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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
40 #include <lustre_dlm.h>
41 #include <linux/pagemap.h>
42 #include <linux/file.h>
43 #include <linux/sched.h>
44 #include <linux/user_namespace.h>
45 #ifdef HAVE_UIDGID_HEADER
46 # include <linux/uidgid.h>
49 #include <uapi/linux/lustre/lustre_ioctl.h>
50 #include <lustre_swab.h>
52 #include "cl_object.h"
53 #include "llite_internal.h"
54 #include "vvp_internal.h"
57 struct inode *sp_inode;
62 __u64 pa_data_version;
68 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
70 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
73 static struct ll_file_data *ll_file_data_get(void)
75 struct ll_file_data *fd;
77 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
81 fd->fd_write_failed = false;
82 pcc_file_init(&fd->fd_pcc_file);
87 static void ll_file_data_put(struct ll_file_data *fd)
90 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
94 * Packs all the attributes into @op_data for the CLOSE rpc.
96 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
97 struct obd_client_handle *och)
101 ll_prep_md_op_data(op_data, inode, NULL, NULL,
102 0, 0, LUSTRE_OPC_ANY, NULL);
104 op_data->op_attr.ia_mode = inode->i_mode;
105 op_data->op_attr.ia_atime = inode->i_atime;
106 op_data->op_attr.ia_mtime = inode->i_mtime;
107 op_data->op_attr.ia_ctime = inode->i_ctime;
108 op_data->op_attr.ia_size = i_size_read(inode);
109 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
110 ATTR_MTIME | ATTR_MTIME_SET |
112 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
113 op_data->op_attr_blocks = inode->i_blocks;
114 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
115 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
116 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
117 op_data->op_open_handle = och->och_open_handle;
119 if (och->och_flags & FMODE_WRITE &&
120 ll_file_test_and_clear_flag(ll_i2info(inode), LLIF_DATA_MODIFIED))
121 /* For HSM: if inode data has been modified, pack it so that
122 * MDT can set data dirty flag in the archive. */
123 op_data->op_bias |= MDS_DATA_MODIFIED;
129 * Perform a close, possibly with a bias.
130 * The meaning of "data" depends on the value of "bias".
132 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
133 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
136 static int ll_close_inode_openhandle(struct inode *inode,
137 struct obd_client_handle *och,
138 enum mds_op_bias bias, void *data)
140 struct obd_export *md_exp = ll_i2mdexp(inode);
141 const struct ll_inode_info *lli = ll_i2info(inode);
142 struct md_op_data *op_data;
143 struct ptlrpc_request *req = NULL;
147 if (class_exp2obd(md_exp) == NULL) {
148 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
149 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
153 OBD_ALLOC_PTR(op_data);
154 /* We leak openhandle and request here on error, but not much to be
155 * done in OOM case since app won't retry close on error either. */
157 GOTO(out, rc = -ENOMEM);
159 ll_prepare_close(inode, op_data, och);
161 case MDS_CLOSE_LAYOUT_MERGE:
162 /* merge blocks from the victim inode */
163 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
164 op_data->op_attr.ia_valid |= ATTR_SIZE;
165 op_data->op_xvalid |= OP_XVALID_BLOCKS;
166 case MDS_CLOSE_LAYOUT_SPLIT:
167 case MDS_CLOSE_LAYOUT_SWAP: {
168 struct split_param *sp = data;
170 LASSERT(data != NULL);
171 op_data->op_bias |= bias;
172 op_data->op_data_version = 0;
173 op_data->op_lease_handle = och->och_lease_handle;
174 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
175 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
176 op_data->op_mirror_id = sp->sp_mirror_id;
178 op_data->op_fid2 = *ll_inode2fid(data);
183 case MDS_CLOSE_RESYNC_DONE: {
184 struct ll_ioc_lease *ioc = data;
186 LASSERT(data != NULL);
187 op_data->op_attr_blocks +=
188 ioc->lil_count * op_data->op_attr_blocks;
189 op_data->op_attr.ia_valid |= ATTR_SIZE;
190 op_data->op_xvalid |= OP_XVALID_BLOCKS;
191 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
193 op_data->op_lease_handle = och->och_lease_handle;
194 op_data->op_data = &ioc->lil_ids[0];
195 op_data->op_data_size =
196 ioc->lil_count * sizeof(ioc->lil_ids[0]);
200 case MDS_PCC_ATTACH: {
201 struct pcc_param *param = data;
203 LASSERT(data != NULL);
204 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
205 op_data->op_archive_id = param->pa_archive_id;
206 op_data->op_data_version = param->pa_data_version;
207 op_data->op_lease_handle = och->och_lease_handle;
211 case MDS_HSM_RELEASE:
212 LASSERT(data != NULL);
213 op_data->op_bias |= MDS_HSM_RELEASE;
214 op_data->op_data_version = *(__u64 *)data;
215 op_data->op_lease_handle = och->och_lease_handle;
216 op_data->op_attr.ia_valid |= ATTR_SIZE;
217 op_data->op_xvalid |= OP_XVALID_BLOCKS;
221 LASSERT(data == NULL);
225 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
226 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
227 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
228 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
230 rc = md_close(md_exp, op_data, och->och_mod, &req);
231 if (rc != 0 && rc != -EINTR)
232 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
233 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
235 if (rc == 0 && op_data->op_bias & bias) {
236 struct mdt_body *body;
238 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
239 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
242 if (bias & MDS_PCC_ATTACH) {
243 struct pcc_param *param = data;
245 param->pa_layout_gen = body->mbo_layout_gen;
249 ll_finish_md_op_data(op_data);
253 md_clear_open_replay_data(md_exp, och);
254 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
257 ptlrpc_req_finished(req); /* This is close request */
261 int ll_md_real_close(struct inode *inode, fmode_t fmode)
263 struct ll_inode_info *lli = ll_i2info(inode);
264 struct obd_client_handle **och_p;
265 struct obd_client_handle *och;
270 if (fmode & FMODE_WRITE) {
271 och_p = &lli->lli_mds_write_och;
272 och_usecount = &lli->lli_open_fd_write_count;
273 } else if (fmode & FMODE_EXEC) {
274 och_p = &lli->lli_mds_exec_och;
275 och_usecount = &lli->lli_open_fd_exec_count;
277 LASSERT(fmode & FMODE_READ);
278 och_p = &lli->lli_mds_read_och;
279 och_usecount = &lli->lli_open_fd_read_count;
282 mutex_lock(&lli->lli_och_mutex);
283 if (*och_usecount > 0) {
284 /* There are still users of this handle, so skip
286 mutex_unlock(&lli->lli_och_mutex);
292 mutex_unlock(&lli->lli_och_mutex);
295 /* There might be a race and this handle may already
297 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
303 static int ll_md_close(struct inode *inode, struct file *file)
305 union ldlm_policy_data policy = {
306 .l_inodebits = { MDS_INODELOCK_OPEN },
308 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
309 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
310 struct ll_inode_info *lli = ll_i2info(inode);
311 struct lustre_handle lockh;
312 enum ldlm_mode lockmode;
316 /* clear group lock, if present */
317 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
318 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
320 if (fd->fd_lease_och != NULL) {
323 /* Usually the lease is not released when the
324 * application crashed, we need to release here. */
325 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
326 CDEBUG(rc ? D_ERROR : D_INODE, "Clean up lease "DFID" %d/%d\n",
327 PFID(&lli->lli_fid), rc, lease_broken);
329 fd->fd_lease_och = NULL;
332 if (fd->fd_och != NULL) {
333 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
338 /* Let's see if we have good enough OPEN lock on the file and if
339 we can skip talking to MDS */
340 mutex_lock(&lli->lli_och_mutex);
341 if (fd->fd_omode & FMODE_WRITE) {
343 LASSERT(lli->lli_open_fd_write_count);
344 lli->lli_open_fd_write_count--;
345 } else if (fd->fd_omode & FMODE_EXEC) {
347 LASSERT(lli->lli_open_fd_exec_count);
348 lli->lli_open_fd_exec_count--;
351 LASSERT(lli->lli_open_fd_read_count);
352 lli->lli_open_fd_read_count--;
354 mutex_unlock(&lli->lli_och_mutex);
356 if (!md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
357 LDLM_IBITS, &policy, lockmode, &lockh))
358 rc = ll_md_real_close(inode, fd->fd_omode);
361 LUSTRE_FPRIVATE(file) = NULL;
362 ll_file_data_put(fd);
367 /* While this returns an error code, fput() the caller does not, so we need
368 * to make every effort to clean up all of our state here. Also, applications
369 * rarely check close errors and even if an error is returned they will not
370 * re-try the close call.
372 int ll_file_release(struct inode *inode, struct file *file)
374 struct ll_file_data *fd;
375 struct ll_sb_info *sbi = ll_i2sbi(inode);
376 struct ll_inode_info *lli = ll_i2info(inode);
380 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
381 PFID(ll_inode2fid(inode)), inode);
383 if (inode->i_sb->s_root != file_dentry(file))
384 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
385 fd = LUSTRE_FPRIVATE(file);
388 /* The last ref on @file, maybe not the the owner pid of statahead,
389 * because parent and child process can share the same file handle. */
390 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
391 ll_deauthorize_statahead(inode, fd);
393 if (inode->i_sb->s_root == file_dentry(file)) {
394 LUSTRE_FPRIVATE(file) = NULL;
395 ll_file_data_put(fd);
399 pcc_file_release(inode, file);
401 if (!S_ISDIR(inode->i_mode)) {
402 if (lli->lli_clob != NULL)
403 lov_read_and_clear_async_rc(lli->lli_clob);
404 lli->lli_async_rc = 0;
407 rc = ll_md_close(inode, file);
409 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
410 libcfs_debug_dumplog();
415 static inline int ll_dom_readpage(void *data, struct page *page)
417 struct niobuf_local *lnb = data;
420 kaddr = ll_kmap_atomic(page, KM_USER0);
421 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
422 if (lnb->lnb_len < PAGE_SIZE)
423 memset(kaddr + lnb->lnb_len, 0,
424 PAGE_SIZE - lnb->lnb_len);
425 flush_dcache_page(page);
426 SetPageUptodate(page);
427 ll_kunmap_atomic(kaddr, KM_USER0);
433 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
434 struct lookup_intent *it)
436 struct ll_inode_info *lli = ll_i2info(inode);
437 struct cl_object *obj = lli->lli_clob;
438 struct address_space *mapping = inode->i_mapping;
440 struct niobuf_remote *rnb;
441 struct mdt_body *body;
443 unsigned long index, start;
444 struct niobuf_local lnb;
451 if (!req_capsule_has_field(&req->rq_pill, &RMF_NIOBUF_INLINE,
455 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
456 if (rnb == NULL || rnb->rnb_len == 0)
459 /* LU-11595: Server may return whole file and that is OK always or
460 * it may return just file tail and its offset must be aligned with
461 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
462 * smaller then offset may be not aligned and that data is just ignored.
464 if (rnb->rnb_offset % PAGE_SIZE)
467 /* Server returns whole file or just file tail if it fills in reply
468 * buffer, in both cases total size should be equal to the file size.
470 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
471 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size) {
472 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
473 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
474 rnb->rnb_len, body->mbo_dom_size);
478 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
479 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
481 data = (char *)rnb + sizeof(*rnb);
483 lnb.lnb_file_offset = rnb->rnb_offset;
484 start = lnb.lnb_file_offset / PAGE_SIZE;
486 LASSERT(lnb.lnb_file_offset % PAGE_SIZE == 0);
487 lnb.lnb_page_offset = 0;
489 lnb.lnb_data = data + (index << PAGE_SHIFT);
490 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
491 if (lnb.lnb_len > PAGE_SIZE)
492 lnb.lnb_len = PAGE_SIZE;
494 vmpage = read_cache_page(mapping, index + start,
495 ll_dom_readpage, &lnb);
496 if (IS_ERR(vmpage)) {
497 CWARN("%s: cannot fill page %lu for "DFID
498 " with data: rc = %li\n",
499 ll_i2sbi(inode)->ll_fsname, index + start,
500 PFID(lu_object_fid(&obj->co_lu)),
506 } while (rnb->rnb_len > (index << PAGE_SHIFT));
510 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
511 struct lookup_intent *itp)
513 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
514 struct dentry *parent = de->d_parent;
517 struct md_op_data *op_data;
518 struct ptlrpc_request *req = NULL;
522 LASSERT(parent != NULL);
523 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
525 /* if server supports open-by-fid, or file name is invalid, don't pack
526 * name in open request */
527 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
528 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
530 len = de->d_name.len;
531 name = kmalloc(len + 1, GFP_NOFS);
536 spin_lock(&de->d_lock);
537 if (len != de->d_name.len) {
538 spin_unlock(&de->d_lock);
542 memcpy(name, de->d_name.name, len);
544 spin_unlock(&de->d_lock);
546 if (!lu_name_is_valid_2(name, len)) {
552 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
553 name, len, 0, LUSTRE_OPC_ANY, NULL);
554 if (IS_ERR(op_data)) {
556 RETURN(PTR_ERR(op_data));
558 op_data->op_data = lmm;
559 op_data->op_data_size = lmmsize;
561 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
562 &ll_md_blocking_ast, 0);
564 ll_finish_md_op_data(op_data);
566 /* reason for keep own exit path - don`t flood log
567 * with messages with -ESTALE errors.
569 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
570 it_open_error(DISP_OPEN_OPEN, itp))
572 ll_release_openhandle(de, itp);
576 if (it_disposition(itp, DISP_LOOKUP_NEG))
577 GOTO(out, rc = -ENOENT);
579 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
580 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
581 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
585 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
587 if (!rc && itp->it_lock_mode) {
588 struct lustre_handle handle = {.cookie = itp->it_lock_handle};
589 struct ldlm_lock *lock;
590 bool has_dom_bit = false;
592 /* If we got a lock back and it has a LOOKUP bit set,
593 * make sure the dentry is marked as valid so we can find it.
594 * We don't need to care about actual hashing since other bits
595 * of kernel will deal with that later.
597 lock = ldlm_handle2lock(&handle);
599 has_dom_bit = ldlm_has_dom(lock);
600 if (lock->l_policy_data.l_inodebits.bits &
601 MDS_INODELOCK_LOOKUP)
602 d_lustre_revalidate(de);
606 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
608 ll_dom_finish_open(de->d_inode, req, itp);
612 ptlrpc_req_finished(req);
613 ll_intent_drop_lock(itp);
615 /* We did open by fid, but by the time we got to the server,
616 * the object disappeared. If this is a create, we cannot really
617 * tell the userspace that the file it was trying to create
618 * does not exist. Instead let's return -ESTALE, and the VFS will
619 * retry the create with LOOKUP_REVAL that we are going to catch
620 * in ll_revalidate_dentry() and use lookup then.
622 if (rc == -ENOENT && itp->it_op & IT_CREAT)
628 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
629 struct obd_client_handle *och)
631 struct mdt_body *body;
633 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
634 och->och_open_handle = body->mbo_open_handle;
635 och->och_fid = body->mbo_fid1;
636 och->och_lease_handle.cookie = it->it_lock_handle;
637 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
638 och->och_flags = it->it_flags;
640 return md_set_open_replay_data(md_exp, och, it);
643 static int ll_local_open(struct file *file, struct lookup_intent *it,
644 struct ll_file_data *fd, struct obd_client_handle *och)
646 struct inode *inode = file_inode(file);
649 LASSERT(!LUSTRE_FPRIVATE(file));
656 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
661 LUSTRE_FPRIVATE(file) = fd;
662 ll_readahead_init(inode, &fd->fd_ras);
663 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
665 /* ll_cl_context initialize */
666 rwlock_init(&fd->fd_lock);
667 INIT_LIST_HEAD(&fd->fd_lccs);
672 /* Open a file, and (for the very first open) create objects on the OSTs at
673 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
674 * creation or open until ll_lov_setstripe() ioctl is called.
