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(const struct lu_env *env, struct inode *inode)
772 struct ll_inode_info *lli = ll_i2info(inode);
773 struct cl_object *obj = lli->lli_clob;
774 struct cl_attr *attr = ccc_env_thread_attr(env);
780 ll_inode_size_lock(inode);
781 /* merge timestamps the most recently obtained from mds with
782 timestamps obtained from osts */
783 LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime;
784 LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime;
785 LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime;
786 inode_init_lvb(inode, &lvb);
788 cl_object_attr_lock(obj);
789 rc = cl_object_attr_get(env, obj, attr);
790 cl_object_attr_unlock(obj);
793 if (lvb.lvb_atime < attr->cat_atime)
794 lvb.lvb_atime = attr->cat_atime;
795 if (lvb.lvb_ctime < attr->cat_ctime)
796 lvb.lvb_ctime = attr->cat_ctime;
797 if (lvb.lvb_mtime < attr->cat_mtime)
798 lvb.lvb_mtime = attr->cat_mtime;
800 CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n",
801 PFID(&lli->lli_fid), attr->cat_size);
802 cl_isize_write_nolock(inode, attr->cat_size);
804 inode->i_blocks = attr->cat_blocks;
806 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
807 LTIME_S(inode->i_atime) = lvb.lvb_atime;
808 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
810 ll_inode_size_unlock(inode);
815 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
818 struct obdo obdo = { 0 };
821 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0);
823 st->st_size = obdo.o_size;
824 st->st_blocks = obdo.o_blocks;
825 st->st_mtime = obdo.o_mtime;
826 st->st_atime = obdo.o_atime;
827 st->st_ctime = obdo.o_ctime;
832 void ll_io_init(struct cl_io *io, const struct file *file, int write)
834 struct inode *inode = file->f_dentry->d_inode;
836 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
838 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
839 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
840 file->f_flags & O_DIRECT ||
843 io->ci_obj = ll_i2info(inode)->lli_clob;
844 io->ci_lockreq = CILR_MAYBE;
845 if (ll_file_nolock(file)) {
846 io->ci_lockreq = CILR_NEVER;
847 io->ci_no_srvlock = 1;
848 } else if (file->f_flags & O_APPEND) {
849 io->ci_lockreq = CILR_MANDATORY;
854 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
855 struct file *file, enum cl_io_type iot,
856 loff_t *ppos, size_t count)
858 struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
859 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
865 io = ccc_env_thread_io(env);
866 ll_io_init(io, file, iot == CIT_WRITE);
868 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
869 struct vvp_io *vio = vvp_env_io(env);
870 struct ccc_io *cio = ccc_env_io(env);
871 int write_mutex_locked = 0;
873 cio->cui_fd = LUSTRE_FPRIVATE(file);
874 vio->cui_io_subtype = args->via_io_subtype;
876 switch (vio->cui_io_subtype) {
878 cio->cui_iov = args->u.normal.via_iov;
879 cio->cui_nrsegs = args->u.normal.via_nrsegs;
880 cio->cui_tot_nrsegs = cio->cui_nrsegs;
881 #ifndef HAVE_FILE_WRITEV
882 cio->cui_iocb = args->u.normal.via_iocb;
884 if ((iot == CIT_WRITE) &&
885 !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
886 if (mutex_lock_interruptible(&lli->
888 GOTO(out, result = -ERESTARTSYS);
889 write_mutex_locked = 1;
890 } else if (iot == CIT_READ) {
891 down_read(&lli->lli_trunc_sem);
895 vio->u.sendfile.cui_actor = args->u.sendfile.via_actor;
896 vio->u.sendfile.cui_target = args->u.sendfile.via_target;
899 vio->u.splice.cui_pipe = args->u.splice.via_pipe;
900 vio->u.splice.cui_flags = args->u.splice.via_flags;
903 CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
906 result = cl_io_loop(env, io);
907 if (write_mutex_locked)
908 mutex_unlock(&lli->lli_write_mutex);
909 else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ)
910 up_read(&lli->lli_trunc_sem);
912 /* cl_io_rw_init() handled IO */
913 result = io->ci_result;
916 if (io->ci_nob > 0) {
918 *ppos = io->u.ci_wr.wr.crw_pos;
923 /* If any bit been read/written (result != 0), we just return
924 * short read/write instead of restart io. */
925 if (result == 0 && io->ci_need_restart) {
926 CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n",
927 iot == CIT_READ ? "read" : "write",
928 file->f_dentry->d_name.name, *ppos, count);
929 LASSERTF(io->u.ci_rw.crw_count == count, "%zd != %zd\n",
930 io->u.ci_rw.crw_count, count);
934 if (iot == CIT_READ) {
936 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
937 LPROC_LL_READ_BYTES, result);
938 } else if (iot == CIT_WRITE) {
940 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
941 LPROC_LL_WRITE_BYTES, result);
942 fd->fd_write_failed = false;
943 } else if (result != -ERESTARTSYS) {
944 fd->fd_write_failed = true;
953 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
955 static int ll_file_get_iov_count(const struct iovec *iov,
956 unsigned long *nr_segs, size_t *count)
961 for (seg = 0; seg < *nr_segs; seg++) {
962 const struct iovec *iv = &iov[seg];
965 * If any segment has a negative length, or the cumulative
966 * length ever wraps negative then return -EINVAL.
969 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
971 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
976 cnt -= iv->iov_len; /* This segment is no good */
983 #ifdef HAVE_FILE_READV
984 static ssize_t ll_file_readv(struct file *file, const struct iovec *iov,
985 unsigned long nr_segs, loff_t *ppos)
988 struct vvp_io_args *args;
994 result = ll_file_get_iov_count(iov, &nr_segs, &count);
998 env = cl_env_get(&refcheck);
1000 RETURN(PTR_ERR(env));
1002 args = vvp_env_args(env, IO_NORMAL);
1003 args->u.normal.via_iov = (struct iovec *)iov;
1004 args->u.normal.via_nrsegs = nr_segs;
1006 result = ll_file_io_generic(env, args, file, CIT_READ, ppos, count);
1007 cl_env_put(env, &refcheck);
1011 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1015 struct iovec *local_iov;
1020 env = cl_env_get(&refcheck);
1022 RETURN(PTR_ERR(env));
1024 local_iov = &vvp_env_info(env)->vti_local_iov;
1025 local_iov->iov_base = (void __user *)buf;
1026 local_iov->iov_len = count;
1027 result = ll_file_readv(file, local_iov, 1, ppos);
1028 cl_env_put(env, &refcheck);
1033 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1034 unsigned long nr_segs, loff_t pos)
1037 struct vvp_io_args *args;
1043 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1047 env = cl_env_get(&refcheck);
1049 RETURN(PTR_ERR(env));
1051 args = vvp_env_args(env, IO_NORMAL);
1052 args->u.normal.via_iov = (struct iovec *)iov;
1053 args->u.normal.via_nrsegs = nr_segs;
1054 args->u.normal.via_iocb = iocb;
1056 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1057 &iocb->ki_pos, count);
1058 cl_env_put(env, &refcheck);
1062 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1066 struct iovec *local_iov;
1067 struct kiocb *kiocb;
1072 env = cl_env_get(&refcheck);
1074 RETURN(PTR_ERR(env));
1076 local_iov = &vvp_env_info(env)->vti_local_iov;
1077 kiocb = &vvp_env_info(env)->vti_kiocb;
1078 local_iov->iov_base = (void __user *)buf;
1079 local_iov->iov_len = count;
1080 init_sync_kiocb(kiocb, file);
1081 kiocb->ki_pos = *ppos;
1082 kiocb->ki_left = count;
1084 result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos);
1085 *ppos = kiocb->ki_pos;
1087 cl_env_put(env, &refcheck);
1093 * Write to a file (through the page cache).
