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 #include <linux/uidgid.h>
46 #include <linux/falloc.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <uapi/linux/llcrypt.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;
167 case MDS_CLOSE_LAYOUT_SPLIT:
168 case MDS_CLOSE_LAYOUT_SWAP: {
169 struct split_param *sp = data;
171 LASSERT(data != NULL);
172 op_data->op_bias |= bias;
173 op_data->op_data_version = 0;
174 op_data->op_lease_handle = och->och_lease_handle;
175 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
176 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
177 op_data->op_mirror_id = sp->sp_mirror_id;
179 op_data->op_fid2 = *ll_inode2fid(data);
184 case MDS_CLOSE_RESYNC_DONE: {
185 struct ll_ioc_lease *ioc = data;
187 LASSERT(data != NULL);
188 op_data->op_attr_blocks +=
189 ioc->lil_count * op_data->op_attr_blocks;
190 op_data->op_attr.ia_valid |= ATTR_SIZE;
191 op_data->op_xvalid |= OP_XVALID_BLOCKS;
192 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
194 op_data->op_lease_handle = och->och_lease_handle;
195 op_data->op_data = &ioc->lil_ids[0];
196 op_data->op_data_size =
197 ioc->lil_count * sizeof(ioc->lil_ids[0]);
201 case MDS_PCC_ATTACH: {
202 struct pcc_param *param = data;
204 LASSERT(data != NULL);
205 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
206 op_data->op_archive_id = param->pa_archive_id;
207 op_data->op_data_version = param->pa_data_version;
208 op_data->op_lease_handle = och->och_lease_handle;
212 case MDS_HSM_RELEASE:
213 LASSERT(data != NULL);
214 op_data->op_bias |= MDS_HSM_RELEASE;
215 op_data->op_data_version = *(__u64 *)data;
216 op_data->op_lease_handle = och->och_lease_handle;
217 op_data->op_attr.ia_valid |= ATTR_SIZE;
218 op_data->op_xvalid |= OP_XVALID_BLOCKS;
222 LASSERT(data == NULL);
226 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
227 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
228 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
229 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
231 rc = md_close(md_exp, op_data, och->och_mod, &req);
232 if (rc != 0 && rc != -EINTR)
233 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
234 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
236 if (rc == 0 && op_data->op_bias & bias) {
237 struct mdt_body *body;
239 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
240 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
243 if (bias & MDS_PCC_ATTACH) {
244 struct pcc_param *param = data;
246 param->pa_layout_gen = body->mbo_layout_gen;
250 ll_finish_md_op_data(op_data);
254 md_clear_open_replay_data(md_exp, och);
255 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
258 ptlrpc_req_finished(req); /* This is close request */
262 int ll_md_real_close(struct inode *inode, fmode_t fmode)
264 struct ll_inode_info *lli = ll_i2info(inode);
265 struct obd_client_handle **och_p;
266 struct obd_client_handle *och;
271 if (fmode & FMODE_WRITE) {
272 och_p = &lli->lli_mds_write_och;
273 och_usecount = &lli->lli_open_fd_write_count;
274 } else if (fmode & FMODE_EXEC) {
275 och_p = &lli->lli_mds_exec_och;
276 och_usecount = &lli->lli_open_fd_exec_count;
278 LASSERT(fmode & FMODE_READ);
279 och_p = &lli->lli_mds_read_och;
280 och_usecount = &lli->lli_open_fd_read_count;
283 mutex_lock(&lli->lli_och_mutex);
284 if (*och_usecount > 0) {
285 /* There are still users of this handle, so skip
287 mutex_unlock(&lli->lli_och_mutex);
293 mutex_unlock(&lli->lli_och_mutex);
296 /* There might be a race and this handle may already
298 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
304 static int ll_md_close(struct inode *inode, struct file *file)
306 union ldlm_policy_data policy = {
307 .l_inodebits = { MDS_INODELOCK_OPEN },
309 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
310 struct ll_file_data *fd = file->private_data;
311 struct ll_inode_info *lli = ll_i2info(inode);
312 struct lustre_handle lockh;
313 enum ldlm_mode lockmode;
317 /* clear group lock, if present */
318 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
319 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
321 if (fd->fd_lease_och != NULL) {
324 /* Usually the lease is not released when the
325 * application crashed, we need to release here. */
326 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
327 CDEBUG(rc ? D_ERROR : D_INODE, "Clean up lease "DFID" %d/%d\n",
328 PFID(&lli->lli_fid), rc, lease_broken);
330 fd->fd_lease_och = NULL;
333 if (fd->fd_och != NULL) {
334 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
339 /* Let's see if we have good enough OPEN lock on the file and if
340 we can skip talking to MDS */
341 mutex_lock(&lli->lli_och_mutex);
342 if (fd->fd_omode & FMODE_WRITE) {
344 LASSERT(lli->lli_open_fd_write_count);
345 lli->lli_open_fd_write_count--;
346 } else if (fd->fd_omode & FMODE_EXEC) {
348 LASSERT(lli->lli_open_fd_exec_count);
349 lli->lli_open_fd_exec_count--;
352 LASSERT(lli->lli_open_fd_read_count);
353 lli->lli_open_fd_read_count--;
355 mutex_unlock(&lli->lli_och_mutex);
357 /* LU-4398: do not cache write open lock if the file has exec bit */
358 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
359 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
360 LDLM_IBITS, &policy, lockmode, &lockh))
361 rc = ll_md_real_close(inode, fd->fd_omode);
364 file->private_data = NULL;
365 ll_file_data_put(fd);
370 /* While this returns an error code, fput() the caller does not, so we need
371 * to make every effort to clean up all of our state here. Also, applications
372 * rarely check close errors and even if an error is returned they will not
373 * re-try the close call.
375 int ll_file_release(struct inode *inode, struct file *file)
377 struct ll_file_data *fd;
378 struct ll_sb_info *sbi = ll_i2sbi(inode);
379 struct ll_inode_info *lli = ll_i2info(inode);
380 ktime_t kstart = ktime_get();
385 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
386 PFID(ll_inode2fid(inode)), inode);
388 fd = file->private_data;
391 /* The last ref on @file, maybe not the the owner pid of statahead,
392 * because parent and child process can share the same file handle. */
393 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
394 ll_deauthorize_statahead(inode, fd);
396 if (inode->i_sb->s_root == file_dentry(file)) {
397 file->private_data = NULL;
398 ll_file_data_put(fd);
402 pcc_file_release(inode, file);
404 if (!S_ISDIR(inode->i_mode)) {
405 if (lli->lli_clob != NULL)
406 lov_read_and_clear_async_rc(lli->lli_clob);
407 lli->lli_async_rc = 0;
410 rc = ll_md_close(inode, file);
412 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
413 libcfs_debug_dumplog();
416 if (!rc && inode->i_sb->s_root != file_dentry(file))
417 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
418 ktime_us_delta(ktime_get(), kstart));
422 static inline int ll_dom_readpage(void *data, struct page *page)
424 struct niobuf_local *lnb = data;
428 struct inode *inode = page2inode(page);
430 kaddr = kmap_atomic(page);
431 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
432 if (lnb->lnb_len < PAGE_SIZE)
433 memset(kaddr + lnb->lnb_len, 0,
434 PAGE_SIZE - lnb->lnb_len);
435 flush_dcache_page(page);
436 SetPageUptodate(page);
437 kunmap_atomic(kaddr);
439 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
440 if (!llcrypt_has_encryption_key(inode))
441 CDEBUG(D_SEC, "no enc key for "DFID"\n",
442 PFID(ll_inode2fid(inode)));
444 unsigned int offs = 0;
446 while (offs < PAGE_SIZE) {
447 /* decrypt only if page is not empty */
448 if (memcmp(page_address(page) + offs,
449 page_address(ZERO_PAGE(0)),
450 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
453 rc = llcrypt_decrypt_pagecache_blocks(page,
454 LUSTRE_ENCRYPTION_UNIT_SIZE,
459 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
468 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
472 struct ll_inode_info *lli = ll_i2info(inode);
473 struct cl_object *obj = lli->lli_clob;
474 struct address_space *mapping = inode->i_mapping;
476 struct niobuf_remote *rnb;
477 struct mdt_body *body;
479 unsigned long index, start;
480 struct niobuf_local lnb;
489 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
493 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
494 if (rnb == NULL || rnb->rnb_len == 0)
497 /* LU-11595: Server may return whole file and that is OK always or
498 * it may return just file tail and its offset must be aligned with
499 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
500 * smaller then offset may be not aligned and that data is just ignored.
502 if (rnb->rnb_offset & ~PAGE_MASK)
505 /* Server returns whole file or just file tail if it fills in reply
506 * buffer, in both cases total size should be equal to the file size.
508 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
509 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
510 !(inode && IS_ENCRYPTED(inode))) {
511 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
512 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
513 rnb->rnb_len, body->mbo_dom_size);
517 env = cl_env_get(&refcheck);
520 io = vvp_env_thread_io(env);
522 io->ci_ignore_layout = 1;
523 rc = cl_io_init(env, io, CIT_MISC, obj);
527 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
528 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
530 data = (char *)rnb + sizeof(*rnb);
532 lnb.lnb_file_offset = rnb->rnb_offset;
533 start = lnb.lnb_file_offset >> PAGE_SHIFT;
535 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
536 lnb.lnb_page_offset = 0;
538 struct cl_page *page;
540 lnb.lnb_data = data + (index << PAGE_SHIFT);
541 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
542 if (lnb.lnb_len > PAGE_SIZE)
543 lnb.lnb_len = PAGE_SIZE;
545 vmpage = read_cache_page(mapping, index + start,
546 ll_dom_readpage, &lnb);
547 if (IS_ERR(vmpage)) {
548 CWARN("%s: cannot fill page %lu for "DFID
549 " with data: rc = %li\n",
550 ll_i2sbi(inode)->ll_fsname, index + start,
551 PFID(lu_object_fid(&obj->co_lu)),
556 if (vmpage->mapping == NULL) {
559 /* page was truncated */
562 /* attach VM page to CL page cache */
563 page = cl_page_find(env, obj, vmpage->index, vmpage,
566 ClearPageUptodate(vmpage);
571 cl_page_export(env, page, 1);
572 cl_page_put(env, page);
576 } while (rnb->rnb_len > (index << PAGE_SHIFT));
580 cl_env_put(env, &refcheck);
585 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
586 struct lookup_intent *itp)
588 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
589 struct dentry *parent = de->d_parent;
592 struct md_op_data *op_data;
593 struct ptlrpc_request *req = NULL;
597 LASSERT(parent != NULL);
598 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
600 /* if server supports open-by-fid, or file name is invalid, don't pack
601 * name in open request */
602 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
603 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
605 len = de->d_name.len;
606 name = kmalloc(len + 1, GFP_NOFS);
611 spin_lock(&de->d_lock);
612 if (len != de->d_name.len) {
613 spin_unlock(&de->d_lock);
617 memcpy(name, de->d_name.name, len);
619 spin_unlock(&de->d_lock);
621 if (!lu_name_is_valid_2(name, len)) {
627 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
628 name, len, 0, LUSTRE_OPC_ANY, NULL);
629 if (IS_ERR(op_data)) {
631 RETURN(PTR_ERR(op_data));
633 op_data->op_data = lmm;
634 op_data->op_data_size = lmmsize;
636 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
637 &ll_md_blocking_ast, 0);
639 ll_finish_md_op_data(op_data);
641 /* reason for keep own exit path - don`t flood log
642 * with messages with -ESTALE errors.
644 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
645 it_open_error(DISP_OPEN_OPEN, itp))
647 ll_release_openhandle(de, itp);
651 if (it_disposition(itp, DISP_LOOKUP_NEG))
652 GOTO(out, rc = -ENOENT);
654 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
655 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
656 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
660 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
662 if (!rc && itp->it_lock_mode) {
665 /* If we got a lock back and it has a LOOKUP bit set,
666 * make sure the dentry is marked as valid so we can find it.
667 * We don't need to care about actual hashing since other bits
668 * of kernel will deal with that later.
670 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
671 if (bits & MDS_INODELOCK_LOOKUP)
672 d_lustre_revalidate(de);
673 /* if DoM bit returned along with LAYOUT bit then there
674 * can be read-on-open data returned.
676 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
677 ll_dom_finish_open(de->d_inode, req);
681 ptlrpc_req_finished(req);
682 ll_intent_drop_lock(itp);
684 /* We did open by fid, but by the time we got to the server,
685 * the object disappeared. If this is a create, we cannot really
686 * tell the userspace that the file it was trying to create
687 * does not exist. Instead let's return -ESTALE, and the VFS will
688 * retry the create with LOOKUP_REVAL that we are going to catch
689 * in ll_revalidate_dentry() and use lookup then.
691 if (rc == -ENOENT && itp->it_op & IT_CREAT)
697 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
698 struct obd_client_handle *och)
700 struct mdt_body *body;
702 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
703 och->och_open_handle = body->mbo_open_handle;
704 och->och_fid = body->mbo_fid1;
705 och->och_lease_handle.cookie = it->it_lock_handle;
706 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
707 och->och_flags = it->it_flags;
709 return md_set_open_replay_data(md_exp, och, it);
712 static int ll_local_open(struct file *file, struct lookup_intent *it,
713 struct ll_file_data *fd, struct obd_client_handle *och)
715 struct inode *inode = file_inode(file);
718 LASSERT(!file->private_data);
725 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
730 file->private_data = fd;
731 ll_readahead_init(inode, &fd->fd_ras);
732 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
734 /* ll_cl_context initialize */
735 rwlock_init(&fd->fd_lock);
736 INIT_LIST_HEAD(&fd->fd_lccs);
741 /* Open a file, and (for the very first open) create objects on the OSTs at
742 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
743 * creation or open until ll_lov_setstripe() ioctl is called.
745 * If we already have the stripe MD locally then we don't request it in
746 * md_open(), by passing a lmm_size = 0.
748 * It is up to the application to ensure no other processes open this file
749 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
750 * used. We might be able to avoid races of that sort by getting lli_open_sem
751 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
752 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
754 int ll_file_open(struct inode *inode, struct file *file)
756 struct ll_inode_info *lli = ll_i2info(inode);
757 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
758 .it_flags = file->f_flags };
759 struct obd_client_handle **och_p = NULL;
760 __u64 *och_usecount = NULL;
761 struct ll_file_data *fd;
762 ktime_t kstart = ktime_get();
766 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
767 PFID(ll_inode2fid(inode)), inode, file->f_flags);
769 it = file->private_data; /* XXX: compat macro */
770 file->private_data = NULL; /* prevent ll_local_open assertion */
772 if (S_ISREG(inode->i_mode)) {
773 rc = llcrypt_file_open(inode, file);
775 GOTO(out_nofiledata, rc);
778 fd = ll_file_data_get();
780 GOTO(out_nofiledata, rc = -ENOMEM);
783 if (S_ISDIR(inode->i_mode))
784 ll_authorize_statahead(inode, fd);
786 if (inode->i_sb->s_root == file_dentry(file)) {
787 file->private_data = fd;
791 if (!it || !it->it_disposition) {
792 /* Convert f_flags into access mode. We cannot use file->f_mode,
793 * because everything but O_ACCMODE mask was stripped from
795 if ((oit.it_flags + 1) & O_ACCMODE)
797 if (file->f_flags & O_TRUNC)
798 oit.it_flags |= FMODE_WRITE;
800 /* kernel only call f_op->open in dentry_open. filp_open calls
801 * dentry_open after call to open_namei that checks permissions.
802 * Only nfsd_open call dentry_open directly without checking
803 * permissions and because of that this code below is safe.