676 * If we already have the stripe MD locally then we don't request it in
677 * md_open(), by passing a lmm_size = 0.
679 * It is up to the application to ensure no other processes open this file
680 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
681 * used. We might be able to avoid races of that sort by getting lli_open_sem
682 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
683 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
685 int ll_file_open(struct inode *inode, struct file *file)
687 struct ll_inode_info *lli = ll_i2info(inode);
688 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
689 .it_flags = file->f_flags };
690 struct obd_client_handle **och_p = NULL;
691 __u64 *och_usecount = NULL;
692 struct ll_file_data *fd;
696 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
697 PFID(ll_inode2fid(inode)), inode, file->f_flags);
699 it = file->private_data; /* XXX: compat macro */
700 file->private_data = NULL; /* prevent ll_local_open assertion */
702 fd = ll_file_data_get();
704 GOTO(out_nofiledata, rc = -ENOMEM);
707 if (S_ISDIR(inode->i_mode))
708 ll_authorize_statahead(inode, fd);
710 if (inode->i_sb->s_root == file_dentry(file)) {
711 LUSTRE_FPRIVATE(file) = fd;
715 if (!it || !it->it_disposition) {
716 /* Convert f_flags into access mode. We cannot use file->f_mode,
717 * because everything but O_ACCMODE mask was stripped from
719 if ((oit.it_flags + 1) & O_ACCMODE)
721 if (file->f_flags & O_TRUNC)
722 oit.it_flags |= FMODE_WRITE;
724 /* kernel only call f_op->open in dentry_open. filp_open calls
725 * dentry_open after call to open_namei that checks permissions.
726 * Only nfsd_open call dentry_open directly without checking
727 * permissions and because of that this code below is safe.
729 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
730 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
732 /* We do not want O_EXCL here, presumably we opened the file
733 * already? XXX - NFS implications? */
734 oit.it_flags &= ~O_EXCL;
736 /* bug20584, if "it_flags" contains O_CREAT, the file will be
737 * created if necessary, then "IT_CREAT" should be set to keep
738 * consistent with it */
739 if (oit.it_flags & O_CREAT)
740 oit.it_op |= IT_CREAT;
746 /* Let's see if we have file open on MDS already. */
747 if (it->it_flags & FMODE_WRITE) {
748 och_p = &lli->lli_mds_write_och;
749 och_usecount = &lli->lli_open_fd_write_count;
750 } else if (it->it_flags & FMODE_EXEC) {
751 och_p = &lli->lli_mds_exec_och;
752 och_usecount = &lli->lli_open_fd_exec_count;
754 och_p = &lli->lli_mds_read_och;
755 och_usecount = &lli->lli_open_fd_read_count;
758 mutex_lock(&lli->lli_och_mutex);
759 if (*och_p) { /* Open handle is present */
760 if (it_disposition(it, DISP_OPEN_OPEN)) {
761 /* Well, there's extra open request that we do not need,
762 let's close it somehow. This will decref request. */
763 rc = it_open_error(DISP_OPEN_OPEN, it);
765 mutex_unlock(&lli->lli_och_mutex);
766 GOTO(out_openerr, rc);
769 ll_release_openhandle(file_dentry(file), it);
773 rc = ll_local_open(file, it, fd, NULL);
776 mutex_unlock(&lli->lli_och_mutex);
777 GOTO(out_openerr, rc);
780 LASSERT(*och_usecount == 0);
781 if (!it->it_disposition) {
782 struct ll_dentry_data *ldd = ll_d2d(file->f_path.dentry);
783 /* We cannot just request lock handle now, new ELC code
784 means that one of other OPEN locks for this file
785 could be cancelled, and since blocking ast handler
786 would attempt to grab och_mutex as well, that would
787 result in a deadlock */
788 mutex_unlock(&lli->lli_och_mutex);
790 * Normally called under two situations:
792 * 2. A race/condition on MDS resulting in no open
793 * handle to be returned from LOOKUP|OPEN request,
794 * for example if the target entry was a symlink.
796 * Only fetch MDS_OPEN_LOCK if this is in NFS path,
797 * marked by a bit set in ll_iget_for_nfs. Clear the
798 * bit so that it's not confusing later callers.
800 * NB; when ldd is NULL, it must have come via normal
801 * lookup path only, since ll_iget_for_nfs always calls
804 if (ldd && ldd->lld_nfs_dentry) {
805 ldd->lld_nfs_dentry = 0;
806 it->it_flags |= MDS_OPEN_LOCK;
810 * Always specify MDS_OPEN_BY_FID because we don't want
811 * to get file with different fid.
813 it->it_flags |= MDS_OPEN_BY_FID;
814 rc = ll_intent_file_open(file_dentry(file), NULL, 0,
817 GOTO(out_openerr, rc);
821 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
823 GOTO(out_och_free, rc = -ENOMEM);
827 /* md_intent_lock() didn't get a request ref if there was an
828 * open error, so don't do cleanup on the request here
830 /* XXX (green): Should not we bail out on any error here, not
831 * just open error? */
832 rc = it_open_error(DISP_OPEN_OPEN, it);
834 GOTO(out_och_free, rc);
836 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
837 "inode %p: disposition %x, status %d\n", inode,
838 it_disposition(it, ~0), it->it_status);
840 rc = ll_local_open(file, it, fd, *och_p);
842 GOTO(out_och_free, rc);
845 rc = pcc_file_open(inode, file);
847 GOTO(out_och_free, rc);
849 mutex_unlock(&lli->lli_och_mutex);
852 /* Must do this outside lli_och_mutex lock to prevent deadlock where
853 different kind of OPEN lock for this same inode gets cancelled
854 by ldlm_cancel_lru */
855 if (!S_ISREG(inode->i_mode))
856 GOTO(out_och_free, rc);
858 cl_lov_delay_create_clear(&file->f_flags);
859 GOTO(out_och_free, rc);
863 if (och_p && *och_p) {
864 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
865 *och_p = NULL; /* OBD_FREE writes some magic there */
868 mutex_unlock(&lli->lli_och_mutex);
871 if (lli->lli_opendir_key == fd)
872 ll_deauthorize_statahead(inode, fd);
875 ll_file_data_put(fd);
877 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
881 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
882 ptlrpc_req_finished(it->it_request);
883 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
889 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
890 struct ldlm_lock_desc *desc, void *data, int flag)
893 struct lustre_handle lockh;
897 case LDLM_CB_BLOCKING:
898 ldlm_lock2handle(lock, &lockh);
899 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
901 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
905 case LDLM_CB_CANCELING:
913 * When setting a lease on a file, we take ownership of the lli_mds_*_och
914 * and save it as fd->fd_och so as to force client to reopen the file even
915 * if it has an open lock in cache already.
917 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
918 struct lustre_handle *old_open_handle)
920 struct ll_inode_info *lli = ll_i2info(inode);
921 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
922 struct obd_client_handle **och_p;
927 /* Get the openhandle of the file */
928 mutex_lock(&lli->lli_och_mutex);
929 if (fd->fd_lease_och != NULL)
930 GOTO(out_unlock, rc = -EBUSY);
932 if (fd->fd_och == NULL) {
933 if (file->f_mode & FMODE_WRITE) {
934 LASSERT(lli->lli_mds_write_och != NULL);
935 och_p = &lli->lli_mds_write_och;
936 och_usecount = &lli->lli_open_fd_write_count;
938 LASSERT(lli->lli_mds_read_och != NULL);
939 och_p = &lli->lli_mds_read_och;
940 och_usecount = &lli->lli_open_fd_read_count;
943 if (*och_usecount > 1)
944 GOTO(out_unlock, rc = -EBUSY);
951 *old_open_handle = fd->fd_och->och_open_handle;
955 mutex_unlock(&lli->lli_och_mutex);
960 * Release ownership on lli_mds_*_och when putting back a file lease.
962 static int ll_lease_och_release(struct inode *inode, struct file *file)
964 struct ll_inode_info *lli = ll_i2info(inode);
965 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
966 struct obd_client_handle **och_p;
967 struct obd_client_handle *old_och = NULL;
972 mutex_lock(&lli->lli_och_mutex);
973 if (file->f_mode & FMODE_WRITE) {
974 och_p = &lli->lli_mds_write_och;
975 och_usecount = &lli->lli_open_fd_write_count;
977 och_p = &lli->lli_mds_read_och;
978 och_usecount = &lli->lli_open_fd_read_count;
981 /* The file may have been open by another process (broken lease) so
982 * *och_p is not NULL. In this case we should simply increase usecount
985 if (*och_p != NULL) {
986 old_och = fd->fd_och;
993 mutex_unlock(&lli->lli_och_mutex);
996 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1002 * Acquire a lease and open the file.
1004 static struct obd_client_handle *
1005 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1008 struct lookup_intent it = { .it_op = IT_OPEN };
1009 struct ll_sb_info *sbi = ll_i2sbi(inode);
1010 struct md_op_data *op_data;
1011 struct ptlrpc_request *req = NULL;
1012 struct lustre_handle old_open_handle = { 0 };
1013 struct obd_client_handle *och = NULL;
1018 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1019 RETURN(ERR_PTR(-EINVAL));
1022 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1023 RETURN(ERR_PTR(-EPERM));
1025 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1027 RETURN(ERR_PTR(rc));
1032 RETURN(ERR_PTR(-ENOMEM));
1034 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1035 LUSTRE_OPC_ANY, NULL);
1036 if (IS_ERR(op_data))
1037 GOTO(out, rc = PTR_ERR(op_data));
1039 /* To tell the MDT this openhandle is from the same owner */
1040 op_data->op_open_handle = old_open_handle;
1042 it.it_flags = fmode | open_flags;
1043 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1044 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1045 &ll_md_blocking_lease_ast,
1046 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1047 * it can be cancelled which may mislead applications that the lease is
1049 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1050 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1051 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1052 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1053 ll_finish_md_op_data(op_data);
1054 ptlrpc_req_finished(req);
1056 GOTO(out_release_it, rc);
1058 if (it_disposition(&it, DISP_LOOKUP_NEG))
1059 GOTO(out_release_it, rc = -ENOENT);
1061 rc = it_open_error(DISP_OPEN_OPEN, &it);
1063 GOTO(out_release_it, rc);
1065 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1066 ll_och_fill(sbi->ll_md_exp, &it, och);
1068 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1069 GOTO(out_close, rc = -EOPNOTSUPP);
1071 /* already get lease, handle lease lock */
1072 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1073 if (it.it_lock_mode == 0 ||
1074 it.it_lock_bits != MDS_INODELOCK_OPEN) {
1075 /* open lock must return for lease */
1076 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1077 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1079 GOTO(out_close, rc = -EPROTO);
1082 ll_intent_release(&it);
1086 /* Cancel open lock */
1087 if (it.it_lock_mode != 0) {
1088 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1090 it.it_lock_mode = 0;
1091 och->och_lease_handle.cookie = 0ULL;
1093 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1095 CERROR("%s: error closing file "DFID": %d\n",
1096 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1097 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1099 ll_intent_release(&it);
1103 RETURN(ERR_PTR(rc));
1107 * Check whether a layout swap can be done between two inodes.
1109 * \param[in] inode1 First inode to check
1110 * \param[in] inode2 Second inode to check
1112 * \retval 0 on success, layout swap can be performed between both inodes
1113 * \retval negative error code if requirements are not met
1115 static int ll_check_swap_layouts_validity(struct inode *inode1,
1116 struct inode *inode2)
1118 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1121 if (inode_permission(inode1, MAY_WRITE) ||
1122 inode_permission(inode2, MAY_WRITE))
1125 if (inode1->i_sb != inode2->i_sb)
1131 static int ll_swap_layouts_close(struct obd_client_handle *och,
1132 struct inode *inode, struct inode *inode2)
1134 const struct lu_fid *fid1 = ll_inode2fid(inode);
1135 const struct lu_fid *fid2;
1139 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1140 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1142 rc = ll_check_swap_layouts_validity(inode, inode2);
1144 GOTO(out_free_och, rc);
1146 /* We now know that inode2 is a lustre inode */
1147 fid2 = ll_inode2fid(inode2);
1149 rc = lu_fid_cmp(fid1, fid2);
1151 GOTO(out_free_och, rc = -EINVAL);
1153 /* Close the file and {swap,merge} layouts between inode & inode2.
1154 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1155 * because we still need it to pack l_remote_handle to MDT. */
1156 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1159 och = NULL; /* freed in ll_close_inode_openhandle() */
1169 * Release lease and close the file.
1170 * It will check if the lease has ever broken.
1172 static int ll_lease_close_intent(struct obd_client_handle *och,
1173 struct inode *inode,
1174 bool *lease_broken, enum mds_op_bias bias,
1177 struct ldlm_lock *lock;
1178 bool cancelled = true;
1182 lock = ldlm_handle2lock(&och->och_lease_handle);
1184 lock_res_and_lock(lock);
1185 cancelled = ldlm_is_cancel(lock);
1186 unlock_res_and_lock(lock);
1187 LDLM_LOCK_PUT(lock);
1190 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1191 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1193 if (lease_broken != NULL)
1194 *lease_broken = cancelled;
1196 if (!cancelled && !bias)
1197 ldlm_cli_cancel(&och->och_lease_handle, 0);
1199 if (cancelled) { /* no need to excute intent */
1204 rc = ll_close_inode_openhandle(inode, och, bias, data);
1208 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1211 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1215 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1217 static int ll_lease_file_resync(struct obd_client_handle *och,
1218 struct inode *inode, unsigned long arg)
1220 struct ll_sb_info *sbi = ll_i2sbi(inode);
1221 struct md_op_data *op_data;
1222 struct ll_ioc_lease_id ioc;
1223 __u64 data_version_unused;
1227 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1228 LUSTRE_OPC_ANY, NULL);
1229 if (IS_ERR(op_data))
1230 RETURN(PTR_ERR(op_data));
1232 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1236 /* before starting file resync, it's necessary to clean up page cache
1237 * in client memory, otherwise once the layout version is increased,
1238 * writing back cached data will be denied the OSTs. */
1239 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1243 op_data->op_lease_handle = och->och_lease_handle;
1244 op_data->op_mirror_id = ioc.lil_mirror_id;
1245 rc = md_file_resync(sbi->ll_md_exp, op_data);
1251 ll_finish_md_op_data(op_data);
1255 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1257 struct ll_inode_info *lli = ll_i2info(inode);
1258 struct cl_object *obj = lli->lli_clob;
1259 struct cl_attr *attr = vvp_env_thread_attr(env);
1267 ll_inode_size_lock(inode);
1269 /* Merge timestamps the most recently obtained from MDS with
1270 * timestamps obtained from OSTs.
1272 * Do not overwrite atime of inode because it may be refreshed
1273 * by file_accessed() function. If the read was served by cache
1274 * data, there is no RPC to be sent so that atime may not be
1275 * transferred to OSTs at all. MDT only updates atime at close time
1276 * if it's at least 'mdd.*.atime_diff' older.
1277 * All in all, the atime in Lustre does not strictly comply with
1278 * POSIX. Solving this problem needs to send an RPC to MDT for each
1279 * read, this will hurt performance.
1281 if (inode->i_atime.tv_sec < lli->lli_atime ||
1282 lli->lli_update_atime) {
1283 inode->i_atime.tv_sec = lli->lli_atime;
1284 lli->lli_update_atime = 0;
1286 inode->i_mtime.tv_sec = lli->lli_mtime;
1287 inode->i_ctime.tv_sec = lli->lli_ctime;
1289 mtime = inode->i_mtime.tv_sec;
1290 atime = inode->i_atime.tv_sec;
1291 ctime = inode->i_ctime.tv_sec;
1293 cl_object_attr_lock(obj);
1294 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1297 rc = cl_object_attr_get(env, obj, attr);
1298 cl_object_attr_unlock(obj);
1301 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1303 if (atime < attr->cat_atime)
1304 atime = attr->cat_atime;
1306 if (ctime < attr->cat_ctime)
1307 ctime = attr->cat_ctime;
1309 if (mtime < attr->cat_mtime)
1310 mtime = attr->cat_mtime;
1312 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1313 PFID(&lli->lli_fid), attr->cat_size);
1315 i_size_write(inode, attr->cat_size);
1316 inode->i_blocks = attr->cat_blocks;
1318 inode->i_mtime.tv_sec = mtime;
1319 inode->i_atime.tv_sec = atime;
1320 inode->i_ctime.tv_sec = ctime;
1323 ll_inode_size_unlock(inode);
1329 * Set designated mirror for I/O.