1095 #ifdef HAVE_FILE_WRITEV
1096 static ssize_t ll_file_writev(struct file *file, const struct iovec *iov,
1097 unsigned long nr_segs, loff_t *ppos)
1100 struct vvp_io_args *args;
1106 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1110 env = cl_env_get(&refcheck);
1112 RETURN(PTR_ERR(env));
1114 args = vvp_env_args(env, IO_NORMAL);
1115 args->u.normal.via_iov = (struct iovec *)iov;
1116 args->u.normal.via_nrsegs = nr_segs;
1118 result = ll_file_io_generic(env, args, file, CIT_WRITE, ppos, count);
1119 cl_env_put(env, &refcheck);
1123 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1127 struct iovec *local_iov;
1132 env = cl_env_get(&refcheck);
1134 RETURN(PTR_ERR(env));
1136 local_iov = &vvp_env_info(env)->vti_local_iov;
1137 local_iov->iov_base = (void __user *)buf;
1138 local_iov->iov_len = count;
1140 result = ll_file_writev(file, local_iov, 1, ppos);
1141 cl_env_put(env, &refcheck);
1145 #else /* AIO stuff */
1146 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1147 unsigned long nr_segs, loff_t pos)
1150 struct vvp_io_args *args;
1156 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1160 env = cl_env_get(&refcheck);
1162 RETURN(PTR_ERR(env));
1164 args = vvp_env_args(env, IO_NORMAL);
1165 args->u.normal.via_iov = (struct iovec *)iov;
1166 args->u.normal.via_nrsegs = nr_segs;
1167 args->u.normal.via_iocb = iocb;
1169 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1170 &iocb->ki_pos, count);
1171 cl_env_put(env, &refcheck);
1175 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1179 struct iovec *local_iov;
1180 struct kiocb *kiocb;
1185 env = cl_env_get(&refcheck);
1187 RETURN(PTR_ERR(env));
1189 local_iov = &vvp_env_info(env)->vti_local_iov;
1190 kiocb = &vvp_env_info(env)->vti_kiocb;
1191 local_iov->iov_base = (void __user *)buf;
1192 local_iov->iov_len = count;
1193 init_sync_kiocb(kiocb, file);
1194 kiocb->ki_pos = *ppos;
1195 kiocb->ki_left = count;
1197 result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos);
1198 *ppos = kiocb->ki_pos;
1200 cl_env_put(env, &refcheck);
1206 #ifdef HAVE_KERNEL_SENDFILE
1208 * Send file content (through pagecache) somewhere with helper
1210 static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count,
1211 read_actor_t actor, void *target)
1214 struct vvp_io_args *args;
1219 env = cl_env_get(&refcheck);
1221 RETURN(PTR_ERR(env));
1223 args = vvp_env_args(env, IO_SENDFILE);
1224 args->u.sendfile.via_target = target;
1225 args->u.sendfile.via_actor = actor;
1227 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1228 cl_env_put(env, &refcheck);
1233 #ifdef HAVE_KERNEL_SPLICE_READ
1235 * Send file content (through pagecache) somewhere with helper
1237 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1238 struct pipe_inode_info *pipe, size_t count,
1242 struct vvp_io_args *args;
1247 env = cl_env_get(&refcheck);
1249 RETURN(PTR_ERR(env));
1251 args = vvp_env_args(env, IO_SPLICE);
1252 args->u.splice.via_pipe = pipe;
1253 args->u.splice.via_flags = flags;
1255 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1256 cl_env_put(env, &refcheck);
1261 static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq,
1264 struct obd_export *exp = ll_i2dtexp(inode);
1265 struct obd_trans_info oti = { 0 };
1266 struct obdo *oa = NULL;
1269 struct lov_stripe_md *lsm = NULL, *lsm2;
1276 lsm = ccc_inode_lsm_get(inode);
1278 GOTO(out, rc = -ENOENT);
1280 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1281 (lsm->lsm_stripe_count));
1283 OBD_ALLOC_LARGE(lsm2, lsm_size);
1285 GOTO(out, rc = -ENOMEM);
1289 oa->o_nlink = ost_idx;
1290 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1291 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1292 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1293 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1294 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1295 memcpy(lsm2, lsm, lsm_size);
1296 ll_inode_size_lock(inode);
1297 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1298 ll_inode_size_unlock(inode);
1300 OBD_FREE_LARGE(lsm2, lsm_size);
1303 ccc_inode_lsm_put(inode, lsm);
1308 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1310 struct ll_recreate_obj ucreat;
1313 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1316 if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
1320 RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0,
1321 ucreat.lrc_ost_idx));
1324 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1331 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1334 if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid)))
1337 id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32);
1338 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1339 RETURN(ll_lov_recreate(inode, id, 0, ost_idx));
1342 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1343 int flags, struct lov_user_md *lum, int lum_size)
1345 struct lov_stripe_md *lsm = NULL;
1346 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1350 lsm = ccc_inode_lsm_get(inode);
1352 ccc_inode_lsm_put(inode, lsm);
1353 CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n",
1358 ll_inode_size_lock(inode);
1359 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1362 rc = oit.d.lustre.it_status;
1364 GOTO(out_req_free, rc);
1366 ll_release_openhandle(file->f_dentry, &oit);
1369 ll_inode_size_unlock(inode);
1370 ll_intent_release(&oit);
1371 ccc_inode_lsm_put(inode, lsm);
1374 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1378 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1379 struct lov_mds_md **lmmp, int *lmm_size,
1380 struct ptlrpc_request **request)
1382 struct ll_sb_info *sbi = ll_i2sbi(inode);
1383 struct mdt_body *body;
1384 struct lov_mds_md *lmm = NULL;
1385 struct ptlrpc_request *req = NULL;
1386 struct md_op_data *op_data;
1389 rc = ll_get_max_mdsize(sbi, &lmmsize);
1393 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1394 strlen(filename), lmmsize,
1395 LUSTRE_OPC_ANY, NULL);
1396 if (IS_ERR(op_data))
1397 RETURN(PTR_ERR(op_data));
1399 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1400 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1401 ll_finish_md_op_data(op_data);
1403 CDEBUG(D_INFO, "md_getattr_name failed "
1404 "on %s: rc %d\n", filename, rc);
1408 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1409 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1411 lmmsize = body->eadatasize;
1413 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1415 GOTO(out, rc = -ENODATA);
1418 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1419 LASSERT(lmm != NULL);
1421 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1422 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1423 GOTO(out, rc = -EPROTO);
1427 * This is coming from the MDS, so is probably in
1428 * little endian. We convert it to host endian before
1429 * passing it to userspace.