805 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
806 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
808 /* We do not want O_EXCL here, presumably we opened the file
809 * already? XXX - NFS implications? */
810 oit.it_flags &= ~O_EXCL;
812 /* bug20584, if "it_flags" contains O_CREAT, the file will be
813 * created if necessary, then "IT_CREAT" should be set to keep
814 * consistent with it */
815 if (oit.it_flags & O_CREAT)
816 oit.it_op |= IT_CREAT;
822 /* Let's see if we have file open on MDS already. */
823 if (it->it_flags & FMODE_WRITE) {
824 och_p = &lli->lli_mds_write_och;
825 och_usecount = &lli->lli_open_fd_write_count;
826 } else if (it->it_flags & FMODE_EXEC) {
827 och_p = &lli->lli_mds_exec_och;
828 och_usecount = &lli->lli_open_fd_exec_count;
830 och_p = &lli->lli_mds_read_och;
831 och_usecount = &lli->lli_open_fd_read_count;
834 mutex_lock(&lli->lli_och_mutex);
835 if (*och_p) { /* Open handle is present */
836 if (it_disposition(it, DISP_OPEN_OPEN)) {
837 /* Well, there's extra open request that we do not need,
838 * let's close it somehow. This will decref request. */
839 rc = it_open_error(DISP_OPEN_OPEN, it);
841 mutex_unlock(&lli->lli_och_mutex);
842 GOTO(out_openerr, rc);
845 ll_release_openhandle(file_dentry(file), it);
849 rc = ll_local_open(file, it, fd, NULL);
852 mutex_unlock(&lli->lli_och_mutex);
853 GOTO(out_openerr, rc);
856 LASSERT(*och_usecount == 0);
857 if (!it->it_disposition) {
858 struct dentry *dentry = file_dentry(file);
859 struct ll_dentry_data *ldd;
861 /* We cannot just request lock handle now, new ELC code
862 * means that one of other OPEN locks for this file
863 * could be cancelled, and since blocking ast handler
864 * would attempt to grab och_mutex as well, that would
865 * result in a deadlock
867 mutex_unlock(&lli->lli_och_mutex);
869 * Normally called under two situations:
871 * 2. A race/condition on MDS resulting in no open
872 * handle to be returned from LOOKUP|OPEN request,
873 * for example if the target entry was a symlink.
875 * Only fetch MDS_OPEN_LOCK if this is in NFS path,
876 * marked by a bit set in ll_iget_for_nfs. Clear the
877 * bit so that it's not confusing later callers.
879 * NB; when ldd is NULL, it must have come via normal
880 * lookup path only, since ll_iget_for_nfs always calls
883 ldd = ll_d2d(dentry);
884 if (ldd && ldd->lld_nfs_dentry) {
885 ldd->lld_nfs_dentry = 0;
886 if (!filename_is_volatile(dentry->d_name.name,
889 it->it_flags |= MDS_OPEN_LOCK;
893 * Always specify MDS_OPEN_BY_FID because we don't want
894 * to get file with different fid.
896 it->it_flags |= MDS_OPEN_BY_FID;
897 rc = ll_intent_file_open(dentry, NULL, 0, it);
899 GOTO(out_openerr, rc);
903 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
905 GOTO(out_och_free, rc = -ENOMEM);
909 /* md_intent_lock() didn't get a request ref if there was an
910 * open error, so don't do cleanup on the request here
912 /* XXX (green): Should not we bail out on any error here, not
913 * just open error? */
914 rc = it_open_error(DISP_OPEN_OPEN, it);
916 GOTO(out_och_free, rc);
918 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
919 "inode %p: disposition %x, status %d\n", inode,
920 it_disposition(it, ~0), it->it_status);
922 rc = ll_local_open(file, it, fd, *och_p);
924 GOTO(out_och_free, rc);
927 rc = pcc_file_open(inode, file);
929 GOTO(out_och_free, rc);
931 mutex_unlock(&lli->lli_och_mutex);
933 /* lockless for direct IO so that it can do IO in parallel */
934 if (file->f_flags & O_DIRECT)
935 fd->fd_flags |= LL_FILE_LOCKLESS_IO;
938 /* Must do this outside lli_och_mutex lock to prevent deadlock where
939 different kind of OPEN lock for this same inode gets cancelled
940 by ldlm_cancel_lru */
941 if (!S_ISREG(inode->i_mode))
942 GOTO(out_och_free, rc);
943 cl_lov_delay_create_clear(&file->f_flags);
944 GOTO(out_och_free, rc);
948 if (och_p && *och_p) {
949 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
950 *och_p = NULL; /* OBD_FREE writes some magic there */
953 mutex_unlock(&lli->lli_och_mutex);
956 if (lli->lli_opendir_key == fd)
957 ll_deauthorize_statahead(inode, fd);
960 ll_file_data_put(fd);
962 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
963 ktime_us_delta(ktime_get(), kstart));
967 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
968 ptlrpc_req_finished(it->it_request);
969 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
975 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
976 struct ldlm_lock_desc *desc, void *data, int flag)
979 struct lustre_handle lockh;
983 case LDLM_CB_BLOCKING:
984 ldlm_lock2handle(lock, &lockh);
985 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
987 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
991 case LDLM_CB_CANCELING:
999 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1000 * and save it as fd->fd_och so as to force client to reopen the file even
1001 * if it has an open lock in cache already.
1003 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1004 struct lustre_handle *old_open_handle)
1006 struct ll_inode_info *lli = ll_i2info(inode);
1007 struct ll_file_data *fd = file->private_data;
1008 struct obd_client_handle **och_p;
1009 __u64 *och_usecount;
1013 /* Get the openhandle of the file */
1014 mutex_lock(&lli->lli_och_mutex);
1015 if (fd->fd_lease_och != NULL)
1016 GOTO(out_unlock, rc = -EBUSY);
1018 if (fd->fd_och == NULL) {
1019 if (file->f_mode & FMODE_WRITE) {
1020 LASSERT(lli->lli_mds_write_och != NULL);
1021 och_p = &lli->lli_mds_write_och;
1022 och_usecount = &lli->lli_open_fd_write_count;
1024 LASSERT(lli->lli_mds_read_och != NULL);
1025 och_p = &lli->lli_mds_read_och;
1026 och_usecount = &lli->lli_open_fd_read_count;
1029 if (*och_usecount > 1)
1030 GOTO(out_unlock, rc = -EBUSY);
1032 fd->fd_och = *och_p;
1037 *old_open_handle = fd->fd_och->och_open_handle;
1041 mutex_unlock(&lli->lli_och_mutex);
1046 * Release ownership on lli_mds_*_och when putting back a file lease.
1048 static int ll_lease_och_release(struct inode *inode, struct file *file)
1050 struct ll_inode_info *lli = ll_i2info(inode);
1051 struct ll_file_data *fd = file->private_data;
1052 struct obd_client_handle **och_p;
1053 struct obd_client_handle *old_och = NULL;
1054 __u64 *och_usecount;
1058 mutex_lock(&lli->lli_och_mutex);
1059 if (file->f_mode & FMODE_WRITE) {
1060 och_p = &lli->lli_mds_write_och;
1061 och_usecount = &lli->lli_open_fd_write_count;
1063 och_p = &lli->lli_mds_read_och;
1064 och_usecount = &lli->lli_open_fd_read_count;
1067 /* The file may have been open by another process (broken lease) so
1068 * *och_p is not NULL. In this case we should simply increase usecount
1071 if (*och_p != NULL) {
1072 old_och = fd->fd_och;
1075 *och_p = fd->fd_och;
1079 mutex_unlock(&lli->lli_och_mutex);
1081 if (old_och != NULL)
1082 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1088 * Acquire a lease and open the file.
1090 static struct obd_client_handle *
1091 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1094 struct lookup_intent it = { .it_op = IT_OPEN };
1095 struct ll_sb_info *sbi = ll_i2sbi(inode);
1096 struct md_op_data *op_data;
1097 struct ptlrpc_request *req = NULL;
1098 struct lustre_handle old_open_handle = { 0 };
1099 struct obd_client_handle *och = NULL;
1104 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1105 RETURN(ERR_PTR(-EINVAL));
1108 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1109 RETURN(ERR_PTR(-EPERM));
1111 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1113 RETURN(ERR_PTR(rc));
1118 RETURN(ERR_PTR(-ENOMEM));
1120 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1121 LUSTRE_OPC_ANY, NULL);
1122 if (IS_ERR(op_data))
1123 GOTO(out, rc = PTR_ERR(op_data));
1125 /* To tell the MDT this openhandle is from the same owner */
1126 op_data->op_open_handle = old_open_handle;
1128 it.it_flags = fmode | open_flags;
1129 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1130 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1131 &ll_md_blocking_lease_ast,
1132 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1133 * it can be cancelled which may mislead applications that the lease is
1135 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1136 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1137 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1138 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1139 ll_finish_md_op_data(op_data);
1140 ptlrpc_req_finished(req);
1142 GOTO(out_release_it, rc);
1144 if (it_disposition(&it, DISP_LOOKUP_NEG))
1145 GOTO(out_release_it, rc = -ENOENT);
1147 rc = it_open_error(DISP_OPEN_OPEN, &it);
1149 GOTO(out_release_it, rc);
1151 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1152 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1154 GOTO(out_release_it, rc);
1156 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1157 GOTO(out_close, rc = -EOPNOTSUPP);
1159 /* already get lease, handle lease lock */
1160 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1161 if (!it.it_lock_mode ||
1162 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1163 /* open lock must return for lease */
1164 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1165 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1167 GOTO(out_close, rc = -EPROTO);
1170 ll_intent_release(&it);
1174 /* Cancel open lock */
1175 if (it.it_lock_mode != 0) {
1176 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1178 it.it_lock_mode = 0;
1179 och->och_lease_handle.cookie = 0ULL;
1181 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1183 CERROR("%s: error closing file "DFID": %d\n",
1184 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1185 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1187 ll_intent_release(&it);
1191 RETURN(ERR_PTR(rc));
1195 * Check whether a layout swap can be done between two inodes.
1197 * \param[in] inode1 First inode to check
1198 * \param[in] inode2 Second inode to check
1200 * \retval 0 on success, layout swap can be performed between both inodes
1201 * \retval negative error code if requirements are not met
1203 static int ll_check_swap_layouts_validity(struct inode *inode1,
1204 struct inode *inode2)
1206 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1209 if (inode_permission(inode1, MAY_WRITE) ||
1210 inode_permission(inode2, MAY_WRITE))
1213 if (inode1->i_sb != inode2->i_sb)
1219 static int ll_swap_layouts_close(struct obd_client_handle *och,
1220 struct inode *inode, struct inode *inode2)
1222 const struct lu_fid *fid1 = ll_inode2fid(inode);
1223 const struct lu_fid *fid2;
1227 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1228 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1230 rc = ll_check_swap_layouts_validity(inode, inode2);
1232 GOTO(out_free_och, rc);
1234 /* We now know that inode2 is a lustre inode */
1235 fid2 = ll_inode2fid(inode2);
1237 rc = lu_fid_cmp(fid1, fid2);
1239 GOTO(out_free_och, rc = -EINVAL);
1241 /* Close the file and {swap,merge} layouts between inode & inode2.
1242 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1243 * because we still need it to pack l_remote_handle to MDT. */
1244 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1247 och = NULL; /* freed in ll_close_inode_openhandle() */
1257 * Release lease and close the file.
1258 * It will check if the lease has ever broken.
1260 static int ll_lease_close_intent(struct obd_client_handle *och,
1261 struct inode *inode,
1262 bool *lease_broken, enum mds_op_bias bias,
1265 struct ldlm_lock *lock;
1266 bool cancelled = true;
1270 lock = ldlm_handle2lock(&och->och_lease_handle);
1272 lock_res_and_lock(lock);
1273 cancelled = ldlm_is_cancel(lock);
1274 unlock_res_and_lock(lock);
1275 LDLM_LOCK_PUT(lock);
1278 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1279 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1281 if (lease_broken != NULL)
1282 *lease_broken = cancelled;
1284 if (!cancelled && !bias)
1285 ldlm_cli_cancel(&och->och_lease_handle, 0);
1287 if (cancelled) { /* no need to excute intent */
1292 rc = ll_close_inode_openhandle(inode, och, bias, data);
1296 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1299 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1303 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1305 static int ll_lease_file_resync(struct obd_client_handle *och,
1306 struct inode *inode, unsigned long arg)
1308 struct ll_sb_info *sbi = ll_i2sbi(inode);
1309 struct md_op_data *op_data;
1310 struct ll_ioc_lease_id ioc;
1311 __u64 data_version_unused;
1315 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1316 LUSTRE_OPC_ANY, NULL);
1317 if (IS_ERR(op_data))
1318 RETURN(PTR_ERR(op_data));
1320 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1324 /* before starting file resync, it's necessary to clean up page cache
1325 * in client memory, otherwise once the layout version is increased,
1326 * writing back cached data will be denied the OSTs. */
1327 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1331 op_data->op_lease_handle = och->och_lease_handle;
1332 op_data->op_mirror_id = ioc.lil_mirror_id;
1333 rc = md_file_resync(sbi->ll_md_exp, op_data);
1339 ll_finish_md_op_data(op_data);
1343 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1345 struct ll_inode_info *lli = ll_i2info(inode);
1346 struct cl_object *obj = lli->lli_clob;
1347 struct cl_attr *attr = vvp_env_thread_attr(env);
1355 ll_inode_size_lock(inode);
1357 /* Merge timestamps the most recently obtained from MDS with
1358 * timestamps obtained from OSTs.
1360 * Do not overwrite atime of inode because it may be refreshed
1361 * by file_accessed() function. If the read was served by cache
1362 * data, there is no RPC to be sent so that atime may not be
1363 * transferred to OSTs at all. MDT only updates atime at close time
1364 * if it's at least 'mdd.*.atime_diff' older.
1365 * All in all, the atime in Lustre does not strictly comply with
1366 * POSIX. Solving this problem needs to send an RPC to MDT for each
1367 * read, this will hurt performance.
1369 if (ll_file_test_and_clear_flag(lli, LLIF_UPDATE_ATIME) ||
1370 inode->i_atime.tv_sec < lli->lli_atime)
1371 inode->i_atime.tv_sec = lli->lli_atime;
1373 inode->i_mtime.tv_sec = lli->lli_mtime;
1374 inode->i_ctime.tv_sec = lli->lli_ctime;
1376 mtime = inode->i_mtime.tv_sec;
1377 atime = inode->i_atime.tv_sec;
1378 ctime = inode->i_ctime.tv_sec;
1380 cl_object_attr_lock(obj);
1381 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1384 rc = cl_object_attr_get(env, obj, attr);
1385 cl_object_attr_unlock(obj);
1388 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1390 if (atime < attr->cat_atime)
1391 atime = attr->cat_atime;
1393 if (ctime < attr->cat_ctime)
1394 ctime = attr->cat_ctime;
1396 if (mtime < attr->cat_mtime)
1397 mtime = attr->cat_mtime;
1399 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1400 PFID(&lli->lli_fid), attr->cat_size);
1402 i_size_write(inode, attr->cat_size);
1403 inode->i_blocks = attr->cat_blocks;
1405 inode->i_mtime.tv_sec = mtime;
1406 inode->i_atime.tv_sec = atime;
1407 inode->i_ctime.tv_sec = ctime;
1410 ll_inode_size_unlock(inode);
1416 * Set designated mirror for I/O.
1418 * So far only read, write, and truncated can support to issue I/O to
1419 * designated mirror.