1331 * So far only read, write, and truncated can support to issue I/O to
1332 * designated mirror.
1334 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1336 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1338 /* clear layout version for generic(non-resync) I/O in case it carries
1339 * stale layout version due to I/O restart */
1340 io->ci_layout_version = 0;
1342 /* FLR: disable non-delay for designated mirror I/O because obviously
1343 * only one mirror is available */
1344 if (fd->fd_designated_mirror > 0) {
1346 io->ci_designated_mirror = fd->fd_designated_mirror;
1347 io->ci_layout_version = fd->fd_layout_version;
1350 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1351 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1354 static bool file_is_noatime(const struct file *file)
1356 const struct vfsmount *mnt = file->f_path.mnt;
1357 const struct inode *inode = file_inode((struct file *)file);
1359 /* Adapted from file_accessed() and touch_atime().*/
1360 if (file->f_flags & O_NOATIME)
1363 if (inode->i_flags & S_NOATIME)
1366 if (IS_NOATIME(inode))
1369 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1372 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1375 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1381 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot)
1383 struct inode *inode = file_inode(file);
1384 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1386 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1387 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1389 if (iot == CIT_WRITE) {
1390 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1391 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1392 file->f_flags & O_DIRECT ||
1395 io->ci_obj = ll_i2info(inode)->lli_clob;
1396 io->ci_lockreq = CILR_MAYBE;
1397 if (ll_file_nolock(file)) {
1398 io->ci_lockreq = CILR_NEVER;
1399 io->ci_no_srvlock = 1;
1400 } else if (file->f_flags & O_APPEND) {
1401 io->ci_lockreq = CILR_MANDATORY;
1403 io->ci_noatime = file_is_noatime(file);
1404 io->ci_async_readahead = false;
1406 /* FLR: only use non-delay I/O for read as there is only one
1407 * avaliable mirror for write. */
1408 io->ci_ndelay = !(iot == CIT_WRITE);
1410 ll_io_set_mirror(io, file);
1413 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1416 struct ll_inode_info *lli = ll_i2info(inode);
1417 struct ll_sb_info *sbi = ll_i2sbi(inode);
1418 enum obd_heat_type sample_type;
1419 enum obd_heat_type iobyte_type;
1420 __u64 now = ktime_get_real_seconds();
1422 if (!ll_sbi_has_file_heat(sbi) ||
1423 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1426 if (iot == CIT_READ) {
1427 sample_type = OBD_HEAT_READSAMPLE;
1428 iobyte_type = OBD_HEAT_READBYTE;
1429 } else if (iot == CIT_WRITE) {
1430 sample_type = OBD_HEAT_WRITESAMPLE;
1431 iobyte_type = OBD_HEAT_WRITEBYTE;
1436 spin_lock(&lli->lli_heat_lock);
1437 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1438 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1439 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1440 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1441 spin_unlock(&lli->lli_heat_lock);
1445 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1446 struct file *file, enum cl_io_type iot,
1447 loff_t *ppos, size_t count)
1449 struct vvp_io *vio = vvp_env_io(env);
1450 struct inode *inode = file_inode(file);
1451 struct ll_inode_info *lli = ll_i2info(inode);
1452 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1453 struct range_lock range;
1457 unsigned retried = 0;
1458 bool restarted = false;
1462 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1463 file_dentry(file)->d_name.name,
1464 iot == CIT_READ ? "read" : "write", *ppos, count);
1467 io = vvp_env_thread_io(env);
1468 ll_io_init(io, file, iot);
1469 io->ci_ndelay_tried = retried;
1471 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1472 bool range_locked = false;
1474 if (file->f_flags & O_APPEND)
1475 range_lock_init(&range, 0, LUSTRE_EOF);
1477 range_lock_init(&range, *ppos, *ppos + count - 1);
1479 vio->vui_fd = LUSTRE_FPRIVATE(file);
1480 vio->vui_io_subtype = args->via_io_subtype;
1482 switch (vio->vui_io_subtype) {
1484 vio->vui_iter = args->u.normal.via_iter;
1485 vio->vui_iocb = args->u.normal.via_iocb;
1486 /* Direct IO reads must also take range lock,
1487 * or multiple reads will try to work on the same pages
1488 * See LU-6227 for details. */
1489 if (((iot == CIT_WRITE) ||
1490 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1491 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1492 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1494 rc = range_lock(&lli->lli_write_tree, &range);
1498 range_locked = true;
1502 vio->u.splice.vui_pipe = args->u.splice.via_pipe;
1503 vio->u.splice.vui_flags = args->u.splice.via_flags;
1506 CERROR("unknown IO subtype %u\n", vio->vui_io_subtype);
1510 ll_cl_add(file, env, io, LCC_RW);
1511 rc = cl_io_loop(env, io);
1512 ll_cl_remove(file, env);
1515 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1517 range_unlock(&lli->lli_write_tree, &range);
1520 /* cl_io_rw_init() handled IO */
1524 if (io->ci_nob > 0) {
1525 result += io->ci_nob;
1526 count -= io->ci_nob;
1527 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1529 /* prepare IO restart */
1530 if (count > 0 && args->via_io_subtype == IO_NORMAL)
1531 args->u.normal.via_iter = vio->vui_iter;
1534 cl_io_fini(env, io);
1537 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1538 file->f_path.dentry->d_name.name,
1539 iot, rc, result, io->ci_need_restart);
1541 if ((rc == 0 || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
1543 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1544 file_dentry(file)->d_name.name,
1545 iot == CIT_READ ? "read" : "write",
1546 *ppos, count, result, rc);
1547 /* preserve the tried count for FLR */
1548 retried = io->ci_ndelay_tried;
1553 if (iot == CIT_READ) {
1555 ll_stats_ops_tally(ll_i2sbi(inode),
1556 LPROC_LL_READ_BYTES, result);
1557 } else if (iot == CIT_WRITE) {
1559 ll_stats_ops_tally(ll_i2sbi(inode),
1560 LPROC_LL_WRITE_BYTES, result);
1561 fd->fd_write_failed = false;
1562 } else if (result == 0 && rc == 0) {
1565 fd->fd_write_failed = true;
1567 fd->fd_write_failed = false;
1568 } else if (rc != -ERESTARTSYS) {
1569 fd->fd_write_failed = true;
1573 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1575 ll_heat_add(inode, iot, result);
1577 RETURN(result > 0 ? result : rc);
1581 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1582 * especially for small I/O.
1584 * To serve a read request, CLIO has to create and initialize a cl_io and
1585 * then request DLM lock. This has turned out to have siginificant overhead
1586 * and affects the performance of small I/O dramatically.
1588 * It's not necessary to create a cl_io for each I/O. Under the help of read
1589 * ahead, most of the pages being read are already in memory cache and we can
1590 * read those pages directly because if the pages exist, the corresponding DLM
1591 * lock must exist so that page content must be valid.
1593 * In fast read implementation, the llite speculatively finds and reads pages
1594 * in memory cache. There are three scenarios for fast read:
1595 * - If the page exists and is uptodate, kernel VM will provide the data and
1596 * CLIO won't be intervened;
1597 * - If the page was brought into memory by read ahead, it will be exported
1598 * and read ahead parameters will be updated;
1599 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1600 * it will go back and invoke normal read, i.e., a cl_io will be created
1601 * and DLM lock will be requested.
1603 * POSIX compliance: posix standard states that read is intended to be atomic.
1604 * Lustre read implementation is in line with Linux kernel read implementation
1605 * and neither of them complies with POSIX standard in this matter. Fast read
1606 * doesn't make the situation worse on single node but it may interleave write
1607 * results from multiple nodes due to short read handling in ll_file_aio_read().
1609 * \param env - lu_env
1610 * \param iocb - kiocb from kernel
1611 * \param iter - user space buffers where the data will be copied
1613 * \retval - number of bytes have been read, or error code if error occurred.
1616 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1620 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1623 /* NB: we can't do direct IO for fast read because it will need a lock
1624 * to make IO engine happy. */
1625 if (iocb->ki_filp->f_flags & O_DIRECT)
1628 result = generic_file_read_iter(iocb, iter);
1630 /* If the first page is not in cache, generic_file_aio_read() will be
1631 * returned with -ENODATA.
1632 * See corresponding code in ll_readpage(). */
1633 if (result == -ENODATA)
1637 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1638 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1639 LPROC_LL_READ_BYTES, result);
1646 * Read from a file (through the page cache).
1648 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1651 struct vvp_io_args *args;
1658 * Currently when PCC read failed, we do not fall back to the
1659 * normal read path, just return the error.
1660 * The resaon is that: for RW-PCC, the file data may be modified
1661 * in the PCC and inconsistent with the data on OSTs (or file
1662 * data has been removed from the Lustre file system), at this
1663 * time, fallback to the normal read path may read the wrong
1665 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1666 * path: read data from data copy on OSTs.
1668 result = pcc_file_read_iter(iocb, to, &cached);
1672 ll_ras_enter(iocb->ki_filp);
1674 result = ll_do_fast_read(iocb, to);
1675 if (result < 0 || iov_iter_count(to) == 0)
1678 env = cl_env_get(&refcheck);
1680 return PTR_ERR(env);
1682 args = ll_env_args(env, IO_NORMAL);
1683 args->u.normal.via_iter = to;
1684 args->u.normal.via_iocb = iocb;
1686 rc2 = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1687 &iocb->ki_pos, iov_iter_count(to));
1690 else if (result == 0)
1693 cl_env_put(env, &refcheck);
1699 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
1700 * If a page is already in the page cache and dirty (and some other things -
1701 * See ll_tiny_write_begin for the instantiation of these rules), then we can
1702 * write to it without doing a full I/O, because Lustre already knows about it
1703 * and will write it out. This saves a lot of processing time.
1705 * All writes here are within one page, so exclusion is handled by the page
1706 * lock on the vm page. We do not do tiny writes for writes which touch
1707 * multiple pages because it's very unlikely multiple sequential pages are
1708 * are already dirty.
1710 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
1711 * and are unlikely to be to already dirty pages.
1713 * Attribute updates are important here, we do them in ll_tiny_write_end.
1715 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
1717 ssize_t count = iov_iter_count(iter);
1718 struct file *file = iocb->ki_filp;
1719 struct inode *inode = file_inode(file);
1720 bool lock_inode = !IS_NOSEC(inode);
1725 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
1726 * of function for why.
1728 if (count >= PAGE_SIZE ||
1729 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
1732 if (unlikely(lock_inode))
1734 result = __generic_file_write_iter(iocb, iter);
1736 if (unlikely(lock_inode))
1737 inode_unlock(inode);
1739 /* If the page is not already dirty, ll_tiny_write_begin returns
1740 * -ENODATA. We continue on to normal write.
1742 if (result == -ENODATA)
1746 ll_heat_add(inode, CIT_WRITE, result);
1747 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
1749 ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED);
1752 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
1758 * Write to a file (through the page cache).
1760 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1762 struct vvp_io_args *args;
1764 ssize_t rc_tiny = 0, rc_normal;
1772 * When PCC write failed, we usually do not fall back to the normal
1773 * write path, just return the error. But there is a special case when
1774 * returned error code is -ENOSPC due to running out of space on PCC HSM
1775 * bakcend. At this time, it will fall back to normal I/O path and
1776 * retry the I/O. As the file is in HSM released state, it will restore
1777 * the file data to OSTs first and redo the write again. And the
1778 * restore process will revoke the layout lock and detach the file
1779 * from PCC cache automatically.
1781 result = pcc_file_write_iter(iocb, from, &cached);
1782 if (cached && result != -ENOSPC && result != -EDQUOT)
1785 /* NB: we can't do direct IO for tiny writes because they use the page
1786 * cache, we can't do sync writes because tiny writes can't flush
1787 * pages, and we can't do append writes because we can't guarantee the
1788 * required DLM locks are held to protect file size.
1790 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(iocb->ki_filp))) &&
1791 !(iocb->ki_filp->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
1792 rc_tiny = ll_do_tiny_write(iocb, from);
1794 /* In case of error, go on and try normal write - Only stop if tiny
1795 * write completed I/O.
1797 if (iov_iter_count(from) == 0)
1798 GOTO(out, rc_normal = rc_tiny);
1800 env = cl_env_get(&refcheck);
1802 return PTR_ERR(env);
1804 args = ll_env_args(env, IO_NORMAL);
1805 args->u.normal.via_iter = from;
1806 args->u.normal.via_iocb = iocb;
1808 rc_normal = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1809 &iocb->ki_pos, iov_iter_count(from));
1811 /* On success, combine bytes written. */
1812 if (rc_tiny >= 0 && rc_normal > 0)
1813 rc_normal += rc_tiny;
1814 /* On error, only return error from normal write if tiny write did not
1815 * write any bytes. Otherwise return bytes written by tiny write.
1817 else if (rc_tiny > 0)
1818 rc_normal = rc_tiny;
1820 cl_env_put(env, &refcheck);
1825 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1827 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1829 static int ll_file_get_iov_count(const struct iovec *iov,
1830 unsigned long *nr_segs, size_t *count)
1835 for (seg = 0; seg < *nr_segs; seg++) {
1836 const struct iovec *iv = &iov[seg];
1839 * If any segment has a negative length, or the cumulative
1840 * length ever wraps negative then return -EINVAL.
1843 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1845 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1850 cnt -= iv->iov_len; /* This segment is no good */
1857 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1858 unsigned long nr_segs, loff_t pos)
1865 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1869 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1870 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
1871 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1872 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
1873 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1875 result = ll_file_read_iter(iocb, &to);
1880 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
1883 struct iovec iov = { .iov_base = buf, .iov_len = count };
1888 init_sync_kiocb(&kiocb, file);
1889 kiocb.ki_pos = *ppos;
1890 #ifdef HAVE_KIOCB_KI_LEFT
1891 kiocb.ki_left = count;
1892 #elif defined(HAVE_KI_NBYTES)
1893 kiocb.i_nbytes = count;
1896 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
1897 *ppos = kiocb.ki_pos;
1903 * Write to a file (through the page cache).