1431 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1432 /* if function called for directory - we should
1433 * avoid swab not existent lsm objects */
1434 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1435 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1436 if (S_ISREG(body->mode))
1437 lustre_swab_lov_user_md_objects(
1438 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1439 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1440 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1441 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1442 if (S_ISREG(body->mode))
1443 lustre_swab_lov_user_md_objects(
1444 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1445 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1451 *lmm_size = lmmsize;
1456 static int ll_lov_setea(struct inode *inode, struct file *file,
1459 int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1460 struct lov_user_md *lump;
1461 int lum_size = sizeof(struct lov_user_md) +
1462 sizeof(struct lov_user_ost_data);
1466 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1469 OBD_ALLOC_LARGE(lump, lum_size);
1473 if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) {
1474 OBD_FREE_LARGE(lump, lum_size);
1478 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1480 OBD_FREE_LARGE(lump, lum_size);
1484 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1487 struct lov_user_md_v3 lumv3;
1488 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1489 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1490 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1492 int flags = FMODE_WRITE;
1495 /* first try with v1 which is smaller than v3 */
1496 lum_size = sizeof(struct lov_user_md_v1);
1497 if (copy_from_user(lumv1, lumv1p, lum_size))
1500 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1501 lum_size = sizeof(struct lov_user_md_v3);
1502 if (copy_from_user(&lumv3, lumv3p, lum_size))
1506 rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size);
1508 struct lov_stripe_md *lsm;
1511 put_user(0, &lumv1p->lmm_stripe_count);
1513 ll_layout_refresh(inode, &gen);
1514 lsm = ccc_inode_lsm_get(inode);
1515 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1516 0, lsm, (void *)arg);
1517 ccc_inode_lsm_put(inode, lsm);
1522 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1524 struct lov_stripe_md *lsm;
1528 lsm = ccc_inode_lsm_get(inode);
1530 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1532 ccc_inode_lsm_put(inode, lsm);
1536 int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1538 struct ll_inode_info *lli = ll_i2info(inode);
1539 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1540 struct ccc_grouplock grouplock;
1544 if (ll_file_nolock(file))
1545 RETURN(-EOPNOTSUPP);
1547 spin_lock(&lli->lli_lock);
1548 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1549 CWARN("group lock already existed with gid %lu\n",
1550 fd->fd_grouplock.cg_gid);
1551 spin_unlock(&lli->lli_lock);
1554 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1555 spin_unlock(&lli->lli_lock);
1557 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1558 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1562 spin_lock(&lli->lli_lock);
1563 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1564 spin_unlock(&lli->lli_lock);
1565 CERROR("another thread just won the race\n");
1566 cl_put_grouplock(&grouplock);
1570 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1571 fd->fd_grouplock = grouplock;
1572 spin_unlock(&lli->lli_lock);
1574 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1578 int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1580 struct ll_inode_info *lli = ll_i2info(inode);
1581 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1582 struct ccc_grouplock grouplock;
1585 spin_lock(&lli->lli_lock);
1586 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1587 spin_unlock(&lli->lli_lock);
1588 CWARN("no group lock held\n");
1591 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1593 if (fd->fd_grouplock.cg_gid != arg) {
1594 CWARN("group lock %lu doesn't match current id %lu\n",
1595 arg, fd->fd_grouplock.cg_gid);
1596 spin_unlock(&lli->lli_lock);
1600 grouplock = fd->fd_grouplock;
1601 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1602 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1603 spin_unlock(&lli->lli_lock);
1605 cl_put_grouplock(&grouplock);
1606 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1611 * Close inode open handle
1613 * \param dentry [in] dentry which contains the inode
1614 * \param it [in,out] intent which contains open info and result
1617 * \retval <0 failure
1619 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1621 struct inode *inode = dentry->d_inode;
1622 struct obd_client_handle *och;
1628 /* Root ? Do nothing. */
1629 if (dentry->d_inode->i_sb->s_root == dentry)
1632 /* No open handle to close? Move away */
1633 if (!it_disposition(it, DISP_OPEN_OPEN))
1636 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1638 OBD_ALLOC(och, sizeof(*och));
1640 GOTO(out, rc = -ENOMEM);
1642 ll_och_fill(ll_i2sbi(inode)->ll_md_exp,
1643 ll_i2info(inode), it, och);
1645 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1648 /* this one is in place of ll_file_open */
1649 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1650 ptlrpc_req_finished(it->d.lustre.it_data);
1651 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1657 * Get size for inode for which FIEMAP mapping is requested.
1658 * Make the FIEMAP get_info call and returns the result.
1660 int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1663 struct obd_export *exp = ll_i2dtexp(inode);
1664 struct lov_stripe_md *lsm = NULL;
1665 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1666 int vallen = num_bytes;
1670 /* Checks for fiemap flags */
1671 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1672 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1676 /* Check for FIEMAP_FLAG_SYNC */
1677 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1678 rc = filemap_fdatawrite(inode->i_mapping);
1683 lsm = ccc_inode_lsm_get(inode);
1687 /* If the stripe_count > 1 and the application does not understand
1688 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1690 if (lsm->lsm_stripe_count > 1 &&
1691 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER))
1692 GOTO(out, rc = -EOPNOTSUPP);
1694 fm_key.oa.o_id = lsm->lsm_object_id;
1695 fm_key.oa.o_seq = lsm->lsm_object_seq;
1696 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1698 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1699 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1700 /* If filesize is 0, then there would be no objects for mapping */
1701 if (fm_key.oa.o_size == 0) {
1702 fiemap->fm_mapped_extents = 0;
1706 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1708 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1711 CERROR("obd_get_info failed: rc = %d\n", rc);
1714 ccc_inode_lsm_put(inode, lsm);
1718 int ll_fid2path(struct inode *inode, void *arg)
1720 struct obd_export *exp = ll_i2mdexp(inode);
1721 struct getinfo_fid2path *gfout, *gfin;
1725 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1726 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1729 /* Need to get the buflen */
1730 OBD_ALLOC_PTR(gfin);
1733 if (copy_from_user(gfin, arg, sizeof(*gfin))) {
1738 outsize = sizeof(*gfout) + gfin->gf_pathlen;
1739 OBD_ALLOC(gfout, outsize);
1740 if (gfout == NULL) {
1744 memcpy(gfout, gfin, sizeof(*gfout));
1747 /* Call mdc_iocontrol */
1748 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1752 if (copy_to_user(arg, gfout, outsize))
1756 OBD_FREE(gfout, outsize);
1760 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1762 struct ll_user_fiemap *fiemap_s;
1763 size_t num_bytes, ret_bytes;
1764 unsigned int extent_count;
1767 /* Get the extent count so we can calculate the size of
1768 * required fiemap buffer */
1769 if (get_user(extent_count,
1770 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1772 num_bytes = sizeof(*fiemap_s) + (extent_count *
1773 sizeof(struct ll_fiemap_extent));
1775 OBD_ALLOC_LARGE(fiemap_s, num_bytes);
1776 if (fiemap_s == NULL)
1779 /* get the fiemap value */
1780 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1782 GOTO(error, rc = -EFAULT);
1784 /* If fm_extent_count is non-zero, read the first extent since
1785 * it is used to calculate end_offset and device from previous
1788 if (copy_from_user(&fiemap_s->fm_extents[0],
1789 (char __user *)arg + sizeof(*fiemap_s),
1790 sizeof(struct ll_fiemap_extent)))
1791 GOTO(error, rc = -EFAULT);
1794 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1798 ret_bytes = sizeof(struct ll_user_fiemap);
1800 if (extent_count != 0)
1801 ret_bytes += (fiemap_s->fm_mapped_extents *
1802 sizeof(struct ll_fiemap_extent));
1804 if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
1808 OBD_FREE_LARGE(fiemap_s, num_bytes);
1813 * Read the data_version for inode.
1815 * This value is computed using stripe object version on OST.
1816 * Version is computed using server side locking.
1818 * @param extent_lock Take extent lock. Not needed if a process is already
1819 * holding the OST object group locks.