1421 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1423 struct ll_file_data *fd = file->private_data;
1425 /* clear layout version for generic(non-resync) I/O in case it carries
1426 * stale layout version due to I/O restart */
1427 io->ci_layout_version = 0;
1429 /* FLR: disable non-delay for designated mirror I/O because obviously
1430 * only one mirror is available */
1431 if (fd->fd_designated_mirror > 0) {
1433 io->ci_designated_mirror = fd->fd_designated_mirror;
1434 io->ci_layout_version = fd->fd_layout_version;
1437 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1438 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1441 static bool file_is_noatime(const struct file *file)
1443 const struct vfsmount *mnt = file->f_path.mnt;
1444 const struct inode *inode = file_inode((struct file *)file);
1446 /* Adapted from file_accessed() and touch_atime().*/
1447 if (file->f_flags & O_NOATIME)
1450 if (inode->i_flags & S_NOATIME)
1453 if (IS_NOATIME(inode))
1456 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1459 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1462 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1468 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1469 struct vvp_io_args *args)
1471 struct inode *inode = file_inode(file);
1472 struct ll_file_data *fd = file->private_data;
1474 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1475 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1477 if (iot == CIT_WRITE) {
1478 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1479 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1480 file->f_flags & O_DIRECT ||
1482 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1483 io->u.ci_wr.wr_sync |= !!(args &&
1484 (args->u.normal.via_iocb->ki_flags &
1489 io->ci_obj = ll_i2info(inode)->lli_clob;
1490 io->ci_lockreq = CILR_MAYBE;
1491 if (ll_file_nolock(file)) {
1492 io->ci_lockreq = CILR_NEVER;
1493 io->ci_no_srvlock = 1;
1494 } else if (file->f_flags & O_APPEND) {
1495 io->ci_lockreq = CILR_MANDATORY;
1497 io->ci_noatime = file_is_noatime(file);
1498 io->ci_async_readahead = false;
1500 /* FLR: only use non-delay I/O for read as there is only one
1501 * avaliable mirror for write. */
1502 io->ci_ndelay = !(iot == CIT_WRITE);
1504 ll_io_set_mirror(io, file);
1507 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1510 struct ll_inode_info *lli = ll_i2info(inode);
1511 struct ll_sb_info *sbi = ll_i2sbi(inode);
1512 enum obd_heat_type sample_type;
1513 enum obd_heat_type iobyte_type;
1514 __u64 now = ktime_get_real_seconds();
1516 if (!ll_sbi_has_file_heat(sbi) ||
1517 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1520 if (iot == CIT_READ) {
1521 sample_type = OBD_HEAT_READSAMPLE;
1522 iobyte_type = OBD_HEAT_READBYTE;
1523 } else if (iot == CIT_WRITE) {
1524 sample_type = OBD_HEAT_WRITESAMPLE;
1525 iobyte_type = OBD_HEAT_WRITEBYTE;
1530 spin_lock(&lli->lli_heat_lock);
1531 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1532 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1533 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1534 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1535 spin_unlock(&lli->lli_heat_lock);
1539 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1540 struct file *file, enum cl_io_type iot,
1541 loff_t *ppos, size_t count)
1543 struct vvp_io *vio = vvp_env_io(env);
1544 struct inode *inode = file_inode(file);
1545 struct ll_inode_info *lli = ll_i2info(inode);
1546 struct ll_file_data *fd = file->private_data;
1547 struct range_lock range;
1551 unsigned int retried = 0, ignore_lockless = 0;
1552 bool is_aio = false;
1553 struct cl_dio_aio *ci_aio = NULL;
1557 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1558 file_dentry(file)->d_name.name,
1559 iot == CIT_READ ? "read" : "write", *ppos, count);
1561 io = vvp_env_thread_io(env);
1562 if (file->f_flags & O_DIRECT) {
1563 if (!is_sync_kiocb(args->u.normal.via_iocb))
1565 ci_aio = cl_aio_alloc(args->u.normal.via_iocb);
1567 GOTO(out, rc = -ENOMEM);
1571 io = vvp_env_thread_io(env);
1572 ll_io_init(io, file, iot, args);
1573 io->ci_aio = ci_aio;
1574 io->ci_ignore_lockless = ignore_lockless;
1575 io->ci_ndelay_tried = retried;
1577 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1578 bool range_locked = false;
1580 if (file->f_flags & O_APPEND)
1581 range_lock_init(&range, 0, LUSTRE_EOF);
1583 range_lock_init(&range, *ppos, *ppos + count - 1);
1585 vio->vui_fd = file->private_data;
1586 vio->vui_iter = args->u.normal.via_iter;
1587 vio->vui_iocb = args->u.normal.via_iocb;
1588 /* Direct IO reads must also take range lock,
1589 * or multiple reads will try to work on the same pages
1590 * See LU-6227 for details.
1592 if (((iot == CIT_WRITE) ||
1593 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1594 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1595 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1597 rc = range_lock(&lli->lli_write_tree, &range);
1601 range_locked = true;
1604 ll_cl_add(file, env, io, LCC_RW);
1605 rc = cl_io_loop(env, io);
1606 ll_cl_remove(file, env);
1609 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1611 range_unlock(&lli->lli_write_tree, &range);
1614 /* cl_io_rw_init() handled IO */
1619 * In order to move forward AIO, ci_nob was increased,
1620 * but that doesn't mean io have been finished, it just
1621 * means io have been submited, we will always return
1622 * EIOCBQUEUED to the caller, So we could only return
1623 * number of bytes in non-AIO case.
1625 if (io->ci_nob > 0) {
1627 result += io->ci_nob;
1628 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1630 count -= io->ci_nob;
1632 /* prepare IO restart */
1634 args->u.normal.via_iter = vio->vui_iter;
1637 cl_io_fini(env, io);
1640 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1641 file->f_path.dentry->d_name.name,
1642 iot, rc, result, io->ci_need_restart);
1644 if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
1645 count > 0 && io->ci_need_restart) {
1647 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1648 file_dentry(file)->d_name.name,
1649 iot == CIT_READ ? "read" : "write",
1650 *ppos, count, result, rc);
1651 /* preserve the tried count for FLR */
1652 retried = io->ci_ndelay_tried;
1653 ignore_lockless = io->ci_ignore_lockless;
1659 * VFS will call aio_complete() if no -EIOCBQUEUED
1660 * is returned for AIO, so we can not call aio_complete()
1663 if (rc != -EIOCBQUEUED)
1664 io->ci_aio->cda_no_aio_complete = 1;
1666 * Drop one extra reference so that end_io() could be
1667 * called for this IO context, we could call it after
1668 * we make sure all AIO requests have been proceed.
1670 cl_sync_io_note(env, &io->ci_aio->cda_sync,
1671 rc == -EIOCBQUEUED ? 0 : rc);
1673 cl_aio_free(io->ci_aio);
1678 if (iot == CIT_READ) {
1680 ll_stats_ops_tally(ll_i2sbi(inode),
1681 LPROC_LL_READ_BYTES, result);
1682 } else if (iot == CIT_WRITE) {
1684 ll_stats_ops_tally(ll_i2sbi(inode),
1685 LPROC_LL_WRITE_BYTES, result);
1686 fd->fd_write_failed = false;
1687 } else if (result == 0 && rc == 0) {
1690 fd->fd_write_failed = true;
1692 fd->fd_write_failed = false;
1693 } else if (rc != -ERESTARTSYS) {
1694 fd->fd_write_failed = true;
1698 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1700 ll_heat_add(inode, iot, result);
1702 RETURN(result > 0 ? result : rc);
1706 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1707 * especially for small I/O.
1709 * To serve a read request, CLIO has to create and initialize a cl_io and
1710 * then request DLM lock. This has turned out to have siginificant overhead
1711 * and affects the performance of small I/O dramatically.
1713 * It's not necessary to create a cl_io for each I/O. Under the help of read
1714 * ahead, most of the pages being read are already in memory cache and we can
1715 * read those pages directly because if the pages exist, the corresponding DLM
1716 * lock must exist so that page content must be valid.
1718 * In fast read implementation, the llite speculatively finds and reads pages
1719 * in memory cache. There are three scenarios for fast read:
1720 * - If the page exists and is uptodate, kernel VM will provide the data and
1721 * CLIO won't be intervened;
1722 * - If the page was brought into memory by read ahead, it will be exported
1723 * and read ahead parameters will be updated;
1724 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1725 * it will go back and invoke normal read, i.e., a cl_io will be created
1726 * and DLM lock will be requested.
1728 * POSIX compliance: posix standard states that read is intended to be atomic.
1729 * Lustre read implementation is in line with Linux kernel read implementation
1730 * and neither of them complies with POSIX standard in this matter. Fast read
1731 * doesn't make the situation worse on single node but it may interleave write
1732 * results from multiple nodes due to short read handling in ll_file_aio_read().
1734 * \param env - lu_env
1735 * \param iocb - kiocb from kernel
1736 * \param iter - user space buffers where the data will be copied
1738 * \retval - number of bytes have been read, or error code if error occurred.
1741 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1745 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1748 /* NB: we can't do direct IO for fast read because it will need a lock
1749 * to make IO engine happy. */
1750 if (iocb->ki_filp->f_flags & O_DIRECT)
1753 result = generic_file_read_iter(iocb, iter);
1755 /* If the first page is not in cache, generic_file_aio_read() will be
1756 * returned with -ENODATA.
1757 * See corresponding code in ll_readpage(). */
1758 if (result == -ENODATA)
1762 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1763 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1764 LPROC_LL_READ_BYTES, result);
1771 * Read from a file (through the page cache).
1773 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1776 struct vvp_io_args *args;
1777 struct file *file = iocb->ki_filp;
1781 ktime_t kstart = ktime_get();
1784 if (!iov_iter_count(to))
1788 * Currently when PCC read failed, we do not fall back to the
1789 * normal read path, just return the error.
1790 * The resaon is that: for RW-PCC, the file data may be modified
1791 * in the PCC and inconsistent with the data on OSTs (or file
1792 * data has been removed from the Lustre file system), at this
1793 * time, fallback to the normal read path may read the wrong
1795 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1796 * path: read data from data copy on OSTs.
1798 result = pcc_file_read_iter(iocb, to, &cached);
1802 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
1804 result = ll_do_fast_read(iocb, to);
1805 if (result < 0 || iov_iter_count(to) == 0)
1808 env = cl_env_get(&refcheck);
1810 return PTR_ERR(env);
1812 args = ll_env_args(env);
1813 args->u.normal.via_iter = to;
1814 args->u.normal.via_iocb = iocb;
1816 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
1817 &iocb->ki_pos, iov_iter_count(to));
1820 else if (result == 0)
1823 cl_env_put(env, &refcheck);
1826 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1827 file->private_data, iocb->ki_pos, result,
1829 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
1830 ktime_us_delta(ktime_get(), kstart));
1837 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
1838 * If a page is already in the page cache and dirty (and some other things -
1839 * See ll_tiny_write_begin for the instantiation of these rules), then we can
1840 * write to it without doing a full I/O, because Lustre already knows about it
1841 * and will write it out. This saves a lot of processing time.
1843 * All writes here are within one page, so exclusion is handled by the page
1844 * lock on the vm page. We do not do tiny writes for writes which touch
1845 * multiple pages because it's very unlikely multiple sequential pages are
1846 * are already dirty.
1848 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
1849 * and are unlikely to be to already dirty pages.
1851 * Attribute updates are important here, we do them in ll_tiny_write_end.
1853 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
1855 ssize_t count = iov_iter_count(iter);
1856 struct file *file = iocb->ki_filp;
1857 struct inode *inode = file_inode(file);
1858 bool lock_inode = !IS_NOSEC(inode);
1863 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
1864 * of function for why.
1866 if (count >= PAGE_SIZE ||
1867 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
1870 if (unlikely(lock_inode))
1872 result = __generic_file_write_iter(iocb, iter);
1874 if (unlikely(lock_inode))
1875 inode_unlock(inode);
1877 /* If the page is not already dirty, ll_tiny_write_begin returns
1878 * -ENODATA. We continue on to normal write.
1880 if (result == -ENODATA)
1884 ll_heat_add(inode, CIT_WRITE, result);
1885 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
1887 ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED);
1890 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
1896 * Write to a file (through the page cache).
1898 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1900 struct vvp_io_args *args;
1902 ssize_t rc_tiny = 0, rc_normal;
1903 struct file *file = iocb->ki_filp;
1906 ktime_t kstart = ktime_get();
1911 if (!iov_iter_count(from))
1912 GOTO(out, rc_normal = 0);
1915 * When PCC write failed, we usually do not fall back to the normal
1916 * write path, just return the error. But there is a special case when
1917 * returned error code is -ENOSPC due to running out of space on PCC HSM
1918 * bakcend. At this time, it will fall back to normal I/O path and
1919 * retry the I/O. As the file is in HSM released state, it will restore
1920 * the file data to OSTs first and redo the write again. And the
1921 * restore process will revoke the layout lock and detach the file
1922 * from PCC cache automatically.
1924 result = pcc_file_write_iter(iocb, from, &cached);
1925 if (cached && result != -ENOSPC && result != -EDQUOT)
1926 GOTO(out, rc_normal = result);
1928 /* NB: we can't do direct IO for tiny writes because they use the page
1929 * cache, we can't do sync writes because tiny writes can't flush
1930 * pages, and we can't do append writes because we can't guarantee the
1931 * required DLM locks are held to protect file size.
1933 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
1934 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
1935 rc_tiny = ll_do_tiny_write(iocb, from);
1937 /* In case of error, go on and try normal write - Only stop if tiny
1938 * write completed I/O.
1940 if (iov_iter_count(from) == 0)
1941 GOTO(out, rc_normal = rc_tiny);
1943 env = cl_env_get(&refcheck);
1945 return PTR_ERR(env);
1947 args = ll_env_args(env);
1948 args->u.normal.via_iter = from;
1949 args->u.normal.via_iocb = iocb;
1951 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
1952 &iocb->ki_pos, iov_iter_count(from));
1954 /* On success, combine bytes written. */
1955 if (rc_tiny >= 0 && rc_normal > 0)
1956 rc_normal += rc_tiny;
1957 /* On error, only return error from normal write if tiny write did not
1958 * write any bytes. Otherwise return bytes written by tiny write.
1960 else if (rc_tiny > 0)
1961 rc_normal = rc_tiny;
1963 cl_env_put(env, &refcheck);
1965 if (rc_normal > 0) {
1966 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1967 file->private_data, iocb->ki_pos,
1969 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
1970 ktime_us_delta(ktime_get(), kstart));
1976 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1978 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1980 static int ll_file_get_iov_count(const struct iovec *iov,
1981 unsigned long *nr_segs, size_t *count,
1987 for (seg = 0; seg < *nr_segs; seg++) {
1988 const struct iovec *iv = &iov[seg];
1991 * If any segment has a negative length, or the cumulative
1992 * length ever wraps negative then return -EINVAL.
1995 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1997 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2002 cnt -= iv->iov_len; /* This segment is no good */
2009 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2010 unsigned long nr_segs, loff_t pos)
2017 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2024 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2025 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2026 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2027 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2028 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2030 result = ll_file_read_iter(iocb, &to);
2035 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2038 struct iovec iov = { .iov_base = buf, .iov_len = count };
2047 init_sync_kiocb(&kiocb, file);
2048 kiocb.ki_pos = *ppos;
2049 #ifdef HAVE_KIOCB_KI_LEFT
2050 kiocb.ki_left = count;
2051 #elif defined(HAVE_KI_NBYTES)
2052 kiocb.i_nbytes = count;
2055 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2056 *ppos = kiocb.ki_pos;
2062 * Write to a file (through the page cache).
2065 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2066 unsigned long nr_segs, loff_t pos)
2068 struct iov_iter from;
2073 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2080 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2081 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2082 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2083 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2084 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2086 result = ll_file_write_iter(iocb, &from);
2091 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2092 size_t count, loff_t *ppos)
2094 struct iovec iov = { .iov_base = (void __user *)buf,
2104 init_sync_kiocb(&kiocb, file);
2105 kiocb.ki_pos = *ppos;
2106 #ifdef HAVE_KIOCB_KI_LEFT
2107 kiocb.ki_left = count;
2108 #elif defined(HAVE_KI_NBYTES)
2109 kiocb.ki_nbytes = count;
2112 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2113 *ppos = kiocb.ki_pos;
2117 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2119 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2120 __u64 flags, struct lov_user_md *lum, int lum_size)
2122 struct lookup_intent oit = {
2124 .it_flags = flags | MDS_OPEN_BY_FID,
2129 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2130 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2131 /* this code will only exist for big-endian systems */
2132 lustre_swab_lov_user_md(lum, 0);
2135 ll_inode_size_lock(inode);
2136 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2138 GOTO(out_unlock, rc);
2140 ll_release_openhandle(dentry, &oit);
2143 ll_inode_size_unlock(inode);
2144 ll_intent_release(&oit);
2149 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2150 struct lov_mds_md **lmmp, int *lmm_size,
2151 struct ptlrpc_request **request)
2153 struct ll_sb_info *sbi = ll_i2sbi(inode);
2154 struct mdt_body *body;
2155 struct lov_mds_md *lmm = NULL;
2156 struct ptlrpc_request *req = NULL;
2157 struct md_op_data *op_data;
2160 rc = ll_get_default_mdsize(sbi, &lmmsize);
2164 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2165 strlen(filename), lmmsize,
2166 LUSTRE_OPC_ANY, NULL);
2167 if (IS_ERR(op_data))
2168 RETURN(PTR_ERR(op_data));
2170 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2171 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2172 ll_finish_md_op_data(op_data);
2174 CDEBUG(D_INFO, "md_getattr_name failed "
2175 "on %s: rc %d\n", filename, rc);
2179 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2180 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2182 lmmsize = body->mbo_eadatasize;
2184 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2186 GOTO(out, rc = -ENODATA);
2189 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2190 LASSERT(lmm != NULL);
2192 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2193 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2194 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2195 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2196 GOTO(out, rc = -EPROTO);
2199 * This is coming from the MDS, so is probably in
2200 * little endian. We convert it to host endian before
2201 * passing it to userspace.