1906 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1907 unsigned long nr_segs, loff_t pos)
1909 struct iov_iter from;
1914 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1918 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1919 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
1920 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1921 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
1922 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1924 result = ll_file_write_iter(iocb, &from);
1929 static ssize_t ll_file_write(struct file *file, const char __user *buf,
1930 size_t count, loff_t *ppos)
1932 struct iovec iov = { .iov_base = (void __user *)buf,
1939 init_sync_kiocb(&kiocb, file);
1940 kiocb.ki_pos = *ppos;
1941 #ifdef HAVE_KIOCB_KI_LEFT
1942 kiocb.ki_left = count;
1943 #elif defined(HAVE_KI_NBYTES)
1944 kiocb.ki_nbytes = count;
1947 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
1948 *ppos = kiocb.ki_pos;
1952 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
1955 * Send file content (through pagecache) somewhere with helper
1957 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1958 struct pipe_inode_info *pipe, size_t count,
1962 struct vvp_io_args *args;
1969 result = pcc_file_splice_read(in_file, ppos, pipe,
1970 count, flags, &cached);
1974 ll_ras_enter(in_file);
1976 env = cl_env_get(&refcheck);
1978 RETURN(PTR_ERR(env));
1980 args = ll_env_args(env, IO_SPLICE);
1981 args->u.splice.via_pipe = pipe;
1982 args->u.splice.via_flags = flags;
1984 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1985 cl_env_put(env, &refcheck);
1989 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1990 __u64 flags, struct lov_user_md *lum, int lum_size)
1992 struct lookup_intent oit = {
1994 .it_flags = flags | MDS_OPEN_BY_FID,
1999 ll_inode_size_lock(inode);
2000 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2002 GOTO(out_unlock, rc);
2004 ll_release_openhandle(dentry, &oit);
2007 ll_inode_size_unlock(inode);
2008 ll_intent_release(&oit);
2013 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2014 struct lov_mds_md **lmmp, int *lmm_size,
2015 struct ptlrpc_request **request)
2017 struct ll_sb_info *sbi = ll_i2sbi(inode);
2018 struct mdt_body *body;
2019 struct lov_mds_md *lmm = NULL;
2020 struct ptlrpc_request *req = NULL;
2021 struct md_op_data *op_data;
2024 rc = ll_get_default_mdsize(sbi, &lmmsize);
2028 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2029 strlen(filename), lmmsize,
2030 LUSTRE_OPC_ANY, NULL);
2031 if (IS_ERR(op_data))
2032 RETURN(PTR_ERR(op_data));
2034 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2035 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2036 ll_finish_md_op_data(op_data);
2038 CDEBUG(D_INFO, "md_getattr_name failed "
2039 "on %s: rc %d\n", filename, rc);
2043 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2044 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2046 lmmsize = body->mbo_eadatasize;
2048 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2050 GOTO(out, rc = -ENODATA);
2053 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2054 LASSERT(lmm != NULL);
2056 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2057 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2058 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2059 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2060 GOTO(out, rc = -EPROTO);
2063 * This is coming from the MDS, so is probably in
2064 * little endian. We convert it to host endian before
2065 * passing it to userspace.
2067 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
2070 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2071 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2072 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2073 if (le32_to_cpu(lmm->lmm_pattern) &
2074 LOV_PATTERN_F_RELEASED)
2078 /* if function called for directory - we should
2079 * avoid swab not existent lsm objects */
2080 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
2081 lustre_swab_lov_user_md_v1(
2082 (struct lov_user_md_v1 *)lmm);
2083 if (S_ISREG(body->mbo_mode))
2084 lustre_swab_lov_user_md_objects(
2085 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2087 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2088 lustre_swab_lov_user_md_v3(
2089 (struct lov_user_md_v3 *)lmm);
2090 if (S_ISREG(body->mbo_mode))
2091 lustre_swab_lov_user_md_objects(
2092 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2094 } else if (lmm->lmm_magic ==
2095 cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2096 lustre_swab_lov_comp_md_v1(
2097 (struct lov_comp_md_v1 *)lmm);
2098 } else if (lmm->lmm_magic ==
2099 cpu_to_le32(LOV_MAGIC_FOREIGN)) {
2100 struct lov_foreign_md *lfm;
2102 lfm = (struct lov_foreign_md *)lmm;
2103 __swab32s(&lfm->lfm_magic);
2104 __swab32s(&lfm->lfm_length);
2105 __swab32s(&lfm->lfm_type);
2106 __swab32s(&lfm->lfm_flags);
2112 *lmm_size = lmmsize;
2117 static int ll_lov_setea(struct inode *inode, struct file *file,
2120 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2121 struct lov_user_md *lump;
2122 int lum_size = sizeof(struct lov_user_md) +
2123 sizeof(struct lov_user_ost_data);
2127 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2130 OBD_ALLOC_LARGE(lump, lum_size);
2134 if (copy_from_user(lump, arg, lum_size))
2135 GOTO(out_lump, rc = -EFAULT);
2137 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2139 cl_lov_delay_create_clear(&file->f_flags);
2142 OBD_FREE_LARGE(lump, lum_size);
2146 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2153 env = cl_env_get(&refcheck);
2155 RETURN(PTR_ERR(env));
2157 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2158 cl_env_put(env, &refcheck);
2162 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2165 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2166 struct lov_user_md *klum;
2168 __u64 flags = FMODE_WRITE;
2171 rc = ll_copy_user_md(lum, &klum);
2176 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2181 rc = put_user(0, &lum->lmm_stripe_count);
2185 rc = ll_layout_refresh(inode, &gen);
2189 rc = ll_file_getstripe(inode, arg, lum_size);
2191 cl_lov_delay_create_clear(&file->f_flags);
2194 OBD_FREE(klum, lum_size);
2199 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2201 struct ll_inode_info *lli = ll_i2info(inode);
2202 struct cl_object *obj = lli->lli_clob;
2203 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2204 struct ll_grouplock grouplock;
2209 CWARN("group id for group lock must not be 0\n");
2213 if (ll_file_nolock(file))
2214 RETURN(-EOPNOTSUPP);
2216 spin_lock(&lli->lli_lock);
2217 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2218 CWARN("group lock already existed with gid %lu\n",
2219 fd->fd_grouplock.lg_gid);
2220 spin_unlock(&lli->lli_lock);
2223 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2224 spin_unlock(&lli->lli_lock);
2227 * XXX: group lock needs to protect all OST objects while PFL
2228 * can add new OST objects during the IO, so we'd instantiate
2229 * all OST objects before getting its group lock.
2234 struct cl_layout cl = {
2235 .cl_is_composite = false,
2237 struct lu_extent ext = {
2239 .e_end = OBD_OBJECT_EOF,
2242 env = cl_env_get(&refcheck);
2244 RETURN(PTR_ERR(env));
2246 rc = cl_object_layout_get(env, obj, &cl);
2247 if (!rc && cl.cl_is_composite)
2248 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2251 cl_env_put(env, &refcheck);
2256 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2257 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2261 spin_lock(&lli->lli_lock);
2262 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2263 spin_unlock(&lli->lli_lock);
2264 CERROR("another thread just won the race\n");
2265 cl_put_grouplock(&grouplock);
2269 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2270 fd->fd_grouplock = grouplock;
2271 spin_unlock(&lli->lli_lock);
2273 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2277 static int ll_put_grouplock(struct inode *inode, struct file *file,
2280 struct ll_inode_info *lli = ll_i2info(inode);
2281 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2282 struct ll_grouplock grouplock;
2285 spin_lock(&lli->lli_lock);
2286 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2287 spin_unlock(&lli->lli_lock);
2288 CWARN("no group lock held\n");
2292 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2294 if (fd->fd_grouplock.lg_gid != arg) {
2295 CWARN("group lock %lu doesn't match current id %lu\n",
2296 arg, fd->fd_grouplock.lg_gid);
2297 spin_unlock(&lli->lli_lock);
2301 grouplock = fd->fd_grouplock;
2302 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2303 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2304 spin_unlock(&lli->lli_lock);
2306 cl_put_grouplock(&grouplock);
2307 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2312 * Close inode open handle
2314 * \param dentry [in] dentry which contains the inode
2315 * \param it [in,out] intent which contains open info and result
2318 * \retval <0 failure
2320 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2322 struct inode *inode = dentry->d_inode;
2323 struct obd_client_handle *och;
2329 /* Root ? Do nothing. */
2330 if (dentry->d_inode->i_sb->s_root == dentry)
2333 /* No open handle to close? Move away */
2334 if (!it_disposition(it, DISP_OPEN_OPEN))
2337 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2339 OBD_ALLOC(och, sizeof(*och));
2341 GOTO(out, rc = -ENOMEM);
2343 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2345 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2347 /* this one is in place of ll_file_open */
2348 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2349 ptlrpc_req_finished(it->it_request);
2350 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2356 * Get size for inode for which FIEMAP mapping is requested.
2357 * Make the FIEMAP get_info call and returns the result.
2358 * \param fiemap kernel buffer to hold extens
2359 * \param num_bytes kernel buffer size
2361 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2367 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2370 /* Checks for fiemap flags */
2371 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2372 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2376 /* Check for FIEMAP_FLAG_SYNC */
2377 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2378 rc = filemap_fdatawrite(inode->i_mapping);
2383 env = cl_env_get(&refcheck);
2385 RETURN(PTR_ERR(env));
2387 if (i_size_read(inode) == 0) {
2388 rc = ll_glimpse_size(inode);
2393 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2394 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2395 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2397 /* If filesize is 0, then there would be no objects for mapping */
2398 if (fmkey.lfik_oa.o_size == 0) {
2399 fiemap->fm_mapped_extents = 0;
2403 fmkey.lfik_fiemap = *fiemap;
2405 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2406 &fmkey, fiemap, &num_bytes);
2408 cl_env_put(env, &refcheck);
2412 int ll_fid2path(struct inode *inode, void __user *arg)
2414 struct obd_export *exp = ll_i2mdexp(inode);
2415 const struct getinfo_fid2path __user *gfin = arg;
2417 struct getinfo_fid2path *gfout;
2423 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2424 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2427 /* Only need to get the buflen */
2428 if (get_user(pathlen, &gfin->gf_pathlen))
2431 if (pathlen > PATH_MAX)
2434 outsize = sizeof(*gfout) + pathlen;
2435 OBD_ALLOC(gfout, outsize);
2439 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2440 GOTO(gf_free, rc = -EFAULT);
2441 /* append root FID after gfout to let MDT know the root FID so that it
2442 * can lookup the correct path, this is mainly for fileset.
2443 * old server without fileset mount support will ignore this. */
2444 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2446 /* Call mdc_iocontrol */
2447 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2451 if (copy_to_user(arg, gfout, outsize))
2455 OBD_FREE(gfout, outsize);
2460 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2462 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2470 ioc->idv_version = 0;
2471 ioc->idv_layout_version = UINT_MAX;
2473 /* If no file object initialized, we consider its version is 0. */
2477 env = cl_env_get(&refcheck);
2479 RETURN(PTR_ERR(env));
2481 io = vvp_env_thread_io(env);
2483 io->u.ci_data_version.dv_data_version = 0;
2484 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2485 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2488 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2489 result = cl_io_loop(env, io);
2491 result = io->ci_result;
2493 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2494 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2496 cl_io_fini(env, io);
2498 if (unlikely(io->ci_need_restart))
2501 cl_env_put(env, &refcheck);
2507 * Read the data_version for inode.
2509 * This value is computed using stripe object version on OST.
2510 * Version is computed using server side locking.
2512 * @param flags if do sync on the OST side;
2514 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2515 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2517 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2519 struct ioc_data_version ioc = { .idv_flags = flags };
2522 rc = ll_ioc_data_version(inode, &ioc);
2524 *data_version = ioc.idv_version;
2530 * Trigger a HSM release request for the provided inode.
2532 int ll_hsm_release(struct inode *inode)
2535 struct obd_client_handle *och = NULL;
2536 __u64 data_version = 0;
2541 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2542 ll_i2sbi(inode)->ll_fsname,
2543 PFID(&ll_i2info(inode)->lli_fid));
2545 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2547 GOTO(out, rc = PTR_ERR(och));
2549 /* Grab latest data_version and [am]time values */
2550 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2554 env = cl_env_get(&refcheck);
2556 GOTO(out, rc = PTR_ERR(env));
2558 rc = ll_merge_attr(env, inode);
2559 cl_env_put(env, &refcheck);
2561 /* If error happen, we have the wrong size for a file.
2567 /* Release the file.
2568 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2569 * we still need it to pack l_remote_handle to MDT. */
2570 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2576 if (och != NULL && !IS_ERR(och)) /* close the file */
2577 ll_lease_close(och, inode, NULL);
2582 struct ll_swap_stack {
2585 struct inode *inode1;
2586 struct inode *inode2;
2591 static int ll_swap_layouts(struct file *file1, struct file *file2,
2592 struct lustre_swap_layouts *lsl)
2594 struct mdc_swap_layouts msl;
2595 struct md_op_data *op_data;
2598 struct ll_swap_stack *llss = NULL;
2601 OBD_ALLOC_PTR(llss);
2605 llss->inode1 = file_inode(file1);
2606 llss->inode2 = file_inode(file2);
2608 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2612 /* we use 2 bool because it is easier to swap than 2 bits */
2613 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2614 llss->check_dv1 = true;
2616 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2617 llss->check_dv2 = true;
2619 /* we cannot use lsl->sl_dvX directly because we may swap them */
2620 llss->dv1 = lsl->sl_dv1;
2621 llss->dv2 = lsl->sl_dv2;
2623 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2624 if (rc == 0) /* same file, done! */
2627 if (rc < 0) { /* sequentialize it */
2628 swap(llss->inode1, llss->inode2);
2630 swap(llss->dv1, llss->dv2);
2631 swap(llss->check_dv1, llss->check_dv2);
2635 if (gid != 0) { /* application asks to flush dirty cache */
2636 rc = ll_get_grouplock(llss->inode1, file1, gid);
2640 rc = ll_get_grouplock(llss->inode2, file2, gid);
2642 ll_put_grouplock(llss->inode1, file1, gid);
2647 /* ultimate check, before swaping the layouts we check if
2648 * dataversion has changed (if requested) */
2649 if (llss->check_dv1) {
2650 rc = ll_data_version(llss->inode1, &dv, 0);
2653 if (dv != llss->dv1)
2654 GOTO(putgl, rc = -EAGAIN);
2657 if (llss->check_dv2) {
2658 rc = ll_data_version(llss->inode2, &dv, 0);
2661 if (dv != llss->dv2)
2662 GOTO(putgl, rc = -EAGAIN);
2665 /* struct md_op_data is used to send the swap args to the mdt
2666 * only flags is missing, so we use struct mdc_swap_layouts
2667 * through the md_op_data->op_data */
2668 /* flags from user space have to be converted before they are send to
2669 * server, no flag is sent today, they are only used on the client */
2672 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2673 0, LUSTRE_OPC_ANY, &msl);
2674 if (IS_ERR(op_data))
2675 GOTO(free, rc = PTR_ERR(op_data));
2677 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2678 sizeof(*op_data), op_data, NULL);
2679 ll_finish_md_op_data(op_data);
2686 ll_put_grouplock(llss->inode2, file2, gid);
2687 ll_put_grouplock(llss->inode1, file1, gid);
2697 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2699 struct obd_export *exp = ll_i2mdexp(inode);
2700 struct md_op_data *op_data;
2704 /* Detect out-of range masks */
2705 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2708 /* Non-root users are forbidden to set or clear flags which are
2709 * NOT defined in HSM_USER_MASK. */
2710 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2711 !cfs_capable(CFS_CAP_SYS_ADMIN))
2714 if (!exp_connect_archive_id_array(exp)) {
2715 /* Detect out-of range archive id */
2716 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2717 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
2721 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2722 LUSTRE_OPC_ANY, hss);
2723 if (IS_ERR(op_data))
2724 RETURN(PTR_ERR(op_data));
2726 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
2729 ll_finish_md_op_data(op_data);
2734 static int ll_hsm_import(struct inode *inode, struct file *file,
2735 struct hsm_user_import *hui)
2737 struct hsm_state_set *hss = NULL;
2738 struct iattr *attr = NULL;
2742 if (!S_ISREG(inode->i_mode))
2748 GOTO(out, rc = -ENOMEM);
2750 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2751 hss->hss_archive_id = hui->hui_archive_id;
2752 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2753 rc = ll_hsm_state_set(inode, hss);
2757 OBD_ALLOC_PTR(attr);
2759 GOTO(out, rc = -ENOMEM);
2761 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2762 attr->ia_mode |= S_IFREG;
2763 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2764 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2765 attr->ia_size = hui->hui_size;
2766 attr->ia_mtime.tv_sec = hui->hui_mtime;
2767 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2768 attr->ia_atime.tv_sec = hui->hui_atime;
2769 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2771 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2772 ATTR_UID | ATTR_GID |
2773 ATTR_MTIME | ATTR_MTIME_SET |
2774 ATTR_ATIME | ATTR_ATIME_SET;
2778 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
2782 inode_unlock(inode);
2794 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2796 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2797 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2800 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
2802 struct inode *inode = file_inode(file);
2804 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
2805 ATTR_MTIME | ATTR_MTIME_SET |
2808 .tv_sec = lfu->lfu_atime_sec,
2809 .tv_nsec = lfu->lfu_atime_nsec,
2812 .tv_sec = lfu->lfu_mtime_sec,
2813 .tv_nsec = lfu->lfu_mtime_nsec,
2816 .tv_sec = lfu->lfu_ctime_sec,
2817 .tv_nsec = lfu->lfu_ctime_nsec,
2823 if (!capable(CAP_SYS_ADMIN))
2826 if (!S_ISREG(inode->i_mode))
2830 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
2832 inode_unlock(inode);
2837 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
2840 case MODE_READ_USER:
2842 case MODE_WRITE_USER:
2849 static const char *const user_lockname[] = LOCK_MODE_NAMES;
2851 /* Used to allow the upper layers of the client to request an LDLM lock
2852 * without doing an actual read or write.