1821 int ll_data_version(struct inode *inode, __u64 *data_version,
1824 struct lov_stripe_md *lsm = NULL;
1825 struct ll_sb_info *sbi = ll_i2sbi(inode);
1826 struct obdo *obdo = NULL;
1830 /* If no stripe, we consider version is 0. */
1831 lsm = ccc_inode_lsm_get(inode);
1834 CDEBUG(D_INODE, "No object for inode\n");
1838 OBD_ALLOC_PTR(obdo);
1840 ccc_inode_lsm_put(inode, lsm);
1844 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock);
1846 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
1849 *data_version = obdo->o_data_version;
1853 ccc_inode_lsm_put(inode, lsm);
1858 static int ll_swap_layout(struct file *file, struct file *file2,
1859 struct lustre_swap_layouts *lsl)
1861 struct mdc_swap_layouts msl = { .msl_flags = lsl->sl_flags };
1862 struct md_op_data *op_data;
1863 struct inode *inode = file->f_dentry->d_inode;
1864 struct inode *inode2 = file2->f_dentry->d_inode;
1868 if (!S_ISREG(inode2->i_mode))
1871 if (ll_permission(inode, MAY_WRITE, NULL) ||
1872 ll_permission(inode2, MAY_WRITE, NULL))
1875 if (inode2->i_sb != inode->i_sb)
1878 rc = lu_fid_cmp(ll_inode2fid(inode), ll_inode2fid(inode2));
1879 if (rc == 0) /* same file, done! */
1882 if (rc < 0) { /* sequentialize it */
1883 swap(inode, inode2);
1888 if (gid != 0) { /* application asks to flush dirty cache */
1889 rc = ll_get_grouplock(inode, file, gid);
1893 rc = ll_get_grouplock(inode2, file2, gid);
1895 ll_put_grouplock(inode, file, gid);
1900 /* struct md_op_data is used to send the swap args to the mdt
1901 * only flags is missing, so we use struct mdc_swap_layouts
1902 * through the md_op_data->op_data */
1904 op_data = ll_prep_md_op_data(NULL, inode, inode2, NULL, 0, 0,
1905 LUSTRE_OPC_ANY, &msl);
1906 if (op_data != NULL) {
1907 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(inode),
1908 sizeof(*op_data), op_data, NULL);
1909 ll_finish_md_op_data(op_data);
1913 ll_put_grouplock(inode2, file2, gid);
1914 ll_put_grouplock(inode, file, gid);
1920 long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1922 struct inode *inode = file->f_dentry->d_inode;
1923 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1927 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino,
1928 inode->i_generation, inode, cmd);
1929 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1931 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1932 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1936 case LL_IOC_GETFLAGS:
1937 /* Get the current value of the file flags */
1938 return put_user(fd->fd_flags, (int *)arg);
1939 case LL_IOC_SETFLAGS:
1940 case LL_IOC_CLRFLAGS:
1941 /* Set or clear specific file flags */
1942 /* XXX This probably needs checks to ensure the flags are
1943 * not abused, and to handle any flag side effects.
1945 if (get_user(flags, (int *) arg))
1948 if (cmd == LL_IOC_SETFLAGS) {
1949 if ((flags & LL_FILE_IGNORE_LOCK) &&
1950 !(file->f_flags & O_DIRECT)) {
1951 CERROR("%s: unable to disable locking on "
1952 "non-O_DIRECT file\n", current->comm);
1956 fd->fd_flags |= flags;
1958 fd->fd_flags &= ~flags;
1961 case LL_IOC_LOV_SETSTRIPE:
1962 RETURN(ll_lov_setstripe(inode, file, arg));
1963 case LL_IOC_LOV_SETEA:
1964 RETURN(ll_lov_setea(inode, file, arg));
1965 case LL_IOC_LOV_SWAP_LAYOUTS: {
1967 struct lustre_swap_layouts lsl;
1969 if (cfs_copy_from_user(&lsl, (char *)arg,
1970 sizeof(struct lustre_swap_layouts)))
1973 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
1976 file2 = fget(lsl.sl_fd);
1981 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
1982 rc = ll_swap_layout(file, file2, &lsl);
1986 case LL_IOC_LOV_GETSTRIPE:
1987 RETURN(ll_lov_getstripe(inode, arg));
1988 case LL_IOC_RECREATE_OBJ:
1989 RETURN(ll_lov_recreate_obj(inode, arg));
1990 case LL_IOC_RECREATE_FID:
1991 RETURN(ll_lov_recreate_fid(inode, arg));
1992 case FSFILT_IOC_FIEMAP:
1993 RETURN(ll_ioctl_fiemap(inode, arg));
1994 case FSFILT_IOC_GETFLAGS:
1995 case FSFILT_IOC_SETFLAGS:
1996 RETURN(ll_iocontrol(inode, file, cmd, arg));
1997 case FSFILT_IOC_GETVERSION_OLD:
1998 case FSFILT_IOC_GETVERSION:
1999 RETURN(put_user(inode->i_generation, (int *)arg));
2000 case LL_IOC_GROUP_LOCK:
2001 RETURN(ll_get_grouplock(inode, file, arg));
2002 case LL_IOC_GROUP_UNLOCK:
2003 RETURN(ll_put_grouplock(inode, file, arg));
2004 case IOC_OBD_STATFS:
2005 RETURN(ll_obd_statfs(inode, (void *)arg));
2007 /* We need to special case any other ioctls we want to handle,
2008 * to send them to the MDS/OST as appropriate and to properly
2009 * network encode the arg field.
2010 case FSFILT_IOC_SETVERSION_OLD:
2011 case FSFILT_IOC_SETVERSION:
2013 case LL_IOC_FLUSHCTX:
2014 RETURN(ll_flush_ctx(inode));
2015 case LL_IOC_PATH2FID: {
2016 if (copy_to_user((void *)arg, ll_inode2fid(inode),
2017 sizeof(struct lu_fid)))
2022 case OBD_IOC_FID2PATH:
2023 RETURN(ll_fid2path(inode, (void *)arg));
2024 case LL_IOC_DATA_VERSION: {
2025 struct ioc_data_version idv;
2028 if (copy_from_user(&idv, (char *)arg, sizeof(idv)))
2031 rc = ll_data_version(inode, &idv.idv_version,
2032 !(idv.idv_flags & LL_DV_NOFLUSH));
2034 if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv)))
2040 case LL_IOC_GET_MDTIDX: {
2043 mdtidx = ll_get_mdt_idx(inode);
2047 if (put_user((int)mdtidx, (int*)arg))
2052 case OBD_IOC_GETDTNAME:
2053 case OBD_IOC_GETMDNAME:
2054 RETURN(ll_get_obd_name(inode, cmd, arg));
2055 case LL_IOC_HSM_STATE_GET: {
2056 struct md_op_data *op_data;
2057 struct hsm_user_state *hus;
2064 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2065 LUSTRE_OPC_ANY, hus);
2066 if (op_data == NULL) {
2071 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2074 if (copy_to_user((void *)arg, hus, sizeof(*hus)))
2077 ll_finish_md_op_data(op_data);
2081 case LL_IOC_HSM_STATE_SET: {
2082 struct md_op_data *op_data;
2083 struct hsm_state_set *hss;
2089 if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
2094 /* Non-root users are forbidden to set or clear flags which are
2095 * NOT defined in HSM_USER_MASK. */
2096 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK)
2097 && !cfs_capable(CFS_CAP_SYS_ADMIN)) {
2102 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2103 LUSTRE_OPC_ANY, hss);
2104 if (op_data == NULL) {
2109 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2112 ll_finish_md_op_data(op_data);
2117 case LL_IOC_HSM_ACTION: {
2118 struct md_op_data *op_data;
2119 struct hsm_current_action *hca;
2126 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2127 LUSTRE_OPC_ANY, hca);
2128 if (op_data == NULL) {
2133 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2136 if (cfs_copy_to_user((char *)arg, hca, sizeof(*hca)))
2139 ll_finish_md_op_data(op_data);
2147 ll_iocontrol_call(inode, file, cmd, arg, &err))
2150 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2156 #ifndef HAVE_FILE_LLSEEK_SIZE
2157 static inline loff_t
2158 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2160 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2162 if (offset > maxsize)
2165 if (offset != file->f_pos) {
2166 file->f_pos = offset;
2167 file->f_version = 0;
2173 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2174 loff_t maxsize, loff_t eof)
2176 struct inode *inode = file->f_dentry->d_inode;
2184 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2185 * position-querying operation. Avoid rewriting the "same"
2186 * f_pos value back to the file because a concurrent read(),
2187 * write() or lseek() might have altered it
2192 * f_lock protects against read/modify/write race with other
2193 * SEEK_CURs. Note that parallel writes and reads behave
2196 mutex_lock(&inode->i_mutex);
2197 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2198 mutex_unlock(&inode->i_mutex);
2202 * In the generic case the entire file is data, so as long as
2203 * offset isn't at the end of the file then the offset is data.