2203 if ((lmm->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
2204 __swab32(LOV_MAGIC_MAGIC)) {
2205 int stripe_count = 0;
2207 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2208 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2209 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2210 if (le32_to_cpu(lmm->lmm_pattern) &
2211 LOV_PATTERN_F_RELEASED)
2215 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2217 /* if function called for directory - we should
2218 * avoid swab not existent lsm objects */
2219 if (lmm->lmm_magic == LOV_MAGIC_V1 && S_ISREG(body->mbo_mode))
2220 lustre_swab_lov_user_md_objects(
2221 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2223 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2224 S_ISREG(body->mbo_mode))
2225 lustre_swab_lov_user_md_objects(
2226 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2232 *lmm_size = lmmsize;
2237 static int ll_lov_setea(struct inode *inode, struct file *file,
2240 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2241 struct lov_user_md *lump;
2242 int lum_size = sizeof(struct lov_user_md) +
2243 sizeof(struct lov_user_ost_data);
2247 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2250 OBD_ALLOC_LARGE(lump, lum_size);
2254 if (copy_from_user(lump, arg, lum_size))
2255 GOTO(out_lump, rc = -EFAULT);
2257 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2259 cl_lov_delay_create_clear(&file->f_flags);
2262 OBD_FREE_LARGE(lump, lum_size);
2266 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2273 env = cl_env_get(&refcheck);
2275 RETURN(PTR_ERR(env));
2277 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2278 cl_env_put(env, &refcheck);
2282 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2285 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2286 struct lov_user_md *klum;
2288 __u64 flags = FMODE_WRITE;
2291 rc = ll_copy_user_md(lum, &klum);
2296 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2301 rc = put_user(0, &lum->lmm_stripe_count);
2305 rc = ll_layout_refresh(inode, &gen);
2309 rc = ll_file_getstripe(inode, arg, lum_size);
2310 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2311 ll_i2info(inode)->lli_clob) {
2312 struct iattr attr = { 0 };
2314 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2315 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2318 cl_lov_delay_create_clear(&file->f_flags);
2321 OBD_FREE_LARGE(klum, lum_size);
2327 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2329 struct ll_inode_info *lli = ll_i2info(inode);
2330 struct cl_object *obj = lli->lli_clob;
2331 struct ll_file_data *fd = file->private_data;
2332 struct ll_grouplock grouplock;
2337 CWARN("group id for group lock must not be 0\n");
2341 if (ll_file_nolock(file))
2342 RETURN(-EOPNOTSUPP);
2344 if (file->f_flags & O_NONBLOCK) {
2345 if (!mutex_trylock(&lli->lli_group_mutex))
2348 mutex_lock(&lli->lli_group_mutex);
2350 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2351 CWARN("group lock already existed with gid %lu\n",
2352 fd->fd_grouplock.lg_gid);
2353 GOTO(out, rc = -EINVAL);
2355 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2356 if (file->f_flags & O_NONBLOCK)
2357 GOTO(out, rc = -EAGAIN);
2358 mutex_unlock(&lli->lli_group_mutex);
2359 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2360 GOTO(retry, rc = 0);
2362 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2365 * XXX: group lock needs to protect all OST objects while PFL
2366 * can add new OST objects during the IO, so we'd instantiate
2367 * all OST objects before getting its group lock.
2372 struct cl_layout cl = {
2373 .cl_is_composite = false,
2375 struct lu_extent ext = {
2377 .e_end = OBD_OBJECT_EOF,
2380 env = cl_env_get(&refcheck);
2382 GOTO(out, rc = PTR_ERR(env));
2384 rc = cl_object_layout_get(env, obj, &cl);
2385 if (!rc && cl.cl_is_composite)
2386 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2389 cl_env_put(env, &refcheck);
2394 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2395 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2400 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2401 fd->fd_grouplock = grouplock;
2402 if (lli->lli_group_users == 0)
2403 lli->lli_group_gid = grouplock.lg_gid;
2404 lli->lli_group_users++;
2406 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2408 mutex_unlock(&lli->lli_group_mutex);
2413 static int ll_put_grouplock(struct inode *inode, struct file *file,
2416 struct ll_inode_info *lli = ll_i2info(inode);
2417 struct ll_file_data *fd = file->private_data;
2418 struct ll_grouplock grouplock;
2422 mutex_lock(&lli->lli_group_mutex);
2423 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2424 CWARN("no group lock held\n");
2425 GOTO(out, rc = -EINVAL);
2428 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2430 if (fd->fd_grouplock.lg_gid != arg) {
2431 CWARN("group lock %lu doesn't match current id %lu\n",
2432 arg, fd->fd_grouplock.lg_gid);
2433 GOTO(out, rc = -EINVAL);
2436 grouplock = fd->fd_grouplock;
2437 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2438 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2440 cl_put_grouplock(&grouplock);
2442 lli->lli_group_users--;
2443 if (lli->lli_group_users == 0) {
2444 lli->lli_group_gid = 0;
2445 wake_up_var(&lli->lli_group_users);
2447 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2450 mutex_unlock(&lli->lli_group_mutex);
2456 * Close inode open handle
2458 * \param dentry [in] dentry which contains the inode
2459 * \param it [in,out] intent which contains open info and result
2462 * \retval <0 failure
2464 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2466 struct inode *inode = dentry->d_inode;
2467 struct obd_client_handle *och;
2473 /* Root ? Do nothing. */
2474 if (dentry->d_inode->i_sb->s_root == dentry)
2477 /* No open handle to close? Move away */
2478 if (!it_disposition(it, DISP_OPEN_OPEN))
2481 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2483 OBD_ALLOC(och, sizeof(*och));
2485 GOTO(out, rc = -ENOMEM);
2487 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2491 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2493 /* this one is in place of ll_file_open */
2494 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2495 ptlrpc_req_finished(it->it_request);
2496 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2502 * Get size for inode for which FIEMAP mapping is requested.
2503 * Make the FIEMAP get_info call and returns the result.
2504 * \param fiemap kernel buffer to hold extens
2505 * \param num_bytes kernel buffer size
2507 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2513 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2516 /* Checks for fiemap flags */
2517 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2518 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2522 /* Check for FIEMAP_FLAG_SYNC */
2523 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2524 rc = filemap_fdatawrite(inode->i_mapping);
2529 env = cl_env_get(&refcheck);
2531 RETURN(PTR_ERR(env));
2533 if (i_size_read(inode) == 0) {
2534 rc = ll_glimpse_size(inode);
2539 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2540 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2541 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2543 /* If filesize is 0, then there would be no objects for mapping */
2544 if (fmkey.lfik_oa.o_size == 0) {
2545 fiemap->fm_mapped_extents = 0;
2549 fmkey.lfik_fiemap = *fiemap;
2551 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2552 &fmkey, fiemap, &num_bytes);
2554 cl_env_put(env, &refcheck);
2558 int ll_fid2path(struct inode *inode, void __user *arg)
2560 struct obd_export *exp = ll_i2mdexp(inode);
2561 const struct getinfo_fid2path __user *gfin = arg;
2563 struct getinfo_fid2path *gfout;
2569 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2570 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2573 /* Only need to get the buflen */
2574 if (get_user(pathlen, &gfin->gf_pathlen))
2577 if (pathlen > PATH_MAX)
2580 outsize = sizeof(*gfout) + pathlen;
2581 OBD_ALLOC(gfout, outsize);
2585 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2586 GOTO(gf_free, rc = -EFAULT);
2587 /* append root FID after gfout to let MDT know the root FID so that it
2588 * can lookup the correct path, this is mainly for fileset.
2589 * old server without fileset mount support will ignore this. */
2590 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2592 /* Call mdc_iocontrol */
2593 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2597 if (copy_to_user(arg, gfout, outsize))
2601 OBD_FREE(gfout, outsize);
2606 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2608 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2616 ioc->idv_version = 0;
2617 ioc->idv_layout_version = UINT_MAX;
2619 /* If no file object initialized, we consider its version is 0. */
2623 env = cl_env_get(&refcheck);
2625 RETURN(PTR_ERR(env));
2627 io = vvp_env_thread_io(env);
2629 io->u.ci_data_version.dv_data_version = 0;
2630 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2631 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2634 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2635 result = cl_io_loop(env, io);
2637 result = io->ci_result;
2639 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2640 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2642 cl_io_fini(env, io);
2644 if (unlikely(io->ci_need_restart))
2647 cl_env_put(env, &refcheck);
2653 * Read the data_version for inode.
2655 * This value is computed using stripe object version on OST.
2656 * Version is computed using server side locking.
2658 * @param flags if do sync on the OST side;
2660 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2661 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2663 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2665 struct ioc_data_version ioc = { .idv_flags = flags };
2668 rc = ll_ioc_data_version(inode, &ioc);
2670 *data_version = ioc.idv_version;
2676 * Trigger a HSM release request for the provided inode.
2678 int ll_hsm_release(struct inode *inode)
2681 struct obd_client_handle *och = NULL;
2682 __u64 data_version = 0;
2687 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2688 ll_i2sbi(inode)->ll_fsname,
2689 PFID(&ll_i2info(inode)->lli_fid));
2691 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2693 GOTO(out, rc = PTR_ERR(och));
2695 /* Grab latest data_version and [am]time values */
2696 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2700 env = cl_env_get(&refcheck);
2702 GOTO(out, rc = PTR_ERR(env));
2704 rc = ll_merge_attr(env, inode);
2705 cl_env_put(env, &refcheck);
2707 /* If error happen, we have the wrong size for a file.
2713 /* Release the file.
2714 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2715 * we still need it to pack l_remote_handle to MDT. */
2716 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2722 if (och != NULL && !IS_ERR(och)) /* close the file */
2723 ll_lease_close(och, inode, NULL);
2728 struct ll_swap_stack {
2731 struct inode *inode1;
2732 struct inode *inode2;
2737 static int ll_swap_layouts(struct file *file1, struct file *file2,
2738 struct lustre_swap_layouts *lsl)
2740 struct mdc_swap_layouts msl;
2741 struct md_op_data *op_data;
2744 struct ll_swap_stack *llss = NULL;
2747 OBD_ALLOC_PTR(llss);
2751 llss->inode1 = file_inode(file1);
2752 llss->inode2 = file_inode(file2);
2754 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2758 /* we use 2 bool because it is easier to swap than 2 bits */
2759 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2760 llss->check_dv1 = true;
2762 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2763 llss->check_dv2 = true;
2765 /* we cannot use lsl->sl_dvX directly because we may swap them */
2766 llss->dv1 = lsl->sl_dv1;
2767 llss->dv2 = lsl->sl_dv2;
2769 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2770 if (rc == 0) /* same file, done! */
2773 if (rc < 0) { /* sequentialize it */
2774 swap(llss->inode1, llss->inode2);
2776 swap(llss->dv1, llss->dv2);
2777 swap(llss->check_dv1, llss->check_dv2);
2781 if (gid != 0) { /* application asks to flush dirty cache */
2782 rc = ll_get_grouplock(llss->inode1, file1, gid);
2786 rc = ll_get_grouplock(llss->inode2, file2, gid);
2788 ll_put_grouplock(llss->inode1, file1, gid);
2793 /* ultimate check, before swaping the layouts we check if
2794 * dataversion has changed (if requested) */
2795 if (llss->check_dv1) {
2796 rc = ll_data_version(llss->inode1, &dv, 0);
2799 if (dv != llss->dv1)
2800 GOTO(putgl, rc = -EAGAIN);
2803 if (llss->check_dv2) {
2804 rc = ll_data_version(llss->inode2, &dv, 0);
2807 if (dv != llss->dv2)
2808 GOTO(putgl, rc = -EAGAIN);
2811 /* struct md_op_data is used to send the swap args to the mdt
2812 * only flags is missing, so we use struct mdc_swap_layouts
2813 * through the md_op_data->op_data */
2814 /* flags from user space have to be converted before they are send to
2815 * server, no flag is sent today, they are only used on the client */
2818 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2819 0, LUSTRE_OPC_ANY, &msl);
2820 if (IS_ERR(op_data))
2821 GOTO(free, rc = PTR_ERR(op_data));
2823 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2824 sizeof(*op_data), op_data, NULL);
2825 ll_finish_md_op_data(op_data);
2832 ll_put_grouplock(llss->inode2, file2, gid);
2833 ll_put_grouplock(llss->inode1, file1, gid);
2843 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2845 struct obd_export *exp = ll_i2mdexp(inode);
2846 struct md_op_data *op_data;
2850 /* Detect out-of range masks */
2851 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2854 /* Non-root users are forbidden to set or clear flags which are
2855 * NOT defined in HSM_USER_MASK. */
2856 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2857 !cfs_capable(CFS_CAP_SYS_ADMIN))
2860 if (!exp_connect_archive_id_array(exp)) {
2861 /* Detect out-of range archive id */
2862 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2863 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
2867 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2868 LUSTRE_OPC_ANY, hss);
2869 if (IS_ERR(op_data))
2870 RETURN(PTR_ERR(op_data));
2872 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
2875 ll_finish_md_op_data(op_data);
2880 static int ll_hsm_import(struct inode *inode, struct file *file,
2881 struct hsm_user_import *hui)
2883 struct hsm_state_set *hss = NULL;
2884 struct iattr *attr = NULL;
2888 if (!S_ISREG(inode->i_mode))
2894 GOTO(out, rc = -ENOMEM);
2896 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2897 hss->hss_archive_id = hui->hui_archive_id;
2898 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2899 rc = ll_hsm_state_set(inode, hss);
2903 OBD_ALLOC_PTR(attr);
2905 GOTO(out, rc = -ENOMEM);
2907 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2908 attr->ia_mode |= S_IFREG;
2909 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2910 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2911 attr->ia_size = hui->hui_size;
2912 attr->ia_mtime.tv_sec = hui->hui_mtime;
2913 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2914 attr->ia_atime.tv_sec = hui->hui_atime;
2915 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2917 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2918 ATTR_UID | ATTR_GID |
2919 ATTR_MTIME | ATTR_MTIME_SET |
2920 ATTR_ATIME | ATTR_ATIME_SET;
2924 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
2928 inode_unlock(inode);
2940 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2942 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2943 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2946 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
2948 struct inode *inode = file_inode(file);
2950 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
2951 ATTR_MTIME | ATTR_MTIME_SET |
2954 .tv_sec = lfu->lfu_atime_sec,
2955 .tv_nsec = lfu->lfu_atime_nsec,
2958 .tv_sec = lfu->lfu_mtime_sec,
2959 .tv_nsec = lfu->lfu_mtime_nsec,
2962 .tv_sec = lfu->lfu_ctime_sec,
2963 .tv_nsec = lfu->lfu_ctime_nsec,
2969 if (!capable(CAP_SYS_ADMIN))
2972 if (!S_ISREG(inode->i_mode))
2976 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
2978 inode_unlock(inode);
2983 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
2986 case MODE_READ_USER:
2988 case MODE_WRITE_USER:
2995 static const char *const user_lockname[] = LOCK_MODE_NAMES;
2997 /* Used to allow the upper layers of the client to request an LDLM lock
2998 * without doing an actual read or write.
3000 * Used for ladvise lockahead to manually request specific locks.
3002 * \param[in] file file this ladvise lock request is on
3003 * \param[in] ladvise ladvise struct describing this lock request
3005 * \retval 0 success, no detailed result available (sync requests
3006 * and requests sent to the server [not handled locally]
3007 * cannot return detailed results)
3008 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3009 * see definitions for details.
3010 * \retval negative negative errno on error
3012 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3014 struct lu_env *env = NULL;
3015 struct cl_io *io = NULL;
3016 struct cl_lock *lock = NULL;
3017 struct cl_lock_descr *descr = NULL;
3018 struct dentry *dentry = file->f_path.dentry;
3019 struct inode *inode = dentry->d_inode;
3020 enum cl_lock_mode cl_mode;
3021 off_t start = ladvise->lla_start;
3022 off_t end = ladvise->lla_end;
3029 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3030 dentry, dentry->d_inode,
3031 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3034 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3036 GOTO(out, result = cl_mode);
3038 /* Get IO environment */
3039 result = cl_io_get(inode, &env, &io, &refcheck);
3043 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3046 * nothing to do for this io. This currently happens when
3047 * stripe sub-object's are not yet created.