2854 * Used for ladvise lockahead to manually request specific locks.
2856 * \param[in] file file this ladvise lock request is on
2857 * \param[in] ladvise ladvise struct describing this lock request
2859 * \retval 0 success, no detailed result available (sync requests
2860 * and requests sent to the server [not handled locally]
2861 * cannot return detailed results)
2862 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
2863 * see definitions for details.
2864 * \retval negative negative errno on error
2866 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
2868 struct lu_env *env = NULL;
2869 struct cl_io *io = NULL;
2870 struct cl_lock *lock = NULL;
2871 struct cl_lock_descr *descr = NULL;
2872 struct dentry *dentry = file->f_path.dentry;
2873 struct inode *inode = dentry->d_inode;
2874 enum cl_lock_mode cl_mode;
2875 off_t start = ladvise->lla_start;
2876 off_t end = ladvise->lla_end;
2882 CDEBUG(D_VFSTRACE, "Lock request: file=%.*s, inode=%p, mode=%s "
2883 "start=%llu, end=%llu\n", dentry->d_name.len,
2884 dentry->d_name.name, dentry->d_inode,
2885 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
2888 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
2890 GOTO(out, result = cl_mode);
2892 /* Get IO environment */
2893 result = cl_io_get(inode, &env, &io, &refcheck);
2897 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
2900 * nothing to do for this io. This currently happens when
2901 * stripe sub-object's are not yet created.
2903 result = io->ci_result;
2904 } else if (result == 0) {
2905 lock = vvp_env_lock(env);
2906 descr = &lock->cll_descr;
2908 descr->cld_obj = io->ci_obj;
2909 /* Convert byte offsets to pages */
2910 descr->cld_start = cl_index(io->ci_obj, start);
2911 descr->cld_end = cl_index(io->ci_obj, end);
2912 descr->cld_mode = cl_mode;
2913 /* CEF_MUST is used because we do not want to convert a
2914 * lockahead request to a lockless lock */
2915 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND |
2918 if (ladvise->lla_peradvice_flags & LF_ASYNC)
2919 descr->cld_enq_flags |= CEF_SPECULATIVE;
2921 result = cl_lock_request(env, io, lock);
2923 /* On success, we need to release the lock */
2925 cl_lock_release(env, lock);
2927 cl_io_fini(env, io);
2928 cl_env_put(env, &refcheck);
2930 /* -ECANCELED indicates a matching lock with a different extent
2931 * was already present, and -EEXIST indicates a matching lock
2932 * on exactly the same extent was already present.
2933 * We convert them to positive values for userspace to make
2934 * recognizing true errors easier.
2935 * Note we can only return these detailed results on async requests,
2936 * as sync requests look the same as i/o requests for locking. */
2937 if (result == -ECANCELED)
2938 result = LLA_RESULT_DIFFERENT;
2939 else if (result == -EEXIST)
2940 result = LLA_RESULT_SAME;
2945 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
2947 static int ll_ladvise_sanity(struct inode *inode,
2948 struct llapi_lu_ladvise *ladvise)
2950 struct ll_sb_info *sbi = ll_i2sbi(inode);
2951 enum lu_ladvise_type advice = ladvise->lla_advice;
2952 /* Note the peradvice flags is a 32 bit field, so per advice flags must
2953 * be in the first 32 bits of enum ladvise_flags */
2954 __u32 flags = ladvise->lla_peradvice_flags;
2955 /* 3 lines at 80 characters per line, should be plenty */
2958 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
2960 CDEBUG(D_VFSTRACE, "%s: advice with value '%d' not recognized,"
2961 "last supported advice is %s (value '%d'): rc = %d\n",
2962 sbi->ll_fsname, advice,
2963 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
2967 /* Per-advice checks */
2969 case LU_LADVISE_LOCKNOEXPAND:
2970 if (flags & ~LF_LOCKNOEXPAND_MASK) {
2972 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
2973 "rc = %d\n", sbi->ll_fsname, flags,
2974 ladvise_names[advice], rc);
2978 case LU_LADVISE_LOCKAHEAD:
2979 /* Currently only READ and WRITE modes can be requested */
2980 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
2981 ladvise->lla_lockahead_mode == 0) {
2983 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
2984 "rc = %d\n", sbi->ll_fsname,
2985 ladvise->lla_lockahead_mode,
2986 ladvise_names[advice], rc);
2989 case LU_LADVISE_WILLREAD:
2990 case LU_LADVISE_DONTNEED:
2992 /* Note fall through above - These checks apply to all advices
2993 * except LOCKNOEXPAND */
2994 if (flags & ~LF_DEFAULT_MASK) {
2996 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
2997 "rc = %d\n", sbi->ll_fsname, flags,
2998 ladvise_names[advice], rc);
3001 if (ladvise->lla_start >= ladvise->lla_end) {
3003 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3004 "for %s: rc = %d\n", sbi->ll_fsname,
3005 ladvise->lla_start, ladvise->lla_end,
3006 ladvise_names[advice], rc);
3018 * Give file access advices
3020 * The ladvise interface is similar to Linux fadvise() system call, except it
3021 * forwards the advices directly from Lustre client to server. The server side
3022 * codes will apply appropriate read-ahead and caching techniques for the
3023 * corresponding files.
3025 * A typical workload for ladvise is e.g. a bunch of different clients are
3026 * doing small random reads of a file, so prefetching pages into OSS cache
3027 * with big linear reads before the random IO is a net benefit. Fetching
3028 * all that data into each client cache with fadvise() may not be, due to
3029 * much more data being sent to the client.
3031 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3032 struct llapi_lu_ladvise *ladvise)
3036 struct cl_ladvise_io *lio;
3041 env = cl_env_get(&refcheck);
3043 RETURN(PTR_ERR(env));
3045 io = vvp_env_thread_io(env);
3046 io->ci_obj = ll_i2info(inode)->lli_clob;
3048 /* initialize parameters for ladvise */
3049 lio = &io->u.ci_ladvise;
3050 lio->li_start = ladvise->lla_start;
3051 lio->li_end = ladvise->lla_end;
3052 lio->li_fid = ll_inode2fid(inode);
3053 lio->li_advice = ladvise->lla_advice;
3054 lio->li_flags = flags;
3056 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3057 rc = cl_io_loop(env, io);
3061 cl_io_fini(env, io);
3062 cl_env_put(env, &refcheck);
3066 static int ll_lock_noexpand(struct file *file, int flags)
3068 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3070 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3075 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3078 struct fsxattr fsxattr;
3080 if (copy_from_user(&fsxattr,
3081 (const struct fsxattr __user *)arg,
3085 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3086 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
3087 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3088 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3089 if (copy_to_user((struct fsxattr __user *)arg,
3090 &fsxattr, sizeof(fsxattr)))
3096 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa)
3099 * Project Quota ID state is only allowed to change from within the init
3100 * namespace. Enforce that restriction only if we are trying to change
3101 * the quota ID state. Everything else is allowed in user namespaces.
3103 if (current_user_ns() == &init_user_ns)
3106 if (ll_i2info(inode)->lli_projid != fa->fsx_projid)
3109 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT)) {
3110 if (!(fa->fsx_xflags & FS_XFLAG_PROJINHERIT))
3113 if (fa->fsx_xflags & FS_XFLAG_PROJINHERIT)
3120 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3124 struct md_op_data *op_data;
3125 struct ptlrpc_request *req = NULL;
3127 struct fsxattr fsxattr;
3128 struct cl_object *obj;
3132 if (copy_from_user(&fsxattr,
3133 (const struct fsxattr __user *)arg,
3137 rc = ll_ioctl_check_project(inode, &fsxattr);
3141 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3142 LUSTRE_OPC_ANY, NULL);
3143 if (IS_ERR(op_data))
3144 RETURN(PTR_ERR(op_data));
3146 flags = ll_xflags_to_inode_flags(fsxattr.fsx_xflags);
3147 op_data->op_attr_flags = ll_inode_to_ext_flags(flags);
3148 if (fsxattr.fsx_xflags & FS_XFLAG_PROJINHERIT)
3149 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3150 op_data->op_projid = fsxattr.fsx_projid;
3151 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3152 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL,
3154 ptlrpc_req_finished(req);
3156 GOTO(out_fsxattr, rc);
3157 ll_update_inode_flags(inode, op_data->op_attr_flags);
3158 obj = ll_i2info(inode)->lli_clob;
3160 GOTO(out_fsxattr, rc);
3162 OBD_ALLOC_PTR(attr);
3164 GOTO(out_fsxattr, rc = -ENOMEM);
3166 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS,
3167 fsxattr.fsx_xflags);
3170 ll_finish_md_op_data(op_data);
3174 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3177 struct inode *inode = file_inode(file);
3178 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3179 struct ll_inode_info *lli = ll_i2info(inode);
3180 struct obd_client_handle *och = NULL;
3181 struct split_param sp;
3182 struct pcc_param param;
3183 bool lease_broken = false;
3185 enum mds_op_bias bias = 0;
3186 struct file *layout_file = NULL;
3188 size_t data_size = 0;
3189 bool attached = false;
3194 mutex_lock(&lli->lli_och_mutex);
3195 if (fd->fd_lease_och != NULL) {
3196 och = fd->fd_lease_och;
3197 fd->fd_lease_och = NULL;
3199 mutex_unlock(&lli->lli_och_mutex);
3204 fmode = och->och_flags;
3206 switch (ioc->lil_flags) {
3207 case LL_LEASE_RESYNC_DONE:
3208 if (ioc->lil_count > IOC_IDS_MAX)
3209 GOTO(out_lease_close, rc = -EINVAL);
3211 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3212 OBD_ALLOC(data, data_size);
3214 GOTO(out_lease_close, rc = -ENOMEM);
3216 if (copy_from_user(data, (void __user *)arg, data_size))
3217 GOTO(out_lease_close, rc = -EFAULT);
3219 bias = MDS_CLOSE_RESYNC_DONE;
3221 case LL_LEASE_LAYOUT_MERGE: {
3224 if (ioc->lil_count != 1)
3225 GOTO(out_lease_close, rc = -EINVAL);
3227 arg += sizeof(*ioc);
3228 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3229 GOTO(out_lease_close, rc = -EFAULT);
3231 layout_file = fget(fd);
3233 GOTO(out_lease_close, rc = -EBADF);
3235 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3236 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3237 GOTO(out_lease_close, rc = -EPERM);
3239 data = file_inode(layout_file);
3240 bias = MDS_CLOSE_LAYOUT_MERGE;
3243 case LL_LEASE_LAYOUT_SPLIT: {
3247 if (ioc->lil_count != 2)
3248 GOTO(out_lease_close, rc = -EINVAL);
3250 arg += sizeof(*ioc);
3251 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3252 GOTO(out_lease_close, rc = -EFAULT);
3254 arg += sizeof(__u32);
3255 if (copy_from_user(&mirror_id, (void __user *)arg,
3257 GOTO(out_lease_close, rc = -EFAULT);
3259 layout_file = fget(fdv);
3261 GOTO(out_lease_close, rc = -EBADF);
3263 sp.sp_inode = file_inode(layout_file);
3264 sp.sp_mirror_id = (__u16)mirror_id;
3266 bias = MDS_CLOSE_LAYOUT_SPLIT;
3269 case LL_LEASE_PCC_ATTACH:
3270 if (ioc->lil_count != 1)
3273 arg += sizeof(*ioc);
3274 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3276 GOTO(out_lease_close, rc2 = -EFAULT);
3278 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3280 GOTO(out_lease_close, rc2);
3283 /* Grab latest data version */
3284 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3287 GOTO(out_lease_close, rc2);
3290 bias = MDS_PCC_ATTACH;
3293 /* without close intent */
3298 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3302 rc = ll_lease_och_release(inode, file);
3311 switch (ioc->lil_flags) {
3312 case LL_LEASE_RESYNC_DONE:
3314 OBD_FREE(data, data_size);
3316 case LL_LEASE_LAYOUT_MERGE:
3317 case LL_LEASE_LAYOUT_SPLIT:
3321 case LL_LEASE_PCC_ATTACH:
3324 rc = pcc_readwrite_attach_fini(file, inode,
3325 param.pa_layout_gen,
3332 rc = ll_lease_type_from_fmode(fmode);
3336 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3339 struct inode *inode = file_inode(file);
3340 struct ll_inode_info *lli = ll_i2info(inode);
3341 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3342 struct obd_client_handle *och = NULL;
3343 __u64 open_flags = 0;
3349 switch (ioc->lil_mode) {
3350 case LL_LEASE_WRLCK:
3351 if (!(file->f_mode & FMODE_WRITE))
3353 fmode = FMODE_WRITE;
3355 case LL_LEASE_RDLCK:
3356 if (!(file->f_mode & FMODE_READ))
3360 case LL_LEASE_UNLCK:
3361 RETURN(ll_file_unlock_lease(file, ioc, arg));
3366 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3368 /* apply for lease */
3369 if (ioc->lil_flags & LL_LEASE_RESYNC)
3370 open_flags = MDS_OPEN_RESYNC;
3371 och = ll_lease_open(inode, file, fmode, open_flags);
3373 RETURN(PTR_ERR(och));
3375 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3376 rc = ll_lease_file_resync(och, inode, arg);
3378 ll_lease_close(och, inode, NULL);
3381 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3383 ll_lease_close(och, inode, NULL);
3389 mutex_lock(&lli->lli_och_mutex);
3390 if (fd->fd_lease_och == NULL) {
3391 fd->fd_lease_och = och;
3394 mutex_unlock(&lli->lli_och_mutex);
3396 /* impossible now that only excl is supported for now */
3397 ll_lease_close(och, inode, &lease_broken);
3403 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3405 struct ll_inode_info *lli = ll_i2info(inode);
3406 struct ll_sb_info *sbi = ll_i2sbi(inode);
3407 __u64 now = ktime_get_real_seconds();
3410 spin_lock(&lli->lli_heat_lock);
3411 heat->lh_flags = lli->lli_heat_flags;
3412 for (i = 0; i < heat->lh_count; i++)
3413 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3414 now, sbi->ll_heat_decay_weight,
3415 sbi->ll_heat_period_second);
3416 spin_unlock(&lli->lli_heat_lock);
3419 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3421 struct ll_inode_info *lli = ll_i2info(inode);
3424 spin_lock(&lli->lli_heat_lock);
3425 if (flags & LU_HEAT_FLAG_CLEAR)
3426 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3428 if (flags & LU_HEAT_FLAG_OFF)
3429 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3431 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3433 spin_unlock(&lli->lli_heat_lock);
3439 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3441 struct inode *inode = file_inode(file);
3442 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3446 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3447 PFID(ll_inode2fid(inode)), inode, cmd);
3448 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3450 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3451 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3455 case LL_IOC_GETFLAGS:
3456 /* Get the current value of the file flags */
3457 return put_user(fd->fd_flags, (int __user *)arg);
3458 case LL_IOC_SETFLAGS:
3459 case LL_IOC_CLRFLAGS:
3460 /* Set or clear specific file flags */
3461 /* XXX This probably needs checks to ensure the flags are
3462 * not abused, and to handle any flag side effects.