2210 * There is a virtual hole at the end of the file, so as long as
2211 * offset isn't i_size or larger, return i_size.
2219 return llseek_execute(file, offset, maxsize);
2223 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2225 struct inode *inode = file->f_dentry->d_inode;
2226 loff_t retval, eof = 0;
2229 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2230 (origin == SEEK_CUR) ? file->f_pos : 0);
2231 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%d)\n",
2232 inode->i_ino, inode->i_generation, inode, retval, retval,
2234 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2236 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2237 retval = ll_glimpse_size(inode);
2240 eof = i_size_read(inode);
2243 retval = ll_generic_file_llseek_size(file, offset, origin,
2244 ll_file_maxbytes(inode), eof);
2248 int ll_flush(struct file *file, fl_owner_t id)
2250 struct inode *inode = file->f_dentry->d_inode;
2251 struct ll_inode_info *lli = ll_i2info(inode);
2252 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2255 LASSERT(!S_ISDIR(inode->i_mode));
2257 /* catch async errors that were recorded back when async writeback
2258 * failed for pages in this mapping. */
2259 rc = lli->lli_async_rc;
2260 lli->lli_async_rc = 0;
2261 err = lov_read_and_clear_async_rc(lli->lli_clob);
2265 /* The application has been told write failure already.
2266 * Do not report failure again. */
2267 if (fd->fd_write_failed)
2269 return rc ? -EIO : 0;
2273 * Called to make sure a portion of file has been written out.
2274 * if @local_only is not true, it will send OST_SYNC RPCs to ost.
2276 * Return how many pages have been written.
2278 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2279 enum cl_fsync_mode mode)
2281 struct cl_env_nest nest;
2284 struct obd_capa *capa = NULL;
2285 struct cl_fsync_io *fio;
2289 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2290 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2293 env = cl_env_nested_get(&nest);
2295 RETURN(PTR_ERR(env));
2297 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
2299 io = ccc_env_thread_io(env);
2300 io->ci_obj = cl_i2info(inode)->lli_clob;
2301 io->ci_ignore_layout = 1;
2303 /* initialize parameters for sync */
2304 fio = &io->u.ci_fsync;
2305 fio->fi_capa = capa;
2306 fio->fi_start = start;
2308 fio->fi_fid = ll_inode2fid(inode);
2309 fio->fi_mode = mode;
2310 fio->fi_nr_written = 0;
2312 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2313 result = cl_io_loop(env, io);
2315 result = io->ci_result;
2317 result = fio->fi_nr_written;
2318 cl_io_fini(env, io);
2319 cl_env_nested_put(&nest, env);
2326 #ifdef HAVE_FILE_FSYNC_4ARGS
2327 int ll_fsync(struct file *file, loff_t start, loff_t end, int data)
2328 #elif defined(HAVE_FILE_FSYNC_2ARGS)
2329 int ll_fsync(struct file *file, int data)
2331 int ll_fsync(struct file *file, struct dentry *dentry, int data)
2334 struct inode *inode = file->f_dentry->d_inode;
2335 struct ll_inode_info *lli = ll_i2info(inode);
2336 struct ptlrpc_request *req;
2337 struct obd_capa *oc;
2341 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2342 inode->i_generation, inode);
2343 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2345 #ifdef HAVE_FILE_FSYNC_4ARGS
2346 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2347 mutex_lock(&inode->i_mutex);
2349 /* fsync's caller has already called _fdata{sync,write}, we want
2350 * that IO to finish before calling the osc and mdc sync methods */
2351 rc = filemap_fdatawait(inode->i_mapping);
2354 /* catch async errors that were recorded back when async writeback
2355 * failed for pages in this mapping. */
2356 if (!S_ISDIR(inode->i_mode)) {
2357 err = lli->lli_async_rc;
2358 lli->lli_async_rc = 0;
2361 err = lov_read_and_clear_async_rc(lli->lli_clob);
2366 oc = ll_mdscapa_get(inode);
2367 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2373 ptlrpc_req_finished(req);
2376 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2378 err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2380 if (rc == 0 && err < 0)
2383 fd->fd_write_failed = true;
2385 fd->fd_write_failed = false;
2388 #ifdef HAVE_FILE_FSYNC_4ARGS
2389 mutex_unlock(&inode->i_mutex);
2394 int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2396 struct inode *inode = file->f_dentry->d_inode;
2397 struct ll_sb_info *sbi = ll_i2sbi(inode);
2398 struct ldlm_enqueue_info einfo = { .ei_type = LDLM_FLOCK,
2399 .ei_cb_cp =ldlm_flock_completion_ast,
2400 .ei_cbdata = file_lock };
2401 struct md_op_data *op_data;
2402 struct lustre_handle lockh = {0};
2403 ldlm_policy_data_t flock = {{0}};
2409 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n",
2410 inode->i_ino, file_lock);
2412 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2414 if (file_lock->fl_flags & FL_FLOCK) {
2415 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2416 /* flocks are whole-file locks */
2417 flock.l_flock.end = OFFSET_MAX;
2418 /* For flocks owner is determined by the local file desctiptor*/
2419 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
2420 } else if (file_lock->fl_flags & FL_POSIX) {
2421 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2422 flock.l_flock.start = file_lock->fl_start;
2423 flock.l_flock.end = file_lock->fl_end;
2427 flock.l_flock.pid = file_lock->fl_pid;
2429 /* Somewhat ugly workaround for svc lockd.
2430 * lockd installs custom fl_lmops->lm_compare_owner that checks
2431 * for the fl_owner to be the same (which it always is on local node
2432 * I guess between lockd processes) and then compares pid.