3049 result = io->ci_result;
3050 } else if (result == 0) {
3051 lock = vvp_env_lock(env);
3052 descr = &lock->cll_descr;
3054 descr->cld_obj = io->ci_obj;
3055 /* Convert byte offsets to pages */
3056 descr->cld_start = cl_index(io->ci_obj, start);
3057 descr->cld_end = cl_index(io->ci_obj, end);
3058 descr->cld_mode = cl_mode;
3059 /* CEF_MUST is used because we do not want to convert a
3060 * lockahead request to a lockless lock */
3061 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND |
3064 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3065 descr->cld_enq_flags |= CEF_SPECULATIVE;
3067 result = cl_lock_request(env, io, lock);
3069 /* On success, we need to release the lock */
3071 cl_lock_release(env, lock);
3073 cl_io_fini(env, io);
3074 cl_env_put(env, &refcheck);
3076 /* -ECANCELED indicates a matching lock with a different extent
3077 * was already present, and -EEXIST indicates a matching lock
3078 * on exactly the same extent was already present.
3079 * We convert them to positive values for userspace to make
3080 * recognizing true errors easier.
3081 * Note we can only return these detailed results on async requests,
3082 * as sync requests look the same as i/o requests for locking. */
3083 if (result == -ECANCELED)
3084 result = LLA_RESULT_DIFFERENT;
3085 else if (result == -EEXIST)
3086 result = LLA_RESULT_SAME;
3091 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3093 static int ll_ladvise_sanity(struct inode *inode,
3094 struct llapi_lu_ladvise *ladvise)
3096 struct ll_sb_info *sbi = ll_i2sbi(inode);
3097 enum lu_ladvise_type advice = ladvise->lla_advice;
3098 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3099 * be in the first 32 bits of enum ladvise_flags */
3100 __u32 flags = ladvise->lla_peradvice_flags;
3101 /* 3 lines at 80 characters per line, should be plenty */
3104 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3107 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3108 sbi->ll_fsname, advice,
3109 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3113 /* Per-advice checks */
3115 case LU_LADVISE_LOCKNOEXPAND:
3116 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3118 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3119 "rc = %d\n", sbi->ll_fsname, flags,
3120 ladvise_names[advice], rc);
3124 case LU_LADVISE_LOCKAHEAD:
3125 /* Currently only READ and WRITE modes can be requested */
3126 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3127 ladvise->lla_lockahead_mode == 0) {
3129 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3130 "rc = %d\n", sbi->ll_fsname,
3131 ladvise->lla_lockahead_mode,
3132 ladvise_names[advice], rc);
3136 case LU_LADVISE_WILLREAD:
3137 case LU_LADVISE_DONTNEED:
3139 /* Note fall through above - These checks apply to all advices
3140 * except LOCKNOEXPAND */
3141 if (flags & ~LF_DEFAULT_MASK) {
3143 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3144 "rc = %d\n", sbi->ll_fsname, flags,
3145 ladvise_names[advice], rc);
3148 if (ladvise->lla_start >= ladvise->lla_end) {
3150 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3151 "for %s: rc = %d\n", sbi->ll_fsname,
3152 ladvise->lla_start, ladvise->lla_end,
3153 ladvise_names[advice], rc);
3165 * Give file access advices
3167 * The ladvise interface is similar to Linux fadvise() system call, except it
3168 * forwards the advices directly from Lustre client to server. The server side
3169 * codes will apply appropriate read-ahead and caching techniques for the
3170 * corresponding files.
3172 * A typical workload for ladvise is e.g. a bunch of different clients are
3173 * doing small random reads of a file, so prefetching pages into OSS cache
3174 * with big linear reads before the random IO is a net benefit. Fetching
3175 * all that data into each client cache with fadvise() may not be, due to
3176 * much more data being sent to the client.
3178 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3179 struct llapi_lu_ladvise *ladvise)
3183 struct cl_ladvise_io *lio;
3188 env = cl_env_get(&refcheck);
3190 RETURN(PTR_ERR(env));
3192 io = vvp_env_thread_io(env);
3193 io->ci_obj = ll_i2info(inode)->lli_clob;
3195 /* initialize parameters for ladvise */
3196 lio = &io->u.ci_ladvise;
3197 lio->li_start = ladvise->lla_start;
3198 lio->li_end = ladvise->lla_end;
3199 lio->li_fid = ll_inode2fid(inode);
3200 lio->li_advice = ladvise->lla_advice;
3201 lio->li_flags = flags;
3203 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3204 rc = cl_io_loop(env, io);
3208 cl_io_fini(env, io);
3209 cl_env_put(env, &refcheck);
3213 static int ll_lock_noexpand(struct file *file, int flags)
3215 struct ll_file_data *fd = file->private_data;
3217 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3222 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3225 struct fsxattr fsxattr;
3227 if (copy_from_user(&fsxattr,
3228 (const struct fsxattr __user *)arg,
3232 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3233 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
3234 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3235 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3236 if (copy_to_user((struct fsxattr __user *)arg,
3237 &fsxattr, sizeof(fsxattr)))
3243 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa)
3246 * Project Quota ID state is only allowed to change from within the init
3247 * namespace. Enforce that restriction only if we are trying to change
3248 * the quota ID state. Everything else is allowed in user namespaces.
3250 if (current_user_ns() == &init_user_ns)
3253 if (ll_i2info(inode)->lli_projid != fa->fsx_projid)
3256 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT)) {
3257 if (!(fa->fsx_xflags & FS_XFLAG_PROJINHERIT))
3260 if (fa->fsx_xflags & FS_XFLAG_PROJINHERIT)
3267 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3271 struct md_op_data *op_data;
3272 struct ptlrpc_request *req = NULL;
3274 struct fsxattr fsxattr;
3275 struct cl_object *obj;
3279 if (copy_from_user(&fsxattr,
3280 (const struct fsxattr __user *)arg,
3284 rc = ll_ioctl_check_project(inode, &fsxattr);
3288 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3289 LUSTRE_OPC_ANY, NULL);
3290 if (IS_ERR(op_data))
3291 RETURN(PTR_ERR(op_data));
3293 flags = ll_xflags_to_inode_flags(fsxattr.fsx_xflags);
3294 op_data->op_attr_flags = ll_inode_to_ext_flags(flags);
3295 if (fsxattr.fsx_xflags & FS_XFLAG_PROJINHERIT)
3296 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3297 op_data->op_projid = fsxattr.fsx_projid;
3298 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3299 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL,
3301 ptlrpc_req_finished(req);
3303 GOTO(out_fsxattr, rc);
3304 ll_update_inode_flags(inode, op_data->op_attr_flags);
3305 obj = ll_i2info(inode)->lli_clob;
3307 GOTO(out_fsxattr, rc);
3309 /* Avoiding OST RPC if this is only project ioctl */
3310 if (fsxattr.fsx_xflags == 0 ||
3311 fsxattr.fsx_xflags == FS_XFLAG_PROJINHERIT)
3312 GOTO(out_fsxattr, rc);
3314 OBD_ALLOC_PTR(attr);
3316 GOTO(out_fsxattr, rc = -ENOMEM);
3318 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS,
3319 fsxattr.fsx_xflags);
3322 ll_finish_md_op_data(op_data);
3326 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3329 struct inode *inode = file_inode(file);
3330 struct ll_file_data *fd = file->private_data;
3331 struct ll_inode_info *lli = ll_i2info(inode);
3332 struct obd_client_handle *och = NULL;
3333 struct split_param sp;
3334 struct pcc_param param;
3335 bool lease_broken = false;
3337 enum mds_op_bias bias = 0;
3338 struct file *layout_file = NULL;
3340 size_t data_size = 0;
3341 bool attached = false;
3346 mutex_lock(&lli->lli_och_mutex);
3347 if (fd->fd_lease_och != NULL) {
3348 och = fd->fd_lease_och;
3349 fd->fd_lease_och = NULL;
3351 mutex_unlock(&lli->lli_och_mutex);
3356 fmode = och->och_flags;
3358 switch (ioc->lil_flags) {
3359 case LL_LEASE_RESYNC_DONE:
3360 if (ioc->lil_count > IOC_IDS_MAX)
3361 GOTO(out_lease_close, rc = -EINVAL);
3363 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3364 OBD_ALLOC(data, data_size);
3366 GOTO(out_lease_close, rc = -ENOMEM);
3368 if (copy_from_user(data, (void __user *)arg, data_size))
3369 GOTO(out_lease_close, rc = -EFAULT);
3371 bias = MDS_CLOSE_RESYNC_DONE;
3373 case LL_LEASE_LAYOUT_MERGE: {
3376 if (ioc->lil_count != 1)
3377 GOTO(out_lease_close, rc = -EINVAL);
3379 arg += sizeof(*ioc);
3380 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3381 GOTO(out_lease_close, rc = -EFAULT);
3383 layout_file = fget(fd);
3385 GOTO(out_lease_close, rc = -EBADF);
3387 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3388 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3389 GOTO(out_lease_close, rc = -EPERM);
3391 data = file_inode(layout_file);
3392 bias = MDS_CLOSE_LAYOUT_MERGE;
3395 case LL_LEASE_LAYOUT_SPLIT: {
3399 if (ioc->lil_count != 2)
3400 GOTO(out_lease_close, rc = -EINVAL);
3402 arg += sizeof(*ioc);
3403 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3404 GOTO(out_lease_close, rc = -EFAULT);
3406 arg += sizeof(__u32);
3407 if (copy_from_user(&mirror_id, (void __user *)arg,
3409 GOTO(out_lease_close, rc = -EFAULT);
3411 layout_file = fget(fdv);
3413 GOTO(out_lease_close, rc = -EBADF);
3415 sp.sp_inode = file_inode(layout_file);
3416 sp.sp_mirror_id = (__u16)mirror_id;
3418 bias = MDS_CLOSE_LAYOUT_SPLIT;
3421 case LL_LEASE_PCC_ATTACH:
3422 if (ioc->lil_count != 1)
3425 arg += sizeof(*ioc);
3426 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3428 GOTO(out_lease_close, rc2 = -EFAULT);
3430 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3432 GOTO(out_lease_close, rc2);
3435 /* Grab latest data version */
3436 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3439 GOTO(out_lease_close, rc2);
3442 bias = MDS_PCC_ATTACH;
3445 /* without close intent */
3450 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3454 rc = ll_lease_och_release(inode, file);
3463 switch (ioc->lil_flags) {
3464 case LL_LEASE_RESYNC_DONE:
3466 OBD_FREE(data, data_size);
3468 case LL_LEASE_LAYOUT_MERGE:
3469 case LL_LEASE_LAYOUT_SPLIT:
3473 case LL_LEASE_PCC_ATTACH:
3476 rc = pcc_readwrite_attach_fini(file, inode,
3477 param.pa_layout_gen,
3484 rc = ll_lease_type_from_fmode(fmode);
3488 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3491 struct inode *inode = file_inode(file);
3492 struct ll_inode_info *lli = ll_i2info(inode);
3493 struct ll_file_data *fd = file->private_data;
3494 struct obd_client_handle *och = NULL;
3495 __u64 open_flags = 0;
3501 switch (ioc->lil_mode) {
3502 case LL_LEASE_WRLCK:
3503 if (!(file->f_mode & FMODE_WRITE))
3505 fmode = FMODE_WRITE;
3507 case LL_LEASE_RDLCK:
3508 if (!(file->f_mode & FMODE_READ))
3512 case LL_LEASE_UNLCK:
3513 RETURN(ll_file_unlock_lease(file, ioc, arg));
3518 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3520 /* apply for lease */
3521 if (ioc->lil_flags & LL_LEASE_RESYNC)
3522 open_flags = MDS_OPEN_RESYNC;
3523 och = ll_lease_open(inode, file, fmode, open_flags);
3525 RETURN(PTR_ERR(och));
3527 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3528 rc = ll_lease_file_resync(och, inode, arg);
3530 ll_lease_close(och, inode, NULL);
3533 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3535 ll_lease_close(och, inode, NULL);
3541 mutex_lock(&lli->lli_och_mutex);
3542 if (fd->fd_lease_och == NULL) {
3543 fd->fd_lease_och = och;
3546 mutex_unlock(&lli->lli_och_mutex);
3548 /* impossible now that only excl is supported for now */
3549 ll_lease_close(och, inode, &lease_broken);
3555 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3557 struct ll_inode_info *lli = ll_i2info(inode);
3558 struct ll_sb_info *sbi = ll_i2sbi(inode);
3559 __u64 now = ktime_get_real_seconds();
3562 spin_lock(&lli->lli_heat_lock);
3563 heat->lh_flags = lli->lli_heat_flags;
3564 for (i = 0; i < heat->lh_count; i++)
3565 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3566 now, sbi->ll_heat_decay_weight,
3567 sbi->ll_heat_period_second);
3568 spin_unlock(&lli->lli_heat_lock);
3571 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3573 struct ll_inode_info *lli = ll_i2info(inode);
3576 spin_lock(&lli->lli_heat_lock);
3577 if (flags & LU_HEAT_FLAG_CLEAR)
3578 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3580 if (flags & LU_HEAT_FLAG_OFF)
3581 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3583 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3585 spin_unlock(&lli->lli_heat_lock);
3591 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3593 struct inode *inode = file_inode(file);
3594 struct ll_file_data *fd = file->private_data;
3598 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3599 PFID(ll_inode2fid(inode)), inode, cmd);
3600 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3602 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3603 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3607 case LL_IOC_GETFLAGS:
3608 /* Get the current value of the file flags */
3609 return put_user(fd->fd_flags, (int __user *)arg);
3610 case LL_IOC_SETFLAGS:
3611 case LL_IOC_CLRFLAGS:
3612 /* Set or clear specific file flags */
3613 /* XXX This probably needs checks to ensure the flags are
3614 * not abused, and to handle any flag side effects.