3464 if (get_user(flags, (int __user *) arg))
3467 if (cmd == LL_IOC_SETFLAGS) {
3468 if ((flags & LL_FILE_IGNORE_LOCK) &&
3469 !(file->f_flags & O_DIRECT)) {
3470 CERROR("%s: unable to disable locking on "
3471 "non-O_DIRECT file\n", current->comm);
3475 fd->fd_flags |= flags;
3477 fd->fd_flags &= ~flags;
3480 case LL_IOC_LOV_SETSTRIPE:
3481 case LL_IOC_LOV_SETSTRIPE_NEW:
3482 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3483 case LL_IOC_LOV_SETEA:
3484 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3485 case LL_IOC_LOV_SWAP_LAYOUTS: {
3487 struct lustre_swap_layouts lsl;
3489 if (copy_from_user(&lsl, (char __user *)arg,
3490 sizeof(struct lustre_swap_layouts)))
3493 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3496 file2 = fget(lsl.sl_fd);
3500 /* O_WRONLY or O_RDWR */
3501 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3502 GOTO(out, rc = -EPERM);
3504 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3505 struct inode *inode2;
3506 struct ll_inode_info *lli;
3507 struct obd_client_handle *och = NULL;
3509 lli = ll_i2info(inode);
3510 mutex_lock(&lli->lli_och_mutex);
3511 if (fd->fd_lease_och != NULL) {
3512 och = fd->fd_lease_och;
3513 fd->fd_lease_och = NULL;
3515 mutex_unlock(&lli->lli_och_mutex);
3517 GOTO(out, rc = -ENOLCK);
3518 inode2 = file_inode(file2);
3519 rc = ll_swap_layouts_close(och, inode, inode2);
3521 rc = ll_swap_layouts(file, file2, &lsl);
3527 case LL_IOC_LOV_GETSTRIPE:
3528 case LL_IOC_LOV_GETSTRIPE_NEW:
3529 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3530 case FS_IOC_GETFLAGS:
3531 case FS_IOC_SETFLAGS:
3532 RETURN(ll_iocontrol(inode, file, cmd, arg));
3533 case FSFILT_IOC_GETVERSION:
3534 case FS_IOC_GETVERSION:
3535 RETURN(put_user(inode->i_generation, (int __user *)arg));
3536 /* We need to special case any other ioctls we want to handle,
3537 * to send them to the MDS/OST as appropriate and to properly
3538 * network encode the arg field. */
3539 case FS_IOC_SETVERSION:
3542 case LL_IOC_GROUP_LOCK:
3543 RETURN(ll_get_grouplock(inode, file, arg));
3544 case LL_IOC_GROUP_UNLOCK:
3545 RETURN(ll_put_grouplock(inode, file, arg));
3546 case IOC_OBD_STATFS:
3547 RETURN(ll_obd_statfs(inode, (void __user *)arg));
3549 case LL_IOC_FLUSHCTX:
3550 RETURN(ll_flush_ctx(inode));
3551 case LL_IOC_PATH2FID: {
3552 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
3553 sizeof(struct lu_fid)))
3558 case LL_IOC_GETPARENT:
3559 RETURN(ll_getparent(file, (struct getparent __user *)arg));
3561 case OBD_IOC_FID2PATH:
3562 RETURN(ll_fid2path(inode, (void __user *)arg));
3563 case LL_IOC_DATA_VERSION: {
3564 struct ioc_data_version idv;
3567 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
3570 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
3571 rc = ll_ioc_data_version(inode, &idv);
3574 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
3580 case LL_IOC_GET_MDTIDX: {
3583 mdtidx = ll_get_mdt_idx(inode);
3587 if (put_user((int)mdtidx, (int __user *)arg))
3592 case OBD_IOC_GETDTNAME:
3593 case OBD_IOC_GETMDNAME:
3594 RETURN(ll_get_obd_name(inode, cmd, arg));
3595 case LL_IOC_HSM_STATE_GET: {
3596 struct md_op_data *op_data;
3597 struct hsm_user_state *hus;
3604 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3605 LUSTRE_OPC_ANY, hus);
3606 if (IS_ERR(op_data)) {
3608 RETURN(PTR_ERR(op_data));
3611 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3614 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
3617 ll_finish_md_op_data(op_data);
3621 case LL_IOC_HSM_STATE_SET: {
3622 struct hsm_state_set *hss;
3629 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
3634 rc = ll_hsm_state_set(inode, hss);
3639 case LL_IOC_HSM_ACTION: {
3640 struct md_op_data *op_data;
3641 struct hsm_current_action *hca;
3648 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3649 LUSTRE_OPC_ANY, hca);
3650 if (IS_ERR(op_data)) {
3652 RETURN(PTR_ERR(op_data));
3655 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3658 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
3661 ll_finish_md_op_data(op_data);
3665 case LL_IOC_SET_LEASE_OLD: {
3666 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
3668 RETURN(ll_file_set_lease(file, &ioc, 0));
3670 case LL_IOC_SET_LEASE: {
3671 struct ll_ioc_lease ioc;
3673 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
3676 RETURN(ll_file_set_lease(file, &ioc, arg));
3678 case LL_IOC_GET_LEASE: {
3679 struct ll_inode_info *lli = ll_i2info(inode);
3680 struct ldlm_lock *lock = NULL;
3683 mutex_lock(&lli->lli_och_mutex);
3684 if (fd->fd_lease_och != NULL) {
3685 struct obd_client_handle *och = fd->fd_lease_och;
3687 lock = ldlm_handle2lock(&och->och_lease_handle);
3689 lock_res_and_lock(lock);
3690 if (!ldlm_is_cancel(lock))
3691 fmode = och->och_flags;
3693 unlock_res_and_lock(lock);
3694 LDLM_LOCK_PUT(lock);
3697 mutex_unlock(&lli->lli_och_mutex);
3699 RETURN(ll_lease_type_from_fmode(fmode));
3701 case LL_IOC_HSM_IMPORT: {
3702 struct hsm_user_import *hui;
3708 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
3713 rc = ll_hsm_import(inode, file, hui);
3718 case LL_IOC_FUTIMES_3: {
3719 struct ll_futimes_3 lfu;
3721 if (copy_from_user(&lfu,
3722 (const struct ll_futimes_3 __user *)arg,
3726 RETURN(ll_file_futimes_3(file, &lfu));
3728 case LL_IOC_LADVISE: {
3729 struct llapi_ladvise_hdr *k_ladvise_hdr;
3730 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
3733 int alloc_size = sizeof(*k_ladvise_hdr);
3736 u_ladvise_hdr = (void __user *)arg;
3737 OBD_ALLOC_PTR(k_ladvise_hdr);
3738 if (k_ladvise_hdr == NULL)
3741 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3742 GOTO(out_ladvise, rc = -EFAULT);
3744 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
3745 k_ladvise_hdr->lah_count < 1)
3746 GOTO(out_ladvise, rc = -EINVAL);
3748 num_advise = k_ladvise_hdr->lah_count;
3749 if (num_advise >= LAH_COUNT_MAX)
3750 GOTO(out_ladvise, rc = -EFBIG);
3752 OBD_FREE_PTR(k_ladvise_hdr);
3753 alloc_size = offsetof(typeof(*k_ladvise_hdr),
3754 lah_advise[num_advise]);
3755 OBD_ALLOC(k_ladvise_hdr, alloc_size);
3756 if (k_ladvise_hdr == NULL)
3760 * TODO: submit multiple advices to one server in a single RPC
3762 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3763 GOTO(out_ladvise, rc = -EFAULT);
3765 for (i = 0; i < num_advise; i++) {
3766 struct llapi_lu_ladvise *k_ladvise =
3767 &k_ladvise_hdr->lah_advise[i];
3768 struct llapi_lu_ladvise __user *u_ladvise =
3769 &u_ladvise_hdr->lah_advise[i];
3771 rc = ll_ladvise_sanity(inode, k_ladvise);
3773 GOTO(out_ladvise, rc);
3775 switch (k_ladvise->lla_advice) {
3776 case LU_LADVISE_LOCKNOEXPAND:
3777 rc = ll_lock_noexpand(file,
3778 k_ladvise->lla_peradvice_flags);
3779 GOTO(out_ladvise, rc);
3780 case LU_LADVISE_LOCKAHEAD:
3782 rc = ll_file_lock_ahead(file, k_ladvise);
3785 GOTO(out_ladvise, rc);
3788 &u_ladvise->lla_lockahead_result))
3789 GOTO(out_ladvise, rc = -EFAULT);
3792 rc = ll_ladvise(inode, file,
3793 k_ladvise_hdr->lah_flags,
3796 GOTO(out_ladvise, rc);
3803 OBD_FREE(k_ladvise_hdr, alloc_size);
3806 case LL_IOC_FLR_SET_MIRROR: {
3807 /* mirror I/O must be direct to avoid polluting page cache
3809 if (!(file->f_flags & O_DIRECT))
3812 fd->fd_designated_mirror = (__u32)arg;
3815 case LL_IOC_FSGETXATTR:
3816 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
3817 case LL_IOC_FSSETXATTR:
3818 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
3820 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
3821 case LL_IOC_HEAT_GET: {
3822 struct lu_heat uheat;
3823 struct lu_heat *heat;
3826 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
3829 if (uheat.lh_count > OBD_HEAT_COUNT)
3830 uheat.lh_count = OBD_HEAT_COUNT;
3832 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
3833 OBD_ALLOC(heat, size);
3837 heat->lh_count = uheat.lh_count;
3838 ll_heat_get(inode, heat);
3839 rc = copy_to_user((char __user *)arg, heat, size);
3840 OBD_FREE(heat, size);
3841 RETURN(rc ? -EFAULT : 0);
3843 case LL_IOC_HEAT_SET: {
3846 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
3849 rc = ll_heat_set(inode, flags);
3852 case LL_IOC_PCC_DETACH: {
3853 struct lu_pcc_detach *detach;
3855 OBD_ALLOC_PTR(detach);
3859 if (copy_from_user(detach,
3860 (const struct lu_pcc_detach __user *)arg,
3862 GOTO(out_detach_free, rc = -EFAULT);
3864 if (!S_ISREG(inode->i_mode))
3865 GOTO(out_detach_free, rc = -EINVAL);
3867 if (!inode_owner_or_capable(inode))
3868 GOTO(out_detach_free, rc = -EPERM);
3870 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
3872 OBD_FREE_PTR(detach);
3875 case LL_IOC_PCC_STATE: {
3876 struct lu_pcc_state __user *ustate =
3877 (struct lu_pcc_state __user *)arg;
3878 struct lu_pcc_state *state;
3880 OBD_ALLOC_PTR(state);
3884 if (copy_from_user(state, ustate, sizeof(*state)))
3885 GOTO(out_state, rc = -EFAULT);
3887 rc = pcc_ioctl_state(file, inode, state);
3889 GOTO(out_state, rc);
3891 if (copy_to_user(ustate, state, sizeof(*state)))
3892 GOTO(out_state, rc = -EFAULT);
3895 OBD_FREE_PTR(state);
3899 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
3900 (void __user *)arg));
3904 #ifndef HAVE_FILE_LLSEEK_SIZE
3905 static inline loff_t
3906 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
3908 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
3910 if (offset > maxsize)
3913 if (offset != file->f_pos) {
3914 file->f_pos = offset;
3915 file->f_version = 0;
3921 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
3922 loff_t maxsize, loff_t eof)
3924 struct inode *inode = file_inode(file);
3932 * Here we special-case the lseek(fd, 0, SEEK_CUR)
3933 * position-querying operation. Avoid rewriting the "same"
3934 * f_pos value back to the file because a concurrent read(),
3935 * write() or lseek() might have altered it
3940 * f_lock protects against read/modify/write race with other
3941 * SEEK_CURs. Note that parallel writes and reads behave
3945 offset = llseek_execute(file, file->f_pos + offset, maxsize);
3946 inode_unlock(inode);
3950 * In the generic case the entire file is data, so as long as
3951 * offset isn't at the end of the file then the offset is data.
3958 * There is a virtual hole at the end of the file, so as long as
3959 * offset isn't i_size or larger, return i_size.
3967 return llseek_execute(file, offset, maxsize);
3971 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
3973 struct inode *inode = file_inode(file);
3974 loff_t retval, eof = 0;
3977 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
3978 (origin == SEEK_CUR) ? file->f_pos : 0);
3979 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
3980 PFID(ll_inode2fid(inode)), inode, retval, retval,
3982 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
3984 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
3985 retval = ll_glimpse_size(inode);
3988 eof = i_size_read(inode);
3991 retval = ll_generic_file_llseek_size(file, offset, origin,
3992 ll_file_maxbytes(inode), eof);
3996 static int ll_flush(struct file *file, fl_owner_t id)
3998 struct inode *inode = file_inode(file);
3999 struct ll_inode_info *lli = ll_i2info(inode);
4000 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
4003 LASSERT(!S_ISDIR(inode->i_mode));
4005 /* catch async errors that were recorded back when async writeback
4006 * failed for pages in this mapping. */
4007 rc = lli->lli_async_rc;
4008 lli->lli_async_rc = 0;
4009 if (lli->lli_clob != NULL) {
4010 err = lov_read_and_clear_async_rc(lli->lli_clob);
4015 /* The application has been told write failure already.
4016 * Do not report failure again. */
4017 if (fd->fd_write_failed)
4019 return rc ? -EIO : 0;
4023 * Called to make sure a portion of file has been written out.
4024 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4026 * Return how many pages have been written.
4028 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4029 enum cl_fsync_mode mode, int ignore_layout)
4033 struct cl_fsync_io *fio;
4038 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4039 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4042 env = cl_env_get(&refcheck);
4044 RETURN(PTR_ERR(env));
4046 io = vvp_env_thread_io(env);
4047 io->ci_obj = ll_i2info(inode)->lli_clob;
4048 io->ci_ignore_layout = ignore_layout;
4050 /* initialize parameters for sync */
4051 fio = &io->u.ci_fsync;
4052 fio->fi_start = start;
4054 fio->fi_fid = ll_inode2fid(inode);
4055 fio->fi_mode = mode;
4056 fio->fi_nr_written = 0;
4058 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4059 result = cl_io_loop(env, io);
4061 result = io->ci_result;
4063 result = fio->fi_nr_written;
4064 cl_io_fini(env, io);
4065 cl_env_put(env, &refcheck);
4071 * When dentry is provided (the 'else' case), file_dentry() may be
4072 * null and dentry must be used directly rather than pulled from
4073 * file_dentry() as is done otherwise.