2433 * As such we assign pid to the owner field to make it all work,
2434 * conflict with normal locks is unlikely since pid space and
2435 * pointer space for current->files are not intersecting */
2436 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2437 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2439 switch (file_lock->fl_type) {
2441 einfo.ei_mode = LCK_PR;
2444 /* An unlock request may or may not have any relation to
2445 * existing locks so we may not be able to pass a lock handle
2446 * via a normal ldlm_lock_cancel() request. The request may even
2447 * unlock a byte range in the middle of an existing lock. In
2448 * order to process an unlock request we need all of the same
2449 * information that is given with a normal read or write record
2450 * lock request. To avoid creating another ldlm unlock (cancel)
2451 * message we'll treat a LCK_NL flock request as an unlock. */
2452 einfo.ei_mode = LCK_NL;
2455 einfo.ei_mode = LCK_PW;
2458 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
2459 file_lock->fl_type);
2474 flags = LDLM_FL_BLOCK_NOWAIT;
2480 flags = LDLM_FL_TEST_LOCK;
2481 /* Save the old mode so that if the mode in the lock changes we
2482 * can decrement the appropriate reader or writer refcount. */
2483 file_lock->fl_type = einfo.ei_mode;
2486 CERROR("unknown fcntl lock command: %d\n", cmd);
2490 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2491 LUSTRE_OPC_ANY, NULL);
2492 if (IS_ERR(op_data))
2493 RETURN(PTR_ERR(op_data));
2495 CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#x, mode=%u, "
2496 "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid,
2497 flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2499 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2500 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2502 if ((file_lock->fl_flags & FL_FLOCK) &&
2503 (rc == 0 || file_lock->fl_type == F_UNLCK))
2504 rc2 = flock_lock_file_wait(file, file_lock);
2505 if ((file_lock->fl_flags & FL_POSIX) &&
2506 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
2507 !(flags & LDLM_FL_TEST_LOCK))
2508 rc2 = posix_lock_file_wait(file, file_lock);
2510 if (rc2 && file_lock->fl_type != F_UNLCK) {
2511 einfo.ei_mode = LCK_NL;
2512 md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2513 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2517 ll_finish_md_op_data(op_data);
2522 int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2530 * test if some locks matching bits and l_req_mode are acquired
2531 * - bits can be in different locks
2532 * - if found clear the common lock bits in *bits
2533 * - the bits not found, are kept in *bits
2535 * \param bits [IN] searched lock bits [IN]
2536 * \param l_req_mode [IN] searched lock mode
2537 * \retval boolean, true iff all bits are found
2539 int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode)
2541 struct lustre_handle lockh;
2542 ldlm_policy_data_t policy;
2543 ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
2544 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
2553 fid = &ll_i2info(inode)->lli_fid;
2554 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
2555 ldlm_lockname[mode]);
2557 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2558 for (i = 0; i < MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2559 policy.l_inodebits.bits = *bits & (1 << i);
2560 if (policy.l_inodebits.bits == 0)
2563 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2564 &policy, mode, &lockh)) {
2565 struct ldlm_lock *lock;
2567 lock = ldlm_handle2lock(&lockh);
2570 ~(lock->l_policy_data.l_inodebits.bits);
2571 LDLM_LOCK_PUT(lock);
2573 *bits &= ~policy.l_inodebits.bits;
2580 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
2581 struct lustre_handle *lockh, __u64 flags)
2583 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
2588 fid = &ll_i2info(inode)->lli_fid;
2589 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2591 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
2592 fid, LDLM_IBITS, &policy,
2593 LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
2597 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2599 /* Already unlinked. Just update nlink and return success */
2600 if (rc == -ENOENT) {
2602 /* This path cannot be hit for regular files unless in
2603 * case of obscure races, so no need to to validate
2605 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2607 } else if (rc != 0) {
2608 CERROR("%s: revalidate FID "DFID" error: rc = %d\n",
2609 ll_get_fsname(inode->i_sb, NULL, 0),
2610 PFID(ll_inode2fid(inode)), rc);
2616 int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
2619 struct inode *inode = dentry->d_inode;
2620 struct ptlrpc_request *req = NULL;
2621 struct obd_export *exp;
2625 LASSERT(inode != NULL);
2627 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n",
2628 inode->i_ino, inode->i_generation, inode, dentry->d_name.name);
2630 exp = ll_i2mdexp(inode);
2632 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2633 * But under CMD case, it caused some lock issues, should be fixed
2634 * with new CMD ibits lock. See bug 12718 */
2635 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
2636 struct lookup_intent oit = { .it_op = IT_GETATTR };
2637 struct md_op_data *op_data;
2639 if (ibits == MDS_INODELOCK_LOOKUP)
2640 oit.it_op = IT_LOOKUP;
2642 /* Call getattr by fid, so do not provide name at all. */
2643 op_data = ll_prep_md_op_data(NULL, dentry->d_parent->d_inode,
2644 dentry->d_inode, NULL, 0, 0,
2645 LUSTRE_OPC_ANY, NULL);
2646 if (IS_ERR(op_data))
2647 RETURN(PTR_ERR(op_data));
2649 oit.it_create_mode |= M_CHECK_STALE;
2650 rc = md_intent_lock(exp, op_data, NULL, 0,
2651 /* we are not interested in name
2654 ll_md_blocking_ast, 0);
2655 ll_finish_md_op_data(op_data);
2656 oit.it_create_mode &= ~M_CHECK_STALE;
2658 rc = ll_inode_revalidate_fini(inode, rc);
2662 rc = ll_revalidate_it_finish(req, &oit, dentry);
2664 ll_intent_release(&oit);
2668 /* Unlinked? Unhash dentry, so it is not picked up later by
2669 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2670 here to preserve get_cwd functionality on 2.6.
2672 if (!dentry->d_inode->i_nlink)
2673 d_lustre_invalidate(dentry);
2675 ll_lookup_finish_locks(&oit, dentry);
2676 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
2677 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
2678 obd_valid valid = OBD_MD_FLGETATTR;
2679 struct md_op_data *op_data;
2682 if (S_ISREG(inode->i_mode)) {
2683 rc = ll_get_max_mdsize(sbi, &ealen);
2686 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2689 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2690 0, ealen, LUSTRE_OPC_ANY,
2692 if (IS_ERR(op_data))
2693 RETURN(PTR_ERR(op_data));
2695 op_data->op_valid = valid;
2696 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
2697 * capa for this inode. Because we only keep capas of dirs
2699 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2700 ll_finish_md_op_data(op_data);
2702 rc = ll_inode_revalidate_fini(inode, rc);
2706 rc = ll_prep_inode(&inode, req, NULL, NULL);
2709 ptlrpc_req_finished(req);
2713 int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
2716 struct inode *inode = dentry->d_inode;
2720 rc = __ll_inode_revalidate_it(dentry, it, ibits);
2724 /* if object isn't regular file, don't validate size */
2725 if (!S_ISREG(inode->i_mode)) {
2726 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
2727 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
2728 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
