PFID(&lli->lli_fid), rc);
3104 * This function checks if there exists a LAYOUT lock on the client side,
3105 * or enqueues it if it doesn't have one in cache.
3107 * This function will not hold layout lock so it may be revoked any time after
3108 * this function returns. Any operations depend on layout should be redone
3111 * This function should be called before lov_io_init() to get an uptodate
3112 * layout version, the caller should save the version number and after IO
3113 * is finished, this function should be called again to verify that layout
3114 * is not changed during IO time.
3116 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3118 struct ll_inode_info *lli = ll_i2info(inode);
3119 struct ll_sb_info *sbi = ll_i2sbi(inode);
3120 struct md_op_data *op_data;
3121 struct lookup_intent it;
3122 struct lustre_handle lockh;
3124 struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS,
3126 .ei_cb_bl = ll_md_blocking_ast,
3127 .ei_cb_cp = ldlm_completion_ast,
3128 .ei_cbdata = inode };
3132 *gen = LL_LAYOUT_GEN_ZERO;
3133 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
3137 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3138 LASSERT(S_ISREG(inode->i_mode));
3140 /* mostly layout lock is caching on the local side, so try to match
3141 * it before grabbing layout lock mutex. */
3142 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0);
3143 if (mode != 0) { /* hit cached lock */
3144 rc = ll_layout_lock_set(&lockh, mode, inode, gen, false);
3148 /* better hold lli_layout_mutex to try again otherwise
3149 * it will have starvation problem. */
3152 /* take layout lock mutex to enqueue layout lock exclusively. */
3153 mutex_lock(&lli->lli_layout_mutex);
3156 /* try again. Maybe somebody else has done this. */
3157 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0);
3158 if (mode != 0) { /* hit cached lock */
3159 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3163 mutex_unlock(&lli->lli_layout_mutex);
3167 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3168 0, 0, LUSTRE_OPC_ANY, NULL);
3169 if (IS_ERR(op_data)) {
3170 mutex_unlock(&lli->lli_layout_mutex);
3171 RETURN(PTR_ERR(op_data));
3174 /* have to enqueue one */
3175 memset(&it, 0, sizeof(it));
3176 it.it_op = IT_LAYOUT;
3177 lockh.cookie = 0ULL;
3179 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file %p/"DFID".\n",
3180 ll_get_fsname(inode->i_sb, NULL, 0), inode,
3181 PFID(&lli->lli_fid));
3183 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
3185 if (it.d.lustre.it_data != NULL)
3186 ptlrpc_req_finished(it.d.lustre.it_data);
3187 it.d.lustre.it_data = NULL;
3189 ll_finish_md_op_data(op_data);
3191 mode = it.d.lustre.it_lock_mode;
3192 it.d.lustre.it_lock_mode = 0;
3193 ll_intent_drop_lock(&it);
3196 /* set lock data in case this is a new lock */
3197 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3198 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3202 mutex_unlock(&lli->lli_layout_mutex);