3616 if (get_user(flags, (int __user *) arg))
3619 if (cmd == LL_IOC_SETFLAGS) {
3620 if ((flags & LL_FILE_IGNORE_LOCK) &&
3621 !(file->f_flags & O_DIRECT)) {
3622 CERROR("%s: unable to disable locking on "
3623 "non-O_DIRECT file\n", current->comm);
3627 fd->fd_flags |= flags;
3629 fd->fd_flags &= ~flags;
3632 case LL_IOC_LOV_SETSTRIPE:
3633 case LL_IOC_LOV_SETSTRIPE_NEW:
3634 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3635 case LL_IOC_LOV_SETEA:
3636 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3637 case LL_IOC_LOV_SWAP_LAYOUTS: {
3639 struct lustre_swap_layouts lsl;
3641 if (copy_from_user(&lsl, (char __user *)arg,
3642 sizeof(struct lustre_swap_layouts)))
3645 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3648 file2 = fget(lsl.sl_fd);
3652 /* O_WRONLY or O_RDWR */
3653 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3654 GOTO(out, rc = -EPERM);
3656 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3657 struct inode *inode2;
3658 struct ll_inode_info *lli;
3659 struct obd_client_handle *och = NULL;
3661 lli = ll_i2info(inode);
3662 mutex_lock(&lli->lli_och_mutex);
3663 if (fd->fd_lease_och != NULL) {
3664 och = fd->fd_lease_och;
3665 fd->fd_lease_och = NULL;
3667 mutex_unlock(&lli->lli_och_mutex);
3669 GOTO(out, rc = -ENOLCK);
3670 inode2 = file_inode(file2);
3671 rc = ll_swap_layouts_close(och, inode, inode2);
3673 rc = ll_swap_layouts(file, file2, &lsl);
3679 case LL_IOC_LOV_GETSTRIPE:
3680 case LL_IOC_LOV_GETSTRIPE_NEW:
3681 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3682 case FS_IOC_GETFLAGS:
3683 case FS_IOC_SETFLAGS:
3684 RETURN(ll_iocontrol(inode, file, cmd, arg));
3685 case FSFILT_IOC_GETVERSION:
3686 case FS_IOC_GETVERSION:
3687 RETURN(put_user(inode->i_generation, (int __user *)arg));
3688 /* We need to special case any other ioctls we want to handle,
3689 * to send them to the MDS/OST as appropriate and to properly
3690 * network encode the arg field. */
3691 case FS_IOC_SETVERSION:
3694 case LL_IOC_GROUP_LOCK:
3695 RETURN(ll_get_grouplock(inode, file, arg));
3696 case LL_IOC_GROUP_UNLOCK:
3697 RETURN(ll_put_grouplock(inode, file, arg));
3698 case IOC_OBD_STATFS:
3699 RETURN(ll_obd_statfs(inode, (void __user *)arg));
3701 case LL_IOC_FLUSHCTX:
3702 RETURN(ll_flush_ctx(inode));
3703 case LL_IOC_PATH2FID: {
3704 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
3705 sizeof(struct lu_fid)))
3710 case LL_IOC_GETPARENT:
3711 RETURN(ll_getparent(file, (struct getparent __user *)arg));
3713 case OBD_IOC_FID2PATH:
3714 RETURN(ll_fid2path(inode, (void __user *)arg));
3715 case LL_IOC_DATA_VERSION: {
3716 struct ioc_data_version idv;
3719 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
3722 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
3723 rc = ll_ioc_data_version(inode, &idv);
3726 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
3732 case LL_IOC_GET_MDTIDX: {
3735 mdtidx = ll_get_mdt_idx(inode);
3739 if (put_user((int)mdtidx, (int __user *)arg))
3744 case OBD_IOC_GETDTNAME:
3745 case OBD_IOC_GETMDNAME:
3746 RETURN(ll_get_obd_name(inode, cmd, arg));
3747 case LL_IOC_HSM_STATE_GET: {
3748 struct md_op_data *op_data;
3749 struct hsm_user_state *hus;
3756 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3757 LUSTRE_OPC_ANY, hus);
3758 if (IS_ERR(op_data)) {
3760 RETURN(PTR_ERR(op_data));
3763 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3766 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
3769 ll_finish_md_op_data(op_data);
3773 case LL_IOC_HSM_STATE_SET: {
3774 struct hsm_state_set *hss;
3781 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
3786 rc = ll_hsm_state_set(inode, hss);
3791 case LL_IOC_HSM_ACTION: {
3792 struct md_op_data *op_data;
3793 struct hsm_current_action *hca;
3800 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3801 LUSTRE_OPC_ANY, hca);
3802 if (IS_ERR(op_data)) {
3804 RETURN(PTR_ERR(op_data));
3807 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3810 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
3813 ll_finish_md_op_data(op_data);
3817 case LL_IOC_SET_LEASE_OLD: {
3818 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
3820 RETURN(ll_file_set_lease(file, &ioc, 0));
3822 case LL_IOC_SET_LEASE: {
3823 struct ll_ioc_lease ioc;
3825 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
3828 RETURN(ll_file_set_lease(file, &ioc, arg));
3830 case LL_IOC_GET_LEASE: {
3831 struct ll_inode_info *lli = ll_i2info(inode);
3832 struct ldlm_lock *lock = NULL;
3835 mutex_lock(&lli->lli_och_mutex);
3836 if (fd->fd_lease_och != NULL) {
3837 struct obd_client_handle *och = fd->fd_lease_och;
3839 lock = ldlm_handle2lock(&och->och_lease_handle);
3841 lock_res_and_lock(lock);
3842 if (!ldlm_is_cancel(lock))
3843 fmode = och->och_flags;
3845 unlock_res_and_lock(lock);
3846 LDLM_LOCK_PUT(lock);
3849 mutex_unlock(&lli->lli_och_mutex);
3851 RETURN(ll_lease_type_from_fmode(fmode));
3853 case LL_IOC_HSM_IMPORT: {
3854 struct hsm_user_import *hui;
3860 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
3865 rc = ll_hsm_import(inode, file, hui);
3870 case LL_IOC_FUTIMES_3: {
3871 struct ll_futimes_3 lfu;
3873 if (copy_from_user(&lfu,
3874 (const struct ll_futimes_3 __user *)arg,
3878 RETURN(ll_file_futimes_3(file, &lfu));
3880 case LL_IOC_LADVISE: {
3881 struct llapi_ladvise_hdr *k_ladvise_hdr;
3882 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
3885 int alloc_size = sizeof(*k_ladvise_hdr);
3888 u_ladvise_hdr = (void __user *)arg;
3889 OBD_ALLOC_PTR(k_ladvise_hdr);
3890 if (k_ladvise_hdr == NULL)
3893 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3894 GOTO(out_ladvise, rc = -EFAULT);
3896 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
3897 k_ladvise_hdr->lah_count < 1)
3898 GOTO(out_ladvise, rc = -EINVAL);
3900 num_advise = k_ladvise_hdr->lah_count;
3901 if (num_advise >= LAH_COUNT_MAX)
3902 GOTO(out_ladvise, rc = -EFBIG);
3904 OBD_FREE_PTR(k_ladvise_hdr);
3905 alloc_size = offsetof(typeof(*k_ladvise_hdr),
3906 lah_advise[num_advise]);
3907 OBD_ALLOC(k_ladvise_hdr, alloc_size);
3908 if (k_ladvise_hdr == NULL)
3912 * TODO: submit multiple advices to one server in a single RPC
3914 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3915 GOTO(out_ladvise, rc = -EFAULT);
3917 for (i = 0; i < num_advise; i++) {
3918 struct llapi_lu_ladvise *k_ladvise =
3919 &k_ladvise_hdr->lah_advise[i];
3920 struct llapi_lu_ladvise __user *u_ladvise =
3921 &u_ladvise_hdr->lah_advise[i];
3923 rc = ll_ladvise_sanity(inode, k_ladvise);
3925 GOTO(out_ladvise, rc);
3927 switch (k_ladvise->lla_advice) {
3928 case LU_LADVISE_LOCKNOEXPAND:
3929 rc = ll_lock_noexpand(file,
3930 k_ladvise->lla_peradvice_flags);
3931 GOTO(out_ladvise, rc);
3932 case LU_LADVISE_LOCKAHEAD:
3934 rc = ll_file_lock_ahead(file, k_ladvise);
3937 GOTO(out_ladvise, rc);
3940 &u_ladvise->lla_lockahead_result))
3941 GOTO(out_ladvise, rc = -EFAULT);
3944 rc = ll_ladvise(inode, file,
3945 k_ladvise_hdr->lah_flags,
3948 GOTO(out_ladvise, rc);
3955 OBD_FREE(k_ladvise_hdr, alloc_size);
3958 case LL_IOC_FLR_SET_MIRROR: {
3959 /* mirror I/O must be direct to avoid polluting page cache
3961 if (!(file->f_flags & O_DIRECT))
3964 fd->fd_designated_mirror = (__u32)arg;
3967 case FS_IOC_FSGETXATTR:
3968 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
3969 case FS_IOC_FSSETXATTR:
3970 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
3972 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
3973 case LL_IOC_HEAT_GET: {
3974 struct lu_heat uheat;
3975 struct lu_heat *heat;
3978 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
3981 if (uheat.lh_count > OBD_HEAT_COUNT)
3982 uheat.lh_count = OBD_HEAT_COUNT;
3984 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
3985 OBD_ALLOC(heat, size);
3989 heat->lh_count = uheat.lh_count;
3990 ll_heat_get(inode, heat);
3991 rc = copy_to_user((char __user *)arg, heat, size);
3992 OBD_FREE(heat, size);
3993 RETURN(rc ? -EFAULT : 0);
3995 case LL_IOC_HEAT_SET: {
3998 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
4001 rc = ll_heat_set(inode, flags);
4004 case LL_IOC_PCC_DETACH: {
4005 struct lu_pcc_detach *detach;
4007 OBD_ALLOC_PTR(detach);
4011 if (copy_from_user(detach,
4012 (const struct lu_pcc_detach __user *)arg,
4014 GOTO(out_detach_free, rc = -EFAULT);
4016 if (!S_ISREG(inode->i_mode))
4017 GOTO(out_detach_free, rc = -EINVAL);
4019 if (!inode_owner_or_capable(inode))
4020 GOTO(out_detach_free, rc = -EPERM);
4022 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4024 OBD_FREE_PTR(detach);
4027 case LL_IOC_PCC_STATE: {
4028 struct lu_pcc_state __user *ustate =
4029 (struct lu_pcc_state __user *)arg;
4030 struct lu_pcc_state *state;
4032 OBD_ALLOC_PTR(state);
4036 if (copy_from_user(state, ustate, sizeof(*state)))
4037 GOTO(out_state, rc = -EFAULT);
4039 rc = pcc_ioctl_state(file, inode, state);
4041 GOTO(out_state, rc);
4043 if (copy_to_user(ustate, state, sizeof(*state)))
4044 GOTO(out_state, rc = -EFAULT);
4047 OBD_FREE_PTR(state);
4050 #ifdef HAVE_LUSTRE_CRYPTO
4051 case LL_IOC_SET_ENCRYPTION_POLICY:
4052 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4054 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
4055 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
4056 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4058 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
4059 case LL_IOC_ADD_ENCRYPTION_KEY:
4060 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4062 return llcrypt_ioctl_add_key(file, (void __user *)arg);
4063 case LL_IOC_REMOVE_ENCRYPTION_KEY:
4064 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4066 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
4067 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
4068 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4070 return llcrypt_ioctl_remove_key_all_users(file,
4071 (void __user *)arg);
4072 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
4073 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4075 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
4078 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
4079 (void __user *)arg));
4083 loff_t ll_lseek(struct inode *inode, loff_t offset, int whence)
4087 struct cl_lseek_io *lsio;
4094 env = cl_env_get(&refcheck);
4096 RETURN(PTR_ERR(env));
4098 io = vvp_env_thread_io(env);
4099 io->ci_obj = ll_i2info(inode)->lli_clob;
4101 lsio = &io->u.ci_lseek;
4102 lsio->ls_start = offset;
4103 lsio->ls_whence = whence;
4104 lsio->ls_result = -ENXIO;
4107 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4109 rc = cl_io_loop(env, io);
4112 retval = rc ? : lsio->ls_result;
4113 cl_io_fini(env, io);
4114 } while (unlikely(io->ci_need_restart));
4116 cl_env_put(env, &refcheck);
4121 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4123 struct inode *inode = file_inode(file);
4124 loff_t retval = offset, eof = 0;
4125 ktime_t kstart = ktime_get();
4129 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4130 PFID(ll_inode2fid(inode)), inode, retval, retval,
4133 if (origin == SEEK_END) {
4134 retval = ll_glimpse_size(inode);
4137 eof = i_size_read(inode);
4140 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4144 /* flush local cache first if any */
4145 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4148 retval = ll_lseek(inode, offset, origin);
4151 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4153 retval = generic_file_llseek_size(file, offset, origin,
4154 ll_file_maxbytes(inode), eof);
4157 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4158 ktime_us_delta(ktime_get(), kstart));
4162 static int ll_flush(struct file *file, fl_owner_t id)
4164 struct inode *inode = file_inode(file);
4165 struct ll_inode_info *lli = ll_i2info(inode);
4166 struct ll_file_data *fd = file->private_data;
4169 LASSERT(!S_ISDIR(inode->i_mode));
4171 /* catch async errors that were recorded back when async writeback
4172 * failed for pages in this mapping. */
4173 rc = lli->lli_async_rc;
4174 lli->lli_async_rc = 0;
4175 if (lli->lli_clob != NULL) {
4176 err = lov_read_and_clear_async_rc(lli->lli_clob);
4181 /* The application has been told write failure already.
4182 * Do not report failure again. */
4183 if (fd->fd_write_failed)
4185 return rc ? -EIO : 0;
4189 * Called to make sure a portion of file has been written out.
4190 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4192 * Return how many pages have been written.
4194 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4195 enum cl_fsync_mode mode, int ignore_layout)
4199 struct cl_fsync_io *fio;
4204 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4205 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4208 env = cl_env_get(&refcheck);
4210 RETURN(PTR_ERR(env));
4212 io = vvp_env_thread_io(env);
4213 io->ci_obj = ll_i2info(inode)->lli_clob;
4214 io->ci_ignore_layout = ignore_layout;
4216 /* initialize parameters for sync */
4217 fio = &io->u.ci_fsync;
4218 fio->fi_start = start;
4220 fio->fi_fid = ll_inode2fid(inode);
4221 fio->fi_mode = mode;
4222 fio->fi_nr_written = 0;
4224 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4225 result = cl_io_loop(env, io);
4227 result = io->ci_result;
4229 result = fio->fi_nr_written;
4230 cl_io_fini(env, io);
4231 cl_env_put(env, &refcheck);
4237 * When dentry is provided (the 'else' case), file_dentry() may be
4238 * null and dentry must be used directly rather than pulled from
4239 * file_dentry() as is done otherwise.
4242 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4244 struct dentry *dentry = file_dentry(file);
4245 struct inode *inode = dentry->d_inode;
4246 struct ll_inode_info *lli = ll_i2info(inode);
4247 struct ptlrpc_request *req;
4248 ktime_t kstart = ktime_get();
4254 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4255 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4257 /* fsync's caller has already called _fdata{sync,write}, we want
4258 * that IO to finish before calling the osc and mdc sync methods */
4259 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4262 /* catch async errors that were recorded back when async writeback
4263 * failed for pages in this mapping. */
4264 if (!S_ISDIR(inode->i_mode)) {
4265 err = lli->lli_async_rc;
4266 lli->lli_async_rc = 0;
4269 if (lli->lli_clob != NULL) {
4270 err = lov_read_and_clear_async_rc(lli->lli_clob);
4276 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4280 ptlrpc_req_finished(req);
4282 if (S_ISREG(inode->i_mode)) {
4283 struct ll_file_data *fd = file->private_data;
4286 /* Sync metadata on MDT first, and then sync the cached data
4289 err = pcc_fsync(file, start, end, datasync, &cached);
4291 err = cl_sync_file_range(inode, start, end,
4293 if (rc == 0 && err < 0)
4296 fd->fd_write_failed = true;
4298 fd->fd_write_failed = false;
4301 inode_unlock(inode);
4304 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4305 ktime_us_delta(ktime_get(), kstart));
4310 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4312 struct inode *inode = file_inode(file);
4313 struct ll_sb_info *sbi = ll_i2sbi(inode);
4314 struct ldlm_enqueue_info einfo = {
4315 .ei_type = LDLM_FLOCK,
4316 .ei_cb_cp = ldlm_flock_completion_ast,
4317 .ei_cbdata = file_lock,
4319 struct md_op_data *op_data;
4320 struct lustre_handle lockh = { 0 };
4321 union ldlm_policy_data flock = { { 0 } };
4322 int fl_type = file_lock->fl_type;
4323 ktime_t kstart = ktime_get();
4329 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4330 PFID(ll_inode2fid(inode)), file_lock);
4332 if (file_lock->fl_flags & FL_FLOCK) {
4333 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4334 /* flocks are whole-file locks */
4335 flock.l_flock.end = OFFSET_MAX;
4336 /* For flocks owner is determined by the local file desctiptor*/
4337 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4338 } else if (file_lock->fl_flags & FL_POSIX) {
4339 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4340 flock.l_flock.start = file_lock->fl_start;
4341 flock.l_flock.end = file_lock->fl_end;
4345 flock.l_flock.pid = file_lock->fl_pid;
4347 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4348 /* Somewhat ugly workaround for svc lockd.
4349 * lockd installs custom fl_lmops->lm_compare_owner that checks
4350 * for the fl_owner to be the same (which it always is on local node
4351 * I guess between lockd processes) and then compares pid.