2730 rc = ll_glimpse_size(inode);
2735 int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
2736 struct lookup_intent *it, struct kstat *stat)
2738 struct inode *inode = de->d_inode;
2739 struct ll_sb_info *sbi = ll_i2sbi(inode);
2740 struct ll_inode_info *lli = ll_i2info(inode);
2743 res = ll_inode_revalidate_it(de, it, MDS_INODELOCK_UPDATE |
2744 MDS_INODELOCK_LOOKUP);
2745 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
2750 stat->dev = inode->i_sb->s_dev;
2751 if (ll_need_32bit_api(sbi))
2752 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
2754 stat->ino = inode->i_ino;
2755 stat->mode = inode->i_mode;
2756 stat->nlink = inode->i_nlink;
2757 stat->uid = inode->i_uid;
2758 stat->gid = inode->i_gid;
2759 stat->rdev = inode->i_rdev;
2760 stat->atime = inode->i_atime;
2761 stat->mtime = inode->i_mtime;
2762 stat->ctime = inode->i_ctime;
2763 stat->blksize = 1 << inode->i_blkbits;
2765 stat->size = i_size_read(inode);
2766 stat->blocks = inode->i_blocks;
2770 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
2772 struct lookup_intent it = { .it_op = IT_GETATTR };
2774 return ll_getattr_it(mnt, de, &it, stat);
2777 #ifdef HAVE_LINUX_FIEMAP_H
2778 int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2779 __u64 start, __u64 len)
2783 struct ll_user_fiemap *fiemap;
2784 unsigned int extent_count = fieinfo->fi_extents_max;
2786 num_bytes = sizeof(*fiemap) + (extent_count *
2787 sizeof(struct ll_fiemap_extent));
2788 OBD_ALLOC_LARGE(fiemap, num_bytes);
2793 fiemap->fm_flags = fieinfo->fi_flags;
2794 fiemap->fm_extent_count = fieinfo->fi_extents_max;
2795 fiemap->fm_start = start;
2796 fiemap->fm_length = len;
2797 memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
2798 sizeof(struct ll_fiemap_extent));
2800 rc = ll_do_fiemap(inode, fiemap, num_bytes);
2802 fieinfo->fi_flags = fiemap->fm_flags;
2803 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
2804 memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
2805 fiemap->fm_mapped_extents * sizeof(struct ll_fiemap_extent));
2807 OBD_FREE_LARGE(fiemap, num_bytes);
2812 struct posix_acl * ll_get_acl(struct inode *inode, int type)
2814 struct ll_inode_info *lli = ll_i2info(inode);
2815 struct posix_acl *acl = NULL;
2818 spin_lock(&lli->lli_lock);
2819 /* VFS' acl_permission_check->check_acl will release the refcount */
2820 acl = posix_acl_dup(lli->lli_posix_acl);
2821 spin_unlock(&lli->lli_lock);
2826 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
2828 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
2829 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
2831 ll_check_acl(struct inode *inode, int mask)
2834 # ifdef CONFIG_FS_POSIX_ACL
2835 struct posix_acl *acl;
2839 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
2840 if (flags & IPERM_FLAG_RCU)
2843 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
2848 rc = posix_acl_permission(inode, acl, mask);
2849 posix_acl_release(acl);
2852 # else /* !CONFIG_FS_POSIX_ACL */
2854 # endif /* CONFIG_FS_POSIX_ACL */
2856 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
2858 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
2859 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
2861 # ifdef HAVE_INODE_PERMISION_2ARGS
2862 int ll_inode_permission(struct inode *inode, int mask)
2864 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
2871 #ifdef MAY_NOT_BLOCK
2872 if (mask & MAY_NOT_BLOCK)
2874 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
2875 if (flags & IPERM_FLAG_RCU)
2879 /* as root inode are NOT getting validated in lookup operation,
2880 * need to do it before permission check. */
2882 if (inode == inode->i_sb->s_root->d_inode) {
2883 struct lookup_intent it = { .it_op = IT_LOOKUP };
2885 rc = __ll_inode_revalidate_it(inode->i_sb->s_root, &it,
2886 MDS_INODELOCK_LOOKUP);
2891 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), inode mode %x mask %o\n",
2892 inode->i_ino, inode->i_generation, inode, inode->i_mode, mask);
2894 if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
2895 return lustre_check_remote_perm(inode, mask);
2897 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
2898 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
2903 #ifdef HAVE_FILE_READV
2904 #define READ_METHOD readv
2905 #define READ_FUNCTION ll_file_readv
2906 #define WRITE_METHOD writev
2907 #define WRITE_FUNCTION ll_file_writev
2909 #define READ_METHOD aio_read
2910 #define READ_FUNCTION ll_file_aio_read
2911 #define WRITE_METHOD aio_write
2912 #define WRITE_FUNCTION ll_file_aio_write
2915 /* -o localflock - only provides locally consistent flock locks */
2916 struct file_operations ll_file_operations = {
2917 .read = ll_file_read,
2918 .READ_METHOD = READ_FUNCTION,
2919 .write = ll_file_write,
2920 .WRITE_METHOD = WRITE_FUNCTION,
2921 .unlocked_ioctl = ll_file_ioctl,
2922 .open = ll_file_open,
2923 .release = ll_file_release,
2924 .mmap = ll_file_mmap,
2925 .llseek = ll_file_seek,
2926 #ifdef HAVE_KERNEL_SENDFILE
2927 .sendfile = ll_file_sendfile,
2929 #ifdef HAVE_KERNEL_SPLICE_READ
2930 .splice_read = ll_file_splice_read,
2936 struct file_operations ll_file_operations_flock = {
2937 .read = ll_file_read,
2938 .READ_METHOD = READ_FUNCTION,
2939 .write = ll_file_write,
2940 .WRITE_METHOD = WRITE_FUNCTION,
2941 .unlocked_ioctl = ll_file_ioctl,
2942 .open = ll_file_open,
2943 .release = ll_file_release,
2944 .mmap = ll_file_mmap,
2945 .llseek = ll_file_seek,
2946 #ifdef HAVE_KERNEL_SENDFILE
2947 .sendfile = ll_file_sendfile,
2949 #ifdef HAVE_KERNEL_SPLICE_READ
2950 .splice_read = ll_file_splice_read,
2954 .flock = ll_file_flock,
2955 .lock = ll_file_flock
2958 /* These are for -o noflock - to return ENOSYS on flock calls */
2959 struct file_operations ll_file_operations_noflock = {
2960 .read = ll_file_read,
2961 .READ_METHOD = READ_FUNCTION,
2962 .write = ll_file_write,
2963 .WRITE_METHOD = WRITE_FUNCTION,
2964 .unlocked_ioctl = ll_file_ioctl,
2965 .open = ll_file_open,
2966 .release = ll_file_release,
2967 .mmap = ll_file_mmap,
2968 .llseek = ll_file_seek,
2969 #ifdef HAVE_KERNEL_SENDFILE
2970 .sendfile = ll_file_sendfile,
2972 #ifdef HAVE_KERNEL_SPLICE_READ
2973 .splice_read = ll_file_splice_read,
2977 .flock = ll_file_noflock,
2978 .lock = ll_file_noflock
2981 struct inode_operations ll_file_inode_operations = {
2982 .setattr = ll_setattr,
2983 .getattr = ll_getattr,
2984 .permission = ll_inode_permission,
2985 .setxattr = ll_setxattr,
2986 .getxattr = ll_getxattr,
2987 .listxattr = ll_listxattr,
2988 .removexattr = ll_removexattr,
2989 #ifdef HAVE_LINUX_FIEMAP_H
2990 .fiemap = ll_fiemap,
2992 #ifdef HAVE_IOP_GET_ACL
2993 .get_acl = ll_get_acl,
2997 /* dynamic ioctl number support routins */
2998 static struct llioc_ctl_data {
2999 struct rw_semaphore ioc_sem;
3000 cfs_list_t ioc_head;
3002 __RWSEM_INITIALIZER(llioc.ioc_sem),
3003 CFS_LIST_HEAD_INIT(llioc.ioc_head)
3008 cfs_list_t iocd_list;
3009 unsigned int iocd_size;
3010 llioc_callback_t iocd_cb;
3011 unsigned int iocd_count;
3012 unsigned int iocd_cmd[0];
3015 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3018 struct llioc_data *in_data = NULL;
3021 if (cb == NULL || cmd == NULL ||
3022 count > LLIOC_MAX_CMD || count < 0)
3025 size = sizeof(*in_data) + count * sizeof(unsigned int);
3026 OBD_ALLOC(in_data, size);
3027 if (in_data == NULL)
3030 memset(in_data, 0, sizeof(*in_data));
3031 in_data->iocd_size = size;
3032 in_data->iocd_cb = cb;
3033 in_data->iocd_count = count;