4076 #ifdef HAVE_FILE_FSYNC_4ARGS
4077 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4079 struct dentry *dentry = file_dentry(file);
4080 #elif defined(HAVE_FILE_FSYNC_2ARGS)
4081 int ll_fsync(struct file *file, int datasync)
4083 struct dentry *dentry = file_dentry(file);
4085 loff_t end = LLONG_MAX;
4087 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
4090 loff_t end = LLONG_MAX;
4092 struct inode *inode = dentry->d_inode;
4093 struct ll_inode_info *lli = ll_i2info(inode);
4094 struct ptlrpc_request *req;
4099 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
4100 PFID(ll_inode2fid(inode)), inode);
4101 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
4103 #ifdef HAVE_FILE_FSYNC_4ARGS
4104 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4107 /* fsync's caller has already called _fdata{sync,write}, we want
4108 * that IO to finish before calling the osc and mdc sync methods */
4109 rc = filemap_fdatawait(inode->i_mapping);
4112 /* catch async errors that were recorded back when async writeback
4113 * failed for pages in this mapping. */
4114 if (!S_ISDIR(inode->i_mode)) {
4115 err = lli->lli_async_rc;
4116 lli->lli_async_rc = 0;
4119 if (lli->lli_clob != NULL) {
4120 err = lov_read_and_clear_async_rc(lli->lli_clob);
4126 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4130 ptlrpc_req_finished(req);
4132 if (S_ISREG(inode->i_mode)) {
4133 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
4136 /* Sync metadata on MDT first, and then sync the cached data
4139 err = pcc_fsync(file, start, end, datasync, &cached);
4141 err = cl_sync_file_range(inode, start, end,
4143 if (rc == 0 && err < 0)
4146 fd->fd_write_failed = true;
4148 fd->fd_write_failed = false;
4151 #ifdef HAVE_FILE_FSYNC_4ARGS
4152 inode_unlock(inode);
4158 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4160 struct inode *inode = file_inode(file);
4161 struct ll_sb_info *sbi = ll_i2sbi(inode);
4162 struct ldlm_enqueue_info einfo = {
4163 .ei_type = LDLM_FLOCK,
4164 .ei_cb_cp = ldlm_flock_completion_ast,
4165 .ei_cbdata = file_lock,
4167 struct md_op_data *op_data;
4168 struct lustre_handle lockh = { 0 };
4169 union ldlm_policy_data flock = { { 0 } };
4170 int fl_type = file_lock->fl_type;
4176 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4177 PFID(ll_inode2fid(inode)), file_lock);
4179 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
4181 if (file_lock->fl_flags & FL_FLOCK) {
4182 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4183 /* flocks are whole-file locks */
4184 flock.l_flock.end = OFFSET_MAX;
4185 /* For flocks owner is determined by the local file desctiptor*/
4186 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4187 } else if (file_lock->fl_flags & FL_POSIX) {
4188 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4189 flock.l_flock.start = file_lock->fl_start;
4190 flock.l_flock.end = file_lock->fl_end;
4194 flock.l_flock.pid = file_lock->fl_pid;
4196 /* Somewhat ugly workaround for svc lockd.
4197 * lockd installs custom fl_lmops->lm_compare_owner that checks
4198 * for the fl_owner to be the same (which it always is on local node
4199 * I guess between lockd processes) and then compares pid.
4200 * As such we assign pid to the owner field to make it all work,
4201 * conflict with normal locks is unlikely since pid space and
4202 * pointer space for current->files are not intersecting */
4203 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4204 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4208 einfo.ei_mode = LCK_PR;
4211 /* An unlock request may or may not have any relation to
4212 * existing locks so we may not be able to pass a lock handle
4213 * via a normal ldlm_lock_cancel() request. The request may even
4214 * unlock a byte range in the middle of an existing lock. In
4215 * order to process an unlock request we need all of the same
4216 * information that is given with a normal read or write record
4217 * lock request. To avoid creating another ldlm unlock (cancel)
4218 * message we'll treat a LCK_NL flock request as an unlock. */
4219 einfo.ei_mode = LCK_NL;
4222 einfo.ei_mode = LCK_PW;
4225 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4240 flags = LDLM_FL_BLOCK_NOWAIT;
4246 flags = LDLM_FL_TEST_LOCK;
4249 CERROR("unknown fcntl lock command: %d\n", cmd);
4253 /* Save the old mode so that if the mode in the lock changes we
4254 * can decrement the appropriate reader or writer refcount. */
4255 file_lock->fl_type = einfo.ei_mode;
4257 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4258 LUSTRE_OPC_ANY, NULL);
4259 if (IS_ERR(op_data))
4260 RETURN(PTR_ERR(op_data));
4262 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4263 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4264 flock.l_flock.pid, flags, einfo.ei_mode,
4265 flock.l_flock.start, flock.l_flock.end);
4267 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4270 /* Restore the file lock type if not TEST lock. */
4271 if (!(flags & LDLM_FL_TEST_LOCK))
4272 file_lock->fl_type = fl_type;
4274 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4275 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4276 !(flags & LDLM_FL_TEST_LOCK))
4277 rc2 = locks_lock_file_wait(file, file_lock);
4279 if ((file_lock->fl_flags & FL_FLOCK) &&
4280 (rc == 0 || file_lock->fl_type == F_UNLCK))
4281 rc2 = flock_lock_file_wait(file, file_lock);
4282 if ((file_lock->fl_flags & FL_POSIX) &&
4283 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4284 !(flags & LDLM_FL_TEST_LOCK))
4285 rc2 = posix_lock_file_wait(file, file_lock);
4286 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4288 if (rc2 && file_lock->fl_type != F_UNLCK) {
4289 einfo.ei_mode = LCK_NL;
4290 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4295 ll_finish_md_op_data(op_data);
4300 int ll_get_fid_by_name(struct inode *parent, const char *name,
4301 int namelen, struct lu_fid *fid,
4302 struct inode **inode)
4304 struct md_op_data *op_data = NULL;
4305 struct mdt_body *body;
4306 struct ptlrpc_request *req;
4310 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4311 LUSTRE_OPC_ANY, NULL);
4312 if (IS_ERR(op_data))
4313 RETURN(PTR_ERR(op_data));
4315 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4316 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4317 ll_finish_md_op_data(op_data);
4321 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4323 GOTO(out_req, rc = -EFAULT);
4325 *fid = body->mbo_fid1;
4328 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
4330 ptlrpc_req_finished(req);
4334 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4337 struct dentry *dchild = NULL;
4338 struct inode *child_inode = NULL;
4339 struct md_op_data *op_data;
4340 struct ptlrpc_request *request = NULL;
4341 struct obd_client_handle *och = NULL;
4343 struct mdt_body *body;
4344 __u64 data_version = 0;
4345 size_t namelen = strlen(name);
4346 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4350 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4351 PFID(ll_inode2fid(parent)), name,
4352 lum->lum_stripe_offset, lum->lum_stripe_count);
4354 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4355 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4356 lustre_swab_lmv_user_md(lum);
4358 /* Get child FID first */
4359 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4362 dchild = d_lookup(file_dentry(file), &qstr);
4364 if (dchild->d_inode)
4365 child_inode = igrab(dchild->d_inode);
4370 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4379 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4380 OBD_CONNECT2_DIR_MIGRATE)) {
4381 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4382 ll_dir_striped(child_inode)) {
4383 CERROR("%s: MDT doesn't support stripe directory "
4384 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4385 GOTO(out_iput, rc = -EOPNOTSUPP);
4390 * lfs migrate command needs to be blocked on the client
4391 * by checking the migrate FID against the FID of the
4394 if (child_inode == parent->i_sb->s_root->d_inode)
4395 GOTO(out_iput, rc = -EINVAL);
4397 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4398 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4399 if (IS_ERR(op_data))
4400 GOTO(out_iput, rc = PTR_ERR(op_data));
4402 inode_lock(child_inode);
4403 op_data->op_fid3 = *ll_inode2fid(child_inode);
4404 if (!fid_is_sane(&op_data->op_fid3)) {
4405 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4406 ll_i2sbi(parent)->ll_fsname, name,
4407 PFID(&op_data->op_fid3));
4408 GOTO(out_unlock, rc = -EINVAL);
4411 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4412 op_data->op_data = lum;
4413 op_data->op_data_size = lumlen;
4416 if (S_ISREG(child_inode->i_mode)) {
4417 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4421 GOTO(out_unlock, rc);
4424 rc = ll_data_version(child_inode, &data_version,
4427 GOTO(out_close, rc);
4429 op_data->op_open_handle = och->och_open_handle;
4430 op_data->op_data_version = data_version;
4431 op_data->op_lease_handle = och->och_lease_handle;
4432 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4434 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4435 och->och_mod->mod_open_req->rq_replay = 0;
4436 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4439 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name, namelen,
4440 name, namelen, &request);
4442 LASSERT(request != NULL);
4443 ll_update_times(request, parent);
4446 if (rc == 0 || rc == -EAGAIN) {
4447 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4448 LASSERT(body != NULL);
4450 /* If the server does release layout lock, then we cleanup
4451 * the client och here, otherwise release it in out_close: */
4452 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4453 obd_mod_put(och->och_mod);
4454 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4456 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4462 if (request != NULL) {
4463 ptlrpc_req_finished(request);
4467 /* Try again if the lease has cancelled. */
4468 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4473 ll_lease_close(och, child_inode, NULL);
4475 clear_nlink(child_inode);
4477 inode_unlock(child_inode);
4478 ll_finish_md_op_data(op_data);
4485 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4493 * test if some locks matching bits and l_req_mode are acquired
4494 * - bits can be in different locks
4495 * - if found clear the common lock bits in *bits
4496 * - the bits not found, are kept in *bits
4498 * \param bits [IN] searched lock bits [IN]
4499 * \param l_req_mode [IN] searched lock mode
4500 * \retval boolean, true iff all bits are found
4502 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
4504 struct lustre_handle lockh;
4505 union ldlm_policy_data policy;
4506 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
4507 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
4516 fid = &ll_i2info(inode)->lli_fid;
4517 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
4518 ldlm_lockname[mode]);
4520 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
4521 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
4522 policy.l_inodebits.bits = *bits & (1 << i);
4523 if (policy.l_inodebits.bits == 0)
4526 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
4527 &policy, mode, &lockh)) {
4528 struct ldlm_lock *lock;
4530 lock = ldlm_handle2lock(&lockh);
4533 ~(lock->l_policy_data.l_inodebits.bits);
4534 LDLM_LOCK_PUT(lock);
4536 *bits &= ~policy.l_inodebits.bits;
4543 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
4544 struct lustre_handle *lockh, __u64 flags,
4545 enum ldlm_mode mode)
4547 union ldlm_policy_data policy = { .l_inodebits = { bits } };
4552 fid = &ll_i2info(inode)->lli_fid;
4553 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
4555 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
4556 fid, LDLM_IBITS, &policy, mode, lockh);
4561 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
4563 /* Already unlinked. Just update nlink and return success */
4564 if (rc == -ENOENT) {
4566 /* If it is striped directory, and there is bad stripe
4567 * Let's revalidate the dentry again, instead of returning
4569 if (ll_dir_striped(inode))
4572 /* This path cannot be hit for regular files unless in
4573 * case of obscure races, so no need to to validate
4575 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
4577 } else if (rc != 0) {
4578 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
4579 "%s: revalidate FID "DFID" error: rc = %d\n",
4580 ll_i2sbi(inode)->ll_fsname,
4581 PFID(ll_inode2fid(inode)), rc);
4587 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
4589 struct inode *inode = dentry->d_inode;
4590 struct obd_export *exp = ll_i2mdexp(inode);
4591 struct lookup_intent oit = {
4594 struct ptlrpc_request *req = NULL;
4595 struct md_op_data *op_data;
4599 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
4600 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
4602 /* Call getattr by fid, so do not provide name at all. */
4603 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
4604 LUSTRE_OPC_ANY, NULL);
4605 if (IS_ERR(op_data))
4606 RETURN(PTR_ERR(op_data));
4608 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
4609 ll_finish_md_op_data(op_data);
4611 rc = ll_inode_revalidate_fini(inode, rc);
4615 rc = ll_revalidate_it_finish(req, &oit, dentry);
4617 ll_intent_release(&oit);
4621 /* Unlinked? Unhash dentry, so it is not picked up later by
4622 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
4623 * here to preserve get_cwd functionality on 2.6.
4625 if (!dentry->d_inode->i_nlink) {
4626 ll_lock_dcache(inode);
4627 d_lustre_invalidate(dentry, 0);
4628 ll_unlock_dcache(inode);
4631 ll_lookup_finish_locks(&oit, dentry);
4633 ptlrpc_req_finished(req);
4638 static int ll_merge_md_attr(struct inode *inode)
4640 struct ll_inode_info *lli = ll_i2info(inode);
4641 struct cl_attr attr = { 0 };
4644 LASSERT(lli->lli_lsm_md != NULL);
4646 if (!lmv_dir_striped(lli->lli_lsm_md))
4649 down_read(&lli->lli_lsm_sem);
4650 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
4651 &attr, ll_md_blocking_ast);
4652 up_read(&lli->lli_lsm_sem);
4656 set_nlink(inode, attr.cat_nlink);
4657 inode->i_blocks = attr.cat_blocks;
4658 i_size_write(inode, attr.cat_size);
4660 ll_i2info(inode)->lli_atime = attr.cat_atime;
4661 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
4662 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
4667 int ll_getattr_dentry(struct dentry *de, struct kstat *stat)
4669 struct inode *inode = de->d_inode;
4670 struct ll_sb_info *sbi = ll_i2sbi(inode);
4671 struct ll_inode_info *lli = ll_i2info(inode);
4674 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
4676 rc = ll_inode_revalidate(de, IT_GETATTR);
4680 if (S_ISREG(inode->i_mode)) {
4683 rc = pcc_inode_getattr(inode, &cached);
4684 if (cached && rc < 0)
4687 /* In case of restore, the MDT has the right size and has
4688 * already send it back without granting the layout lock,
4689 * inode is up-to-date so glimpse is useless.