4352 * As such we assign pid to the owner field to make it all work,
4353 * conflict with normal locks is unlikely since pid space and
4354 * pointer space for current->files are not intersecting */
4355 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4356 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4361 einfo.ei_mode = LCK_PR;
4364 /* An unlock request may or may not have any relation to
4365 * existing locks so we may not be able to pass a lock handle
4366 * via a normal ldlm_lock_cancel() request. The request may even
4367 * unlock a byte range in the middle of an existing lock. In
4368 * order to process an unlock request we need all of the same
4369 * information that is given with a normal read or write record
4370 * lock request. To avoid creating another ldlm unlock (cancel)
4371 * message we'll treat a LCK_NL flock request as an unlock. */
4372 einfo.ei_mode = LCK_NL;
4375 einfo.ei_mode = LCK_PW;
4378 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4393 flags = LDLM_FL_BLOCK_NOWAIT;
4399 flags = LDLM_FL_TEST_LOCK;
4402 CERROR("unknown fcntl lock command: %d\n", cmd);
4406 /* Save the old mode so that if the mode in the lock changes we
4407 * can decrement the appropriate reader or writer refcount. */
4408 file_lock->fl_type = einfo.ei_mode;
4410 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4411 LUSTRE_OPC_ANY, NULL);
4412 if (IS_ERR(op_data))
4413 RETURN(PTR_ERR(op_data));
4415 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4416 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4417 flock.l_flock.pid, flags, einfo.ei_mode,
4418 flock.l_flock.start, flock.l_flock.end);
4420 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4423 /* Restore the file lock type if not TEST lock. */
4424 if (!(flags & LDLM_FL_TEST_LOCK))
4425 file_lock->fl_type = fl_type;
4427 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4428 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4429 !(flags & LDLM_FL_TEST_LOCK))
4430 rc2 = locks_lock_file_wait(file, file_lock);
4432 if ((file_lock->fl_flags & FL_FLOCK) &&
4433 (rc == 0 || file_lock->fl_type == F_UNLCK))
4434 rc2 = flock_lock_file_wait(file, file_lock);
4435 if ((file_lock->fl_flags & FL_POSIX) &&
4436 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4437 !(flags & LDLM_FL_TEST_LOCK))
4438 rc2 = posix_lock_file_wait(file, file_lock);
4439 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4441 if (rc2 && file_lock->fl_type != F_UNLCK) {
4442 einfo.ei_mode = LCK_NL;
4443 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4448 ll_finish_md_op_data(op_data);
4451 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4452 ktime_us_delta(ktime_get(), kstart));
4456 int ll_get_fid_by_name(struct inode *parent, const char *name,
4457 int namelen, struct lu_fid *fid,
4458 struct inode **inode)
4460 struct md_op_data *op_data = NULL;
4461 struct mdt_body *body;
4462 struct ptlrpc_request *req;
4466 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4467 LUSTRE_OPC_ANY, NULL);
4468 if (IS_ERR(op_data))
4469 RETURN(PTR_ERR(op_data));
4471 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4472 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4473 ll_finish_md_op_data(op_data);
4477 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4479 GOTO(out_req, rc = -EFAULT);
4481 *fid = body->mbo_fid1;
4484 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
4486 ptlrpc_req_finished(req);
4490 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4493 struct dentry *dchild = NULL;
4494 struct inode *child_inode = NULL;
4495 struct md_op_data *op_data;
4496 struct ptlrpc_request *request = NULL;
4497 struct obd_client_handle *och = NULL;
4499 struct mdt_body *body;
4500 __u64 data_version = 0;
4501 size_t namelen = strlen(name);
4502 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4506 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4507 PFID(ll_inode2fid(parent)), name,
4508 lum->lum_stripe_offset, lum->lum_stripe_count);
4510 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4511 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4512 lustre_swab_lmv_user_md(lum);
4514 /* Get child FID first */
4515 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4518 dchild = d_lookup(file_dentry(file), &qstr);
4520 if (dchild->d_inode)
4521 child_inode = igrab(dchild->d_inode);
4526 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4535 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4536 OBD_CONNECT2_DIR_MIGRATE)) {
4537 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4538 ll_dir_striped(child_inode)) {
4539 CERROR("%s: MDT doesn't support stripe directory "
4540 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4541 GOTO(out_iput, rc = -EOPNOTSUPP);
4546 * lfs migrate command needs to be blocked on the client
4547 * by checking the migrate FID against the FID of the
4550 if (child_inode == parent->i_sb->s_root->d_inode)
4551 GOTO(out_iput, rc = -EINVAL);
4553 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4554 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4555 if (IS_ERR(op_data))
4556 GOTO(out_iput, rc = PTR_ERR(op_data));
4558 inode_lock(child_inode);
4559 op_data->op_fid3 = *ll_inode2fid(child_inode);
4560 if (!fid_is_sane(&op_data->op_fid3)) {
4561 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4562 ll_i2sbi(parent)->ll_fsname, name,
4563 PFID(&op_data->op_fid3));
4564 GOTO(out_unlock, rc = -EINVAL);
4567 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4568 op_data->op_data = lum;
4569 op_data->op_data_size = lumlen;
4572 if (S_ISREG(child_inode->i_mode)) {
4573 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4577 GOTO(out_unlock, rc);
4580 rc = ll_data_version(child_inode, &data_version,
4583 GOTO(out_close, rc);
4585 op_data->op_open_handle = och->och_open_handle;
4586 op_data->op_data_version = data_version;
4587 op_data->op_lease_handle = och->och_lease_handle;
4588 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4590 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4591 och->och_mod->mod_open_req->rq_replay = 0;
4592 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4595 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name, namelen,
4596 name, namelen, &request);
4598 LASSERT(request != NULL);
4599 ll_update_times(request, parent);
4602 if (rc == 0 || rc == -EAGAIN) {
4603 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4604 LASSERT(body != NULL);
4606 /* If the server does release layout lock, then we cleanup
4607 * the client och here, otherwise release it in out_close: */
4608 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4609 obd_mod_put(och->och_mod);
4610 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4612 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4618 if (request != NULL) {
4619 ptlrpc_req_finished(request);
4623 /* Try again if the lease has cancelled. */
4624 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4629 ll_lease_close(och, child_inode, NULL);
4631 clear_nlink(child_inode);
4633 inode_unlock(child_inode);
4634 ll_finish_md_op_data(op_data);
4641 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4643 struct ll_file_data *fd = file->private_data;
4647 * In order to avoid flood of warning messages, only print one message
4648 * for one file. And the entire message rate on the client is limited
4649 * by CDEBUG_LIMIT too.
4651 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
4652 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
4653 CDEBUG_LIMIT(D_TTY | D_CONSOLE,
4654 "flock disabled, mount with '-o [local]flock' to enable\r\n");
4660 * test if some locks matching bits and l_req_mode are acquired
4661 * - bits can be in different locks
4662 * - if found clear the common lock bits in *bits
4663 * - the bits not found, are kept in *bits
4665 * \param bits [IN] searched lock bits [IN]
4666 * \param l_req_mode [IN] searched lock mode
4667 * \retval boolean, true iff all bits are found
4669 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
4671 struct lustre_handle lockh;
4672 union ldlm_policy_data policy;
4673 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
4674 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
4683 fid = &ll_i2info(inode)->lli_fid;
4684 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
4685 ldlm_lockname[mode]);
4687 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
4688 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
4689 policy.l_inodebits.bits = *bits & BIT(i);
4690 if (policy.l_inodebits.bits == 0)
4693 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
4694 &policy, mode, &lockh)) {
4695 struct ldlm_lock *lock;
4697 lock = ldlm_handle2lock(&lockh);
4700 ~(lock->l_policy_data.l_inodebits.bits);
4701 LDLM_LOCK_PUT(lock);
4703 *bits &= ~policy.l_inodebits.bits;
4710 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
4711 struct lustre_handle *lockh, __u64 flags,
4712 enum ldlm_mode mode)
4714 union ldlm_policy_data policy = { .l_inodebits = { bits } };
4719 fid = &ll_i2info(inode)->lli_fid;
4720 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
4722 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
4723 fid, LDLM_IBITS, &policy, mode, lockh);
4728 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
4730 /* Already unlinked. Just update nlink and return success */
4731 if (rc == -ENOENT) {
4733 /* If it is striped directory, and there is bad stripe
4734 * Let's revalidate the dentry again, instead of returning
4736 if (ll_dir_striped(inode))
4739 /* This path cannot be hit for regular files unless in
4740 * case of obscure races, so no need to to validate
4742 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
4744 } else if (rc != 0) {
4745 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
4746 "%s: revalidate FID "DFID" error: rc = %d\n",
4747 ll_i2sbi(inode)->ll_fsname,
4748 PFID(ll_inode2fid(inode)), rc);
4754 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
4756 struct inode *parent;
4757 struct inode *inode = dentry->d_inode;
4758 struct obd_export *exp = ll_i2mdexp(inode);
4759 struct lookup_intent oit = {
4762 struct ptlrpc_request *req = NULL;
4763 struct md_op_data *op_data;
4764 const char *name = NULL;
4769 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
4770 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
4772 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
4773 parent = dentry->d_parent->d_inode;
4774 name = dentry->d_name.name;
4775 namelen = dentry->d_name.len;
4780 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
4781 LUSTRE_OPC_ANY, NULL);
4782 if (IS_ERR(op_data))
4783 RETURN(PTR_ERR(op_data));
4785 /* Call getattr by fid */
4786 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
4787 op_data->op_flags = MF_GETATTR_BY_FID;
4788 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
4789 ll_finish_md_op_data(op_data);
4791 rc = ll_inode_revalidate_fini(inode, rc);
4795 rc = ll_revalidate_it_finish(req, &oit, dentry);
4797 ll_intent_release(&oit);
4801 /* Unlinked? Unhash dentry, so it is not picked up later by
4802 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
4803 * here to preserve get_cwd functionality on 2.6.
4805 if (!dentry->d_inode->i_nlink)
4806 d_lustre_invalidate(dentry);
4808 ll_lookup_finish_locks(&oit, dentry);
4810 ptlrpc_req_finished(req);
4815 static int ll_merge_md_attr(struct inode *inode)
4817 struct ll_inode_info *lli = ll_i2info(inode);
4818 struct cl_attr attr = { 0 };
4821 LASSERT(lli->lli_lsm_md != NULL);
4823 if (!lmv_dir_striped(lli->lli_lsm_md))
4826 down_read(&lli->lli_lsm_sem);
4827 rc = md_merge_attr(ll_i2mdexp(inode), &lli->lli_fid, lli->lli_lsm_md,
4828 &attr, ll_md_blocking_ast);
4829 up_read(&lli->lli_lsm_sem);
4833 set_nlink(inode, attr.cat_nlink);
4834 inode->i_blocks = attr.cat_blocks;
4835 i_size_write(inode, attr.cat_size);
4837 ll_i2info(inode)->lli_atime = attr.cat_atime;
4838 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
4839 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
4844 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
4847 struct inode *inode = de->d_inode;
4848 struct ll_sb_info *sbi = ll_i2sbi(inode);
4849 struct ll_inode_info *lli = ll_i2info(inode);
4850 struct inode *dir = de->d_parent->d_inode;
4851 bool need_glimpse = true;
4852 ktime_t kstart = ktime_get();
4855 /* The OST object(s) determine the file size, blocks and mtime. */
4856 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
4857 request_mask & STATX_MTIME))
4858 need_glimpse = false;
4860 if (dentry_may_statahead(dir, de))
4861 ll_start_statahead(dir, de, need_glimpse &&
4862 !(flags & AT_STATX_DONT_SYNC));
4864 if (flags & AT_STATX_DONT_SYNC)
4865 GOTO(fill_attr, rc = 0);
4867 rc = ll_inode_revalidate(de, IT_GETATTR);
4871 if (S_ISREG(inode->i_mode)) {
4875 GOTO(fill_attr, rc);
4877 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
4878 if (cached && rc < 0)
4882 GOTO(fill_attr, rc);
4885 * If the returned attr is masked with OBD_MD_FLSIZE &
4886 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
4887 * or blocks obtained from MDT is strictly correct, and the file
4888 * is usually not being modified by clients, and the [a|m|c]time
4889 * got from MDT is also strictly correct.
4890 * Under this circumstance, it does not need to send glimpse
4891 * RPCs to OSTs for file attributes such as the size and blocks.
4893 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
4894 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
4895 lli->lli_attr_valid & OBD_MD_FLMTIME) {
4896 inode->i_mtime.tv_sec = lli->lli_mtime;
4897 if (lli->lli_attr_valid & OBD_MD_FLATIME)
4898 inode->i_atime.tv_sec = lli->lli_atime;
4899 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
4900 inode->i_ctime.tv_sec = lli->lli_ctime;
4901 GOTO(fill_attr, rc);
4904 /* In case of restore, the MDT has the right size and has
4905 * already send it back without granting the layout lock,
4906 * inode is up-to-date so glimpse is useless.
4907 * Also to glimpse we need the layout, in case of a running
4908 * restore the MDT holds the layout lock so the glimpse will
4909 * block up to the end of restore (getattr will block)
4911 if (!ll_file_test_flag(lli, LLIF_FILE_RESTORING)) {
4912 rc = ll_glimpse_size(inode);
4917 /* If object isn't regular a file then don't validate size. */
4918 if (ll_dir_striped(inode)) {
4919 rc = ll_merge_md_attr(inode);
4924 if (lli->lli_attr_valid & OBD_MD_FLATIME)
4925 inode->i_atime.tv_sec = lli->lli_atime;
4926 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
4927 inode->i_mtime.tv_sec = lli->lli_mtime;
4928 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
4929 inode->i_ctime.tv_sec = lli->lli_ctime;
4933 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
4935 if (ll_need_32bit_api(sbi)) {
4936 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
4937 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
4938 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
4940 stat->ino = inode->i_ino;
4941 stat->dev = inode->i_sb->s_dev;
4942 stat->rdev = inode->i_rdev;
4945 stat->mode = inode->i_mode;
4946 stat->uid = inode->i_uid;
4947 stat->gid = inode->i_gid;
4948 stat->atime = inode->i_atime;
4949 stat->mtime = inode->i_mtime;
4950 stat->ctime = inode->i_ctime;
4951 /* stat->blksize is used to tell about preferred IO size */
4952 if (sbi->ll_stat_blksize)
4953 stat->blksize = sbi->ll_stat_blksize;
4954 else if (S_ISREG(inode->i_mode))
4955 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
4956 LL_MAX_BLKSIZE_BITS);
4958 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
4960 stat->nlink = inode->i_nlink;
4961 stat->size = i_size_read(inode);
4962 stat->blocks = inode->i_blocks;
4964 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
4965 if (flags & AT_STATX_DONT_SYNC) {
4966 if (stat->size == 0 &&
4967 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
4968 stat->size = lli->lli_lazysize;
4969 if (stat->blocks == 0 &&
4970 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
4971 stat->blocks = lli->lli_lazyblocks;
4974 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
4975 stat->result_mask |= STATX_BTIME;
4976 stat->btime.tv_sec = lli->lli_btime;
4979 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
4980 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
4981 stat->result_mask &= request_mask;
4984 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
4985 ktime_us_delta(ktime_get(), kstart));
4990 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
4991 int ll_getattr(const struct path *path, struct kstat *stat,
4992 u32 request_mask, unsigned int flags)
4994 return ll_getattr_dentry(path->dentry, stat, request_mask, flags);
4997 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
4999 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5000 AT_STATX_SYNC_AS_STAT);
5004 int cl_falloc(struct inode *inode, int mode, loff_t offset, loff_t len)
5009 int rc; loff_t sa_falloc_end;
5010 loff_t size = i_size_read(inode);
5014 env = cl_env_get(&refcheck);
5016 RETURN(PTR_ERR(env));
5018 io = vvp_env_thread_io(env);
5019 io->ci_obj = ll_i2info(inode)->lli_clob;
5020 io->ci_verify_layout = 1;
5021 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5022 io->u.ci_setattr.sa_falloc_mode = mode;
5023 io->u.ci_setattr.sa_falloc_offset = offset;
5024 io->u.ci_setattr.sa_falloc_len = len;
5025 io->u.ci_setattr.sa_falloc_end = io->u.ci_setattr.sa_falloc_offset +
5026 io->u.ci_setattr.sa_falloc_len;
5027 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5028 sa_falloc_end = io->u.ci_setattr.sa_falloc_end;
5029 if (sa_falloc_end > size) {
5030 /* Check new size against VFS/VM file size limit and rlimit */
5031 rc = inode_newsize_ok(inode, sa_falloc_end);
5034 if (sa_falloc_end > ll_file_maxbytes(inode)) {
5035 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5036 (unsigned long long)(sa_falloc_end),
5037 ll_file_maxbytes(inode));
5041 io->u.ci_setattr.sa_attr.lvb_size = sa_falloc_end;
5042 if (!(mode & FALLOC_FL_KEEP_SIZE))
5043 io->u.ci_setattr.sa_avalid |= ATTR_SIZE;
5045 io->u.ci_setattr.sa_attr.lvb_size = size;
5049 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0)
5050 rc = cl_io_loop(env, io);
5054 cl_io_fini(env, io);
5055 if (unlikely(io->ci_need_restart))
5059 cl_env_put(env, &refcheck);
5063 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5065 struct inode *inode = filp->f_path.dentry->d_inode;
5069 * Encrypted inodes can't handle collapse range or zero range or insert
5070 * range since we would need to re-encrypt blocks with a different IV or
5071 * XTS tweak (which are based on the logical block number).