3034 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3036 down_write(&llioc.ioc_sem);
3037 cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3038 up_write(&llioc.ioc_sem);
3043 void ll_iocontrol_unregister(void *magic)
3045 struct llioc_data *tmp;
3050 down_write(&llioc.ioc_sem);
3051 cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3053 unsigned int size = tmp->iocd_size;
3055 cfs_list_del(&tmp->iocd_list);
3056 up_write(&llioc.ioc_sem);
3058 OBD_FREE(tmp, size);
3062 up_write(&llioc.ioc_sem);
3064 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3067 EXPORT_SYMBOL(ll_iocontrol_register);
3068 EXPORT_SYMBOL(ll_iocontrol_unregister);
3070 enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
3071 unsigned int cmd, unsigned long arg, int *rcp)
3073 enum llioc_iter ret = LLIOC_CONT;
3074 struct llioc_data *data;
3075 int rc = -EINVAL, i;
3077 down_read(&llioc.ioc_sem);
3078 cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3079 for (i = 0; i < data->iocd_count; i++) {
3080 if (cmd != data->iocd_cmd[i])
3083 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3087 if (ret == LLIOC_STOP)
3090 up_read(&llioc.ioc_sem);
3097 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3099 struct ll_inode_info *lli = ll_i2info(inode);
3100 struct cl_env_nest nest;
3105 if (lli->lli_clob == NULL)
3108 env = cl_env_nested_get(&nest);
3110 RETURN(PTR_ERR(env));
3112 result = cl_conf_set(env, lli->lli_clob, conf);
3113 cl_env_nested_put(&nest, env);
3115 if (conf->coc_opc == OBJECT_CONF_SET) {
3116 struct ldlm_lock *lock = conf->coc_lock;
3118 LASSERT(lock != NULL);
3119 LASSERT(ldlm_has_layout(lock));
3121 /* it can only be allowed to match after layout is
3122 * applied to inode otherwise false layout would be
3123 * seen. Applying layout shoud happen before dropping
3124 * the intent lock. */
3125 ldlm_lock_allow_match(lock);
3132 * Apply the layout to the inode. Layout lock is held and will be released
3135 static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
3136 struct inode *inode, __u32 *gen, bool reconf)
3138 struct ll_inode_info *lli = ll_i2info(inode);
3139 struct ll_sb_info *sbi = ll_i2sbi(inode);
3140 struct ldlm_lock *lock;
3141 struct lustre_md md = { NULL };
3142 struct cl_object_conf conf;
3147 LASSERT(lustre_handle_is_used(lockh));
3149 lock = ldlm_handle2lock(lockh);
3150 LASSERT(lock != NULL);
3151 LASSERT(ldlm_has_layout(lock));
3153 LDLM_DEBUG(lock, "File %p/"DFID" being reconfigured: %d.\n",
3154 inode, PFID(&lli->lli_fid), reconf);
3156 lock_res_and_lock(lock);
3157 lvb_ready = !!(lock->l_flags & LDLM_FL_LVB_READY);
3158 unlock_res_and_lock(lock);
3159 /* checking lvb_ready is racy but this is okay. The worst case is
3160 * that multi processes may configure the file on the same time. */
3161 if (lvb_ready || !reconf) {
3162 LDLM_LOCK_PUT(lock);
3166 /* layout_gen must be valid if layout lock is not
3167 * cancelled and stripe has already set */
3168 *gen = lli->lli_layout_gen;
3171 ldlm_lock_decref(lockh, mode);
3175 /* for layout lock, lmm is returned in lock's lvb.
3176 * lvb_data is immutable if the lock is held so it's safe to access it
3177 * without res lock. See the description in ldlm_lock_decref_internal()
3178 * for the condition to free lvb_data of layout lock */
3179 if (lock->l_lvb_data != NULL) {
3180 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3181 lock->l_lvb_data, lock->l_lvb_len);
3183 *gen = LL_LAYOUT_GEN_EMPTY;
3185 *gen = md.lsm->lsm_layout_gen;
3188 CERROR("%s: file "DFID" unpackmd error: %d\n",
3189 ll_get_fsname(inode->i_sb, NULL, 0),
3190 PFID(&lli->lli_fid), rc);
3194 LDLM_LOCK_PUT(lock);
3195 ldlm_lock_decref(lockh, mode);
3199 /* set layout to file. Unlikely this will fail as old layout was
3200 * surely eliminated */
3201 memset(&conf, 0, sizeof conf);
3202 conf.coc_opc = OBJECT_CONF_SET;
3203 conf.coc_inode = inode;
3204 conf.coc_lock = lock;
3205 conf.u.coc_md = &md;
3206 rc = ll_layout_conf(inode, &conf);
3207 LDLM_LOCK_PUT(lock);
3209 ldlm_lock_decref(lockh, mode);
3212 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3214 /* wait for IO to complete if it's still being used. */
3216 CDEBUG(D_INODE, "%s: %p/"DFID" wait for layout reconf.\n",
3217 ll_get_fsname(inode->i_sb, NULL, 0),
3218 inode, PFID(&lli->lli_fid));
3220 memset(&conf, 0, sizeof conf);
3221 conf.coc_opc = OBJECT_CONF_WAIT;
3222 conf.coc_inode = inode;
3223 rc = ll_layout_conf(inode, &conf);
3227 CDEBUG(D_INODE, "file: "DFID" waiting layout return: %d.\n",
3228 PFID(&lli->lli_fid), rc);
3235 * This function checks if there exists a LAYOUT lock on the client side,
3236 * or enqueues it if it doesn't have one in cache.
3238 * This function will not hold layout lock so it may be revoked any time after
3239 * this function returns. Any operations depend on layout should be redone
3242 * This function should be called before lov_io_init() to get an uptodate
3243 * layout version, the caller should save the version number and after IO
3244 * is finished, this function should be called again to verify that layout
3245 * is not changed during IO time.
3247 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3249 struct ll_inode_info *lli = ll_i2info(inode);
3250 struct ll_sb_info *sbi = ll_i2sbi(inode);
3251 struct md_op_data *op_data;
3252 struct lookup_intent it;
3253 struct lustre_handle lockh;
3255 struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS,
3257 .ei_cb_bl = ll_md_blocking_ast,
3258 .ei_cb_cp = ldlm_completion_ast,
3259 .ei_cbdata = NULL };
3263 *gen = LL_LAYOUT_GEN_NONE;
3264 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
3268 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3269 LASSERT(S_ISREG(inode->i_mode));
3271 /* mostly layout lock is caching on the local side, so try to match
3272 * it before grabbing layout lock mutex. */
3273 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0);
3274 if (mode != 0) { /* hit cached lock */
3275 rc = ll_layout_lock_set(&lockh, mode, inode, gen, false);
3279 /* better hold lli_layout_mutex to try again otherwise
3280 * it will have starvation problem. */
3283 /* take layout lock mutex to enqueue layout lock exclusively. */
3284 mutex_lock(&lli->lli_layout_mutex);
3287 /* try again. Maybe somebody else has done this. */
3288 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0);
3289 if (mode != 0) { /* hit cached lock */
3290 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3294 mutex_unlock(&lli->lli_layout_mutex);
3298 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3299 0, 0, LUSTRE_OPC_ANY, NULL);
3300 if (IS_ERR(op_data)) {
3301 mutex_unlock(&lli->lli_layout_mutex);
3302 RETURN(PTR_ERR(op_data));
3305 /* have to enqueue one */
3306 memset(&it, 0, sizeof(it));
3307 it.it_op = IT_LAYOUT;
3308 lockh.cookie = 0ULL;
3310 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file %p/"DFID".\n",
3311 ll_get_fsname(inode->i_sb, NULL, 0), inode,
3312 PFID(&lli->lli_fid));
3314 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
3316 if (it.d.lustre.it_data != NULL)
3317 ptlrpc_req_finished(it.d.lustre.it_data);
3318 it.d.lustre.it_data = NULL;
3320 ll_finish_md_op_data(op_data);
3322 md_set_lock_data(sbi->ll_md_exp, &it.d.lustre.it_lock_handle, inode, NULL);
3324 mode = it.d.lustre.it_lock_mode;
3325 it.d.lustre.it_lock_mode = 0;
3326 ll_intent_drop_lock(&it);
3329 /* set lock data in case this is a new lock */
3330 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3331 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3335 mutex_unlock(&lli->lli_layout_mutex);