4690 * Also to glimpse we need the layout, in case of a running
4691 * restore the MDT holds the layout lock so the glimpse will
4692 * block up to the end of restore (getattr will block)
4694 if (!cached && !ll_file_test_flag(lli, LLIF_FILE_RESTORING)) {
4695 rc = ll_glimpse_size(inode);
4700 /* If object isn't regular a file then don't validate size. */
4701 if (ll_dir_striped(inode)) {
4702 rc = ll_merge_md_attr(inode);
4707 inode->i_atime.tv_sec = lli->lli_atime;
4708 inode->i_mtime.tv_sec = lli->lli_mtime;
4709 inode->i_ctime.tv_sec = lli->lli_ctime;
4712 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
4714 if (ll_need_32bit_api(sbi)) {
4715 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
4716 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
4717 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
4719 stat->ino = inode->i_ino;
4720 stat->dev = inode->i_sb->s_dev;
4721 stat->rdev = inode->i_rdev;
4724 stat->mode = inode->i_mode;
4725 stat->uid = inode->i_uid;
4726 stat->gid = inode->i_gid;
4727 stat->atime = inode->i_atime;
4728 stat->mtime = inode->i_mtime;
4729 stat->ctime = inode->i_ctime;
4730 stat->blksize = sbi->ll_stat_blksize ?: 1 << inode->i_blkbits;
4732 stat->nlink = inode->i_nlink;
4733 stat->size = i_size_read(inode);
4734 stat->blocks = inode->i_blocks;
4739 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
4740 int ll_getattr(const struct path *path, struct kstat *stat,
4741 u32 request_mask, unsigned int flags)
4743 struct dentry *de = path->dentry;
4745 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
4748 return ll_getattr_dentry(de, stat);
4751 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4752 __u64 start, __u64 len)
4756 struct fiemap *fiemap;
4757 unsigned int extent_count = fieinfo->fi_extents_max;
4759 num_bytes = sizeof(*fiemap) + (extent_count *
4760 sizeof(struct fiemap_extent));
4761 OBD_ALLOC_LARGE(fiemap, num_bytes);
4766 fiemap->fm_flags = fieinfo->fi_flags;
4767 fiemap->fm_extent_count = fieinfo->fi_extents_max;
4768 fiemap->fm_start = start;
4769 fiemap->fm_length = len;
4770 if (extent_count > 0 &&
4771 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
4772 sizeof(struct fiemap_extent)) != 0)
4773 GOTO(out, rc = -EFAULT);
4775 rc = ll_do_fiemap(inode, fiemap, num_bytes);
4777 fieinfo->fi_flags = fiemap->fm_flags;
4778 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
4779 if (extent_count > 0 &&
4780 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
4781 fiemap->fm_mapped_extents *
4782 sizeof(struct fiemap_extent)) != 0)
4783 GOTO(out, rc = -EFAULT);
4785 OBD_FREE_LARGE(fiemap, num_bytes);
4789 struct posix_acl *ll_get_acl(struct inode *inode, int type)
4791 struct ll_inode_info *lli = ll_i2info(inode);
4792 struct posix_acl *acl = NULL;
4795 spin_lock(&lli->lli_lock);
4796 /* VFS' acl_permission_check->check_acl will release the refcount */
4797 acl = posix_acl_dup(lli->lli_posix_acl);
4798 spin_unlock(&lli->lli_lock);
4803 #ifdef HAVE_IOP_SET_ACL
4804 #ifdef CONFIG_FS_POSIX_ACL
4805 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type)
4807 struct ll_sb_info *sbi = ll_i2sbi(inode);
4808 struct ptlrpc_request *req = NULL;
4809 const char *name = NULL;
4811 size_t value_size = 0;
4816 case ACL_TYPE_ACCESS:
4817 name = XATTR_NAME_POSIX_ACL_ACCESS;
4819 rc = posix_acl_update_mode(inode, &inode->i_mode, &acl);
4822 case ACL_TYPE_DEFAULT:
4823 name = XATTR_NAME_POSIX_ACL_DEFAULT;
4824 if (!S_ISDIR(inode->i_mode))
4825 rc = acl ? -EACCES : 0;
4836 value_size = posix_acl_xattr_size(acl->a_count);
4837 value = kmalloc(value_size, GFP_NOFS);
4839 GOTO(out, rc = -ENOMEM);
4841 rc = posix_acl_to_xattr(&init_user_ns, acl, value, value_size);
4843 GOTO(out_value, rc);
4846 rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode),
4847 value ? OBD_MD_FLXATTR : OBD_MD_FLXATTRRM,
4848 name, value, value_size, 0, 0, &req);
4850 ptlrpc_req_finished(req);
4855 forget_cached_acl(inode, type);
4857 set_cached_acl(inode, type, acl);
4860 #endif /* CONFIG_FS_POSIX_ACL */
4861 #endif /* HAVE_IOP_SET_ACL */
4863 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
4865 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
4866 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
4868 ll_check_acl(struct inode *inode, int mask)
4871 # ifdef CONFIG_FS_POSIX_ACL
4872 struct posix_acl *acl;
4876 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
4877 if (flags & IPERM_FLAG_RCU)
4880 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
4885 rc = posix_acl_permission(inode, acl, mask);
4886 posix_acl_release(acl);
4889 # else /* !CONFIG_FS_POSIX_ACL */
4891 # endif /* CONFIG_FS_POSIX_ACL */
4893 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
4895 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
4896 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
4898 # ifdef HAVE_INODE_PERMISION_2ARGS
4899 int ll_inode_permission(struct inode *inode, int mask)
4901 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
4906 struct ll_sb_info *sbi;
4907 struct root_squash_info *squash;
4908 struct cred *cred = NULL;
4909 const struct cred *old_cred = NULL;
4911 bool squash_id = false;
4914 #ifdef MAY_NOT_BLOCK
4915 if (mask & MAY_NOT_BLOCK)
4917 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
4918 if (flags & IPERM_FLAG_RCU)
4922 /* as root inode are NOT getting validated in lookup operation,
4923 * need to do it before permission check. */
4925 if (inode == inode->i_sb->s_root->d_inode) {
4926 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
4931 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
4932 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
4934 /* squash fsuid/fsgid if needed */
4935 sbi = ll_i2sbi(inode);
4936 squash = &sbi->ll_squash;
4937 if (unlikely(squash->rsi_uid != 0 &&
4938 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
4939 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
4943 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
4944 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
4945 squash->rsi_uid, squash->rsi_gid);
4947 /* update current process's credentials
4948 * and FS capability */
4949 cred = prepare_creds();
4953 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
4954 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
4955 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
4956 if ((1 << cap) & CFS_CAP_FS_MASK)
4957 cap_lower(cred->cap_effective, cap);
4959 old_cred = override_creds(cred);
4962 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
4963 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
4964 /* restore current process's credentials and FS capability */
4966 revert_creds(old_cred);
4973 /* -o localflock - only provides locally consistent flock locks */
4974 struct file_operations ll_file_operations = {
4975 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
4976 # ifdef HAVE_SYNC_READ_WRITE
4977 .read = new_sync_read,
4978 .write = new_sync_write,
4980 .read_iter = ll_file_read_iter,
4981 .write_iter = ll_file_write_iter,
4982 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
4983 .read = ll_file_read,
4984 .aio_read = ll_file_aio_read,
4985 .write = ll_file_write,
4986 .aio_write = ll_file_aio_write,
4987 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
4988 .unlocked_ioctl = ll_file_ioctl,
4989 .open = ll_file_open,
4990 .release = ll_file_release,
4991 .mmap = ll_file_mmap,
4992 .llseek = ll_file_seek,
4993 .splice_read = ll_file_splice_read,
4998 struct file_operations ll_file_operations_flock = {
4999 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5000 # ifdef HAVE_SYNC_READ_WRITE
5001 .read = new_sync_read,
5002 .write = new_sync_write,
5003 # endif /* HAVE_SYNC_READ_WRITE */
5004 .read_iter = ll_file_read_iter,
5005 .write_iter = ll_file_write_iter,
5006 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5007 .read = ll_file_read,
5008 .aio_read = ll_file_aio_read,
5009 .write = ll_file_write,
5010 .aio_write = ll_file_aio_write,
5011 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5012 .unlocked_ioctl = ll_file_ioctl,
5013 .open = ll_file_open,
5014 .release = ll_file_release,
5015 .mmap = ll_file_mmap,
5016 .llseek = ll_file_seek,
5017 .splice_read = ll_file_splice_read,
5020 .flock = ll_file_flock,
5021 .lock = ll_file_flock
5024 /* These are for -o noflock - to return ENOSYS on flock calls */
5025 struct file_operations ll_file_operations_noflock = {
5026 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5027 # ifdef HAVE_SYNC_READ_WRITE
5028 .read = new_sync_read,
5029 .write = new_sync_write,
5030 # endif /* HAVE_SYNC_READ_WRITE */
5031 .read_iter = ll_file_read_iter,
5032 .write_iter = ll_file_write_iter,
5033 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5034 .read = ll_file_read,
5035 .aio_read = ll_file_aio_read,
5036 .write = ll_file_write,
5037 .aio_write = ll_file_aio_write,
5038 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5039 .unlocked_ioctl = ll_file_ioctl,
5040 .open = ll_file_open,
5041 .release = ll_file_release,
5042 .mmap = ll_file_mmap,
5043 .llseek = ll_file_seek,
5044 .splice_read = ll_file_splice_read,
5047 .flock = ll_file_noflock,
5048 .lock = ll_file_noflock
5051 struct inode_operations ll_file_inode_operations = {
5052 .setattr = ll_setattr,
5053 .getattr = ll_getattr,
5054 .permission = ll_inode_permission,
5055 #ifdef HAVE_IOP_XATTR
5056 .setxattr = ll_setxattr,
5057 .getxattr = ll_getxattr,
5058 .removexattr = ll_removexattr,
5060 .listxattr = ll_listxattr,
5061 .fiemap = ll_fiemap,
5062 #ifdef HAVE_IOP_GET_ACL
5063 .get_acl = ll_get_acl,
5065 #ifdef HAVE_IOP_SET_ACL
5066 .set_acl = ll_set_acl,
5070 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5072 struct ll_inode_info *lli = ll_i2info(inode);
5073 struct cl_object *obj = lli->lli_clob;
5082 env = cl_env_get(&refcheck);
5084 RETURN(PTR_ERR(env));
5086 rc = cl_conf_set(env, lli->lli_clob, conf);
5090 if (conf->coc_opc == OBJECT_CONF_SET) {
5091 struct ldlm_lock *lock = conf->coc_lock;
5092 struct cl_layout cl = {
5096 LASSERT(lock != NULL);
5097 LASSERT(ldlm_has_layout(lock));
5099 /* it can only be allowed to match after layout is
5100 * applied to inode otherwise false layout would be
5101 * seen. Applying layout shoud happen before dropping
5102 * the intent lock. */
5103 ldlm_lock_allow_match(lock);
5105 rc = cl_object_layout_get(env, obj, &cl);
5110 DFID": layout version change: %u -> %u\n",
5111 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5113 ll_layout_version_set(lli, cl.cl_layout_gen);
5117 cl_env_put(env, &refcheck);
5122 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5123 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5126 struct ll_sb_info *sbi = ll_i2sbi(inode);
5127 struct ptlrpc_request *req;
5134 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5135 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5136 lock->l_lvb_data, lock->l_lvb_len);
5138 if (lock->l_lvb_data != NULL)
5141 /* if layout lock was granted right away, the layout is returned
5142 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5143 * blocked and then granted via completion ast, we have to fetch
5144 * layout here. Please note that we can't use the LVB buffer in
5145 * completion AST because it doesn't have a large enough buffer */
5146 rc = ll_get_default_mdsize(sbi, &lmmsize);
5150 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5151 XATTR_NAME_LOV, lmmsize, &req);
5154 GOTO(out, rc = 0); /* empty layout */
5161 if (lmmsize == 0) /* empty layout */
5164 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5166 GOTO(out, rc = -EFAULT);
5168 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5169 if (lvbdata == NULL)
5170 GOTO(out, rc = -ENOMEM);
5172 memcpy(lvbdata, lmm, lmmsize);
5173 lock_res_and_lock(lock);
5174 if (unlikely(lock->l_lvb_data == NULL)) {
5175 lock->l_lvb_type = LVB_T_LAYOUT;
5176 lock->l_lvb_data = lvbdata;
5177 lock->l_lvb_len = lmmsize;
5180 unlock_res_and_lock(lock);
5183 OBD_FREE_LARGE(lvbdata, lmmsize);
5188 ptlrpc_req_finished(req);
5193 * Apply the layout to the inode. Layout lock is held and will be released
5196 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5197 struct inode *inode)
5199 struct ll_inode_info *lli = ll_i2info(inode);
5200 struct ll_sb_info *sbi = ll_i2sbi(inode);
5201 struct ldlm_lock *lock;
5202 struct cl_object_conf conf;
5205 bool wait_layout = false;
5208 LASSERT(lustre_handle_is_used(lockh));
5210 lock = ldlm_handle2lock(lockh);
5211 LASSERT(lock != NULL);
5212 LASSERT(ldlm_has_layout(lock));
5214 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5215 PFID(&lli->lli_fid), inode);
5217 /* in case this is a caching lock and reinstate with new inode */
5218 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5220 lock_res_and_lock(lock);
5221 lvb_ready = ldlm_is_lvb_ready(lock);
5222 unlock_res_and_lock(lock);
5224 /* checking lvb_ready is racy but this is okay. The worst case is
5225 * that multi processes may configure the file on the same time. */
5229 rc = ll_layout_fetch(inode, lock);
5233 /* for layout lock, lmm is stored in lock's lvb.
5234 * lvb_data is immutable if the lock is held so it's safe to access it
5237 * set layout to file. Unlikely this will fail as old layout was
5238 * surely eliminated */
5239 memset(&conf, 0, sizeof conf);
5240 conf.coc_opc = OBJECT_CONF_SET;
5241 conf.coc_inode = inode;
5242 conf.coc_lock = lock;
5243 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5244 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5245 rc = ll_layout_conf(inode, &conf);
5247 /* refresh layout failed, need to wait */
5248 wait_layout = rc == -EBUSY;
5251 LDLM_LOCK_PUT(lock);
5252 ldlm_lock_decref(lockh, mode);
5254 /* wait for IO to complete if it's still being used. */
5256 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5257 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5259 memset(&conf, 0, sizeof conf);
5260 conf.coc_opc = OBJECT_CONF_WAIT;
5261 conf.coc_inode = inode;
5262 rc = ll_layout_conf(inode, &conf);
5266 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5267 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5273 * Issue layout intent RPC to MDS.
5274 * \param inode [in] file inode
5275 * \param intent [in] layout intent
5277 * \retval 0 on success
5278 * \retval < 0 error code
5280 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5282 struct ll_inode_info *lli = ll_i2info(inode);
5283 struct ll_sb_info *sbi = ll_i2sbi(inode);
5284 struct md_op_data *op_data;
5285 struct lookup_intent it;
5286 struct ptlrpc_request *req;
5290 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5291 0, 0, LUSTRE_OPC_ANY, NULL);
5292 if (IS_ERR(op_data))
5293 RETURN(PTR_ERR(op_data));
5295 op_data->op_data = intent;
5296 op_data->op_data_size = sizeof(*intent);
5298 memset(&it, 0, sizeof(it));
5299 it.it_op = IT_LAYOUT;
5300 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5301 intent->li_opc == LAYOUT_INTENT_TRUNC)
5302 it.it_flags = FMODE_WRITE;
5304 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5305 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5307 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5308 &ll_md_blocking_ast, 0);
5309 if (it.it_request != NULL)
5310 ptlrpc_req_finished(it.it_request);
5311 it.it_request = NULL;
5313 ll_finish_md_op_data(op_data);
5315 /* set lock data in case this is a new lock */
5317 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5319 ll_intent_drop_lock(&it);
5325 * This function checks if there exists a LAYOUT lock on the client side,
5326 * or enqueues it if it doesn't have one in cache.
5328 * This function will not hold layout lock so it may be revoked any time after
5329 * this function returns. Any operations depend on layout should be redone
5332 * This function should be called before lov_io_init() to get an uptodate
5333 * layout version, the caller should save the version number and after IO
5334 * is finished, this function should be called again to verify that layout
5335 * is not changed during IO time.
5337 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5339 struct ll_inode_info *lli = ll_i2info(inode);
5340 struct ll_sb_info *sbi = ll_i2sbi(inode);
5341 struct lustre_handle lockh;
5342 struct layout_intent intent = {
5343 .li_opc = LAYOUT_INTENT_ACCESS,
5345 enum ldlm_mode mode;
5349 *gen = ll_layout_version_get(lli);
5350 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
5354 LASSERT(fid_is_sane(ll_inode2fid(inode)));
5355 LASSERT(S_ISREG(inode->i_mode));
5357 /* take layout lock mutex to enqueue layout lock exclusively. */
5358 mutex_lock(&lli->lli_layout_mutex);
5361 /* mostly layout lock is caching on the local side, so try to
5362 * match it before grabbing layout lock mutex. */
5363 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
5364 LCK_CR | LCK_CW | LCK_PR |
5366 if (mode != 0) { /* hit cached lock */
5367 rc = ll_layout_lock_set(&lockh, mode, inode);
5373 rc = ll_layout_intent(inode, &intent);
5379 *gen = ll_layout_version_get(lli);
5380 mutex_unlock(&lli->lli_layout_mutex);
5386 * Issue layout intent RPC indicating where in a file an IO is about to write.
5388 * \param[in] inode file inode.
5389 * \param[in] ext write range with start offset of fille in bytes where
5390 * an IO is about to write, and exclusive end offset in
5393 * \retval 0 on success
5394 * \retval < 0 error code
5396 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
5397 struct lu_extent *ext)
5399 struct layout_intent intent = {
5401 .li_extent.e_start = ext->e_start,
5402 .li_extent.e_end = ext->e_end,
5407 rc = ll_layout_intent(inode, &intent);
5413 * This function send a restore request to the MDT
5415 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
5417 struct hsm_user_request *hur;
5421 len = sizeof(struct hsm_user_request) +
5422 sizeof(struct hsm_user_item);
5423 OBD_ALLOC(hur, len);
5427 hur->hur_request.hr_action = HUA_RESTORE;
5428 hur->hur_request.hr_archive_id = 0;
5429 hur->hur_request.hr_flags = 0;
5430 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
5431 sizeof(hur->hur_user_item[0].hui_fid));
5432 hur->hur_user_item[0].hui_extent.offset = offset;
5433 hur->hur_user_item[0].hui_extent.length = length;
5434 hur->hur_request.hr_itemcount = 1;
5435 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,