5072 * Similar to what ext4 does.
5074 if (IS_ENCRYPTED(inode) &&
5075 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5076 FALLOC_FL_ZERO_RANGE)))
5077 RETURN(-EOPNOTSUPP);
5080 * Only mode == 0 (which is standard prealloc) is supported now.
5081 * Punch is not supported yet.
5083 if (mode & ~FALLOC_FL_KEEP_SIZE)
5084 RETURN(-EOPNOTSUPP);
5086 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5088 rc = cl_falloc(inode, mode, offset, len);
5093 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5094 __u64 start, __u64 len)
5098 struct fiemap *fiemap;
5099 unsigned int extent_count = fieinfo->fi_extents_max;
5101 num_bytes = sizeof(*fiemap) + (extent_count *
5102 sizeof(struct fiemap_extent));
5103 OBD_ALLOC_LARGE(fiemap, num_bytes);
5108 fiemap->fm_flags = fieinfo->fi_flags;
5109 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5110 fiemap->fm_start = start;
5111 fiemap->fm_length = len;
5112 if (extent_count > 0 &&
5113 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5114 sizeof(struct fiemap_extent)) != 0)
5115 GOTO(out, rc = -EFAULT);
5117 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5119 fieinfo->fi_flags = fiemap->fm_flags;
5120 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5121 if (extent_count > 0 &&
5122 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5123 fiemap->fm_mapped_extents *
5124 sizeof(struct fiemap_extent)) != 0)
5125 GOTO(out, rc = -EFAULT);
5127 OBD_FREE_LARGE(fiemap, num_bytes);
5131 struct posix_acl *ll_get_acl(struct inode *inode, int type)
5133 struct ll_inode_info *lli = ll_i2info(inode);
5134 struct posix_acl *acl = NULL;
5137 spin_lock(&lli->lli_lock);
5138 /* VFS' acl_permission_check->check_acl will release the refcount */
5139 acl = posix_acl_dup(lli->lli_posix_acl);
5140 spin_unlock(&lli->lli_lock);
5145 #ifdef HAVE_IOP_SET_ACL
5146 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
5147 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type)
5149 struct ll_sb_info *sbi = ll_i2sbi(inode);
5150 struct ptlrpc_request *req = NULL;
5151 const char *name = NULL;
5153 size_t value_size = 0;
5158 case ACL_TYPE_ACCESS:
5159 name = XATTR_NAME_POSIX_ACL_ACCESS;
5161 rc = posix_acl_update_mode(inode, &inode->i_mode, &acl);
5164 case ACL_TYPE_DEFAULT:
5165 name = XATTR_NAME_POSIX_ACL_DEFAULT;
5166 if (!S_ISDIR(inode->i_mode))
5167 rc = acl ? -EACCES : 0;
5178 value_size = posix_acl_xattr_size(acl->a_count);
5179 value = kmalloc(value_size, GFP_NOFS);
5181 GOTO(out, rc = -ENOMEM);
5183 rc = posix_acl_to_xattr(&init_user_ns, acl, value, value_size);
5185 GOTO(out_value, rc);
5188 rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode),
5189 value ? OBD_MD_FLXATTR : OBD_MD_FLXATTRRM,
5190 name, value, value_size, 0, 0, &req);
5192 ptlrpc_req_finished(req);
5197 forget_cached_acl(inode, type);
5199 set_cached_acl(inode, type, acl);
5202 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
5203 #endif /* HAVE_IOP_SET_ACL */
5205 int ll_inode_permission(struct inode *inode, int mask)
5208 struct ll_sb_info *sbi;
5209 struct root_squash_info *squash;
5210 struct cred *cred = NULL;
5211 const struct cred *old_cred = NULL;
5213 bool squash_id = false;
5214 ktime_t kstart = ktime_get();
5218 if (mask & MAY_NOT_BLOCK)
5222 * as root inode are NOT getting validated in lookup operation,
5223 * need to do it before permission check.
5226 if (inode == inode->i_sb->s_root->d_inode) {
5227 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
5232 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5233 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5235 /* squash fsuid/fsgid if needed */
5236 sbi = ll_i2sbi(inode);
5237 squash = &sbi->ll_squash;
5238 if (unlikely(squash->rsi_uid != 0 &&
5239 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5240 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
5244 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5245 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5246 squash->rsi_uid, squash->rsi_gid);
5248 /* update current process's credentials
5249 * and FS capability */
5250 cred = prepare_creds();
5254 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5255 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5256 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
5257 if (BIT(cap) & CFS_CAP_FS_MASK)
5258 cap_lower(cred->cap_effective, cap);
5260 old_cred = override_creds(cred);
5263 rc = generic_permission(inode, mask);
5264 /* restore current process's credentials and FS capability */
5266 revert_creds(old_cred);
5271 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5272 ktime_us_delta(ktime_get(), kstart));
5277 /* -o localflock - only provides locally consistent flock locks */
5278 struct file_operations ll_file_operations = {
5279 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5280 # ifdef HAVE_SYNC_READ_WRITE
5281 .read = new_sync_read,
5282 .write = new_sync_write,
5284 .read_iter = ll_file_read_iter,
5285 .write_iter = ll_file_write_iter,
5286 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5287 .read = ll_file_read,
5288 .aio_read = ll_file_aio_read,
5289 .write = ll_file_write,
5290 .aio_write = ll_file_aio_write,
5291 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5292 .unlocked_ioctl = ll_file_ioctl,
5293 .open = ll_file_open,
5294 .release = ll_file_release,
5295 .mmap = ll_file_mmap,
5296 .llseek = ll_file_seek,
5297 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5298 .splice_read = generic_file_splice_read,
5300 .splice_read = pcc_file_splice_read,
5304 .fallocate = ll_fallocate,
5307 struct file_operations ll_file_operations_flock = {
5308 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5309 # ifdef HAVE_SYNC_READ_WRITE
5310 .read = new_sync_read,
5311 .write = new_sync_write,
5312 # endif /* HAVE_SYNC_READ_WRITE */
5313 .read_iter = ll_file_read_iter,
5314 .write_iter = ll_file_write_iter,
5315 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5316 .read = ll_file_read,
5317 .aio_read = ll_file_aio_read,
5318 .write = ll_file_write,
5319 .aio_write = ll_file_aio_write,
5320 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5321 .unlocked_ioctl = ll_file_ioctl,
5322 .open = ll_file_open,
5323 .release = ll_file_release,
5324 .mmap = ll_file_mmap,
5325 .llseek = ll_file_seek,
5326 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5327 .splice_read = generic_file_splice_read,
5329 .splice_read = pcc_file_splice_read,
5333 .flock = ll_file_flock,
5334 .lock = ll_file_flock,
5335 .fallocate = ll_fallocate,
5338 /* These are for -o noflock - to return ENOSYS on flock calls */
5339 struct file_operations ll_file_operations_noflock = {
5340 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5341 # ifdef HAVE_SYNC_READ_WRITE
5342 .read = new_sync_read,
5343 .write = new_sync_write,
5344 # endif /* HAVE_SYNC_READ_WRITE */
5345 .read_iter = ll_file_read_iter,
5346 .write_iter = ll_file_write_iter,
5347 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5348 .read = ll_file_read,
5349 .aio_read = ll_file_aio_read,
5350 .write = ll_file_write,
5351 .aio_write = ll_file_aio_write,
5352 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5353 .unlocked_ioctl = ll_file_ioctl,
5354 .open = ll_file_open,
5355 .release = ll_file_release,
5356 .mmap = ll_file_mmap,
5357 .llseek = ll_file_seek,
5358 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5359 .splice_read = generic_file_splice_read,
5361 .splice_read = pcc_file_splice_read,
5365 .flock = ll_file_noflock,
5366 .lock = ll_file_noflock,
5367 .fallocate = ll_fallocate,
5370 struct inode_operations ll_file_inode_operations = {
5371 .setattr = ll_setattr,
5372 .getattr = ll_getattr,
5373 .permission = ll_inode_permission,
5374 #ifdef HAVE_IOP_XATTR
5375 .setxattr = ll_setxattr,
5376 .getxattr = ll_getxattr,
5377 .removexattr = ll_removexattr,
5379 .listxattr = ll_listxattr,
5380 .fiemap = ll_fiemap,
5381 .get_acl = ll_get_acl,
5382 #ifdef HAVE_IOP_SET_ACL
5383 .set_acl = ll_set_acl,
5387 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5389 struct ll_inode_info *lli = ll_i2info(inode);
5390 struct cl_object *obj = lli->lli_clob;
5399 env = cl_env_get(&refcheck);
5401 RETURN(PTR_ERR(env));
5403 rc = cl_conf_set(env, lli->lli_clob, conf);
5407 if (conf->coc_opc == OBJECT_CONF_SET) {
5408 struct ldlm_lock *lock = conf->coc_lock;
5409 struct cl_layout cl = {
5413 LASSERT(lock != NULL);
5414 LASSERT(ldlm_has_layout(lock));
5416 /* it can only be allowed to match after layout is
5417 * applied to inode otherwise false layout would be
5418 * seen. Applying layout shoud happen before dropping
5419 * the intent lock. */
5420 ldlm_lock_allow_match(lock);
5422 rc = cl_object_layout_get(env, obj, &cl);
5427 DFID": layout version change: %u -> %u\n",
5428 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5430 ll_layout_version_set(lli, cl.cl_layout_gen);
5434 cl_env_put(env, &refcheck);
5439 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5440 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5443 struct ll_sb_info *sbi = ll_i2sbi(inode);
5444 struct ptlrpc_request *req;
5451 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5452 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5453 lock->l_lvb_data, lock->l_lvb_len);
5455 if (lock->l_lvb_data != NULL)
5458 /* if layout lock was granted right away, the layout is returned
5459 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5460 * blocked and then granted via completion ast, we have to fetch
5461 * layout here. Please note that we can't use the LVB buffer in
5462 * completion AST because it doesn't have a large enough buffer */
5463 rc = ll_get_default_mdsize(sbi, &lmmsize);
5467 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5468 XATTR_NAME_LOV, lmmsize, &req);
5471 GOTO(out, rc = 0); /* empty layout */
5478 if (lmmsize == 0) /* empty layout */
5481 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5483 GOTO(out, rc = -EFAULT);
5485 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5486 if (lvbdata == NULL)
5487 GOTO(out, rc = -ENOMEM);
5489 memcpy(lvbdata, lmm, lmmsize);
5490 lock_res_and_lock(lock);
5491 if (unlikely(lock->l_lvb_data == NULL)) {
5492 lock->l_lvb_type = LVB_T_LAYOUT;
5493 lock->l_lvb_data = lvbdata;
5494 lock->l_lvb_len = lmmsize;
5497 unlock_res_and_lock(lock);
5500 OBD_FREE_LARGE(lvbdata, lmmsize);
5505 ptlrpc_req_finished(req);
5510 * Apply the layout to the inode. Layout lock is held and will be released
5513 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5514 struct inode *inode)
5516 struct ll_inode_info *lli = ll_i2info(inode);
5517 struct ll_sb_info *sbi = ll_i2sbi(inode);
5518 struct ldlm_lock *lock;
5519 struct cl_object_conf conf;
5522 bool wait_layout = false;
5525 LASSERT(lustre_handle_is_used(lockh));
5527 lock = ldlm_handle2lock(lockh);
5528 LASSERT(lock != NULL);
5529 LASSERT(ldlm_has_layout(lock));
5531 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5532 PFID(&lli->lli_fid), inode);
5534 /* in case this is a caching lock and reinstate with new inode */
5535 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5537 lock_res_and_lock(lock);
5538 lvb_ready = ldlm_is_lvb_ready(lock);
5539 unlock_res_and_lock(lock);
5541 /* checking lvb_ready is racy but this is okay. The worst case is
5542 * that multi processes may configure the file on the same time. */
5546 rc = ll_layout_fetch(inode, lock);
5550 /* for layout lock, lmm is stored in lock's lvb.
5551 * lvb_data is immutable if the lock is held so it's safe to access it
5554 * set layout to file. Unlikely this will fail as old layout was
5555 * surely eliminated */
5556 memset(&conf, 0, sizeof conf);
5557 conf.coc_opc = OBJECT_CONF_SET;
5558 conf.coc_inode = inode;
5559 conf.coc_lock = lock;
5560 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5561 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5562 rc = ll_layout_conf(inode, &conf);
5564 /* refresh layout failed, need to wait */
5565 wait_layout = rc == -EBUSY;
5568 LDLM_LOCK_PUT(lock);
5569 ldlm_lock_decref(lockh, mode);
5571 /* wait for IO to complete if it's still being used. */
5573 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5574 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5576 memset(&conf, 0, sizeof conf);
5577 conf.coc_opc = OBJECT_CONF_WAIT;
5578 conf.coc_inode = inode;
5579 rc = ll_layout_conf(inode, &conf);
5583 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5584 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5590 * Issue layout intent RPC to MDS.
5591 * \param inode [in] file inode
5592 * \param intent [in] layout intent
5594 * \retval 0 on success
5595 * \retval < 0 error code
5597 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5599 struct ll_inode_info *lli = ll_i2info(inode);
5600 struct ll_sb_info *sbi = ll_i2sbi(inode);
5601 struct md_op_data *op_data;
5602 struct lookup_intent it;
5603 struct ptlrpc_request *req;
5607 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5608 0, 0, LUSTRE_OPC_ANY, NULL);
5609 if (IS_ERR(op_data))
5610 RETURN(PTR_ERR(op_data));
5612 op_data->op_data = intent;
5613 op_data->op_data_size = sizeof(*intent);
5615 memset(&it, 0, sizeof(it));
5616 it.it_op = IT_LAYOUT;
5617 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5618 intent->li_opc == LAYOUT_INTENT_TRUNC)
5619 it.it_flags = FMODE_WRITE;
5621 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5622 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5624 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5625 &ll_md_blocking_ast, 0);
5626 if (it.it_request != NULL)
5627 ptlrpc_req_finished(it.it_request);
5628 it.it_request = NULL;
5630 ll_finish_md_op_data(op_data);
5632 /* set lock data in case this is a new lock */
5634 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5636 ll_intent_drop_lock(&it);
5642 * This function checks if there exists a LAYOUT lock on the client side,
5643 * or enqueues it if it doesn't have one in cache.
5645 * This function will not hold layout lock so it may be revoked any time after
5646 * this function returns. Any operations depend on layout should be redone
5649 * This function should be called before lov_io_init() to get an uptodate
5650 * layout version, the caller should save the version number and after IO
5651 * is finished, this function should be called again to verify that layout
5652 * is not changed during IO time.
5654 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5656 struct ll_inode_info *lli = ll_i2info(inode);
5657 struct ll_sb_info *sbi = ll_i2sbi(inode);
5658 struct lustre_handle lockh;
5659 struct layout_intent intent = {
5660 .li_opc = LAYOUT_INTENT_ACCESS,
5662 enum ldlm_mode mode;
5666 *gen = ll_layout_version_get(lli);
5667 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
5671 LASSERT(fid_is_sane(ll_inode2fid(inode)));
5672 LASSERT(S_ISREG(inode->i_mode));
5674 /* take layout lock mutex to enqueue layout lock exclusively. */
5675 mutex_lock(&lli->lli_layout_mutex);
5678 /* mostly layout lock is caching on the local side, so try to
5679 * match it before grabbing layout lock mutex. */
5680 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
5681 LCK_CR | LCK_CW | LCK_PR |
5683 if (mode != 0) { /* hit cached lock */
5684 rc = ll_layout_lock_set(&lockh, mode, inode);
5690 rc = ll_layout_intent(inode, &intent);
5696 *gen = ll_layout_version_get(lli);
5697 mutex_unlock(&lli->lli_layout_mutex);
5703 * Issue layout intent RPC indicating where in a file an IO is about to write.
5705 * \param[in] inode file inode.
5706 * \param[in] ext write range with start offset of fille in bytes where
5707 * an IO is about to write, and exclusive end offset in
5710 * \retval 0 on success
5711 * \retval < 0 error code
5713 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
5714 struct lu_extent *ext)
5716 struct layout_intent intent = {
5718 .li_extent.e_start = ext->e_start,
5719 .li_extent.e_end = ext->e_end,
5724 rc = ll_layout_intent(inode, &intent);
5730 * This function send a restore request to the MDT
5732 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
5734 struct hsm_user_request *hur;
5738 len = sizeof(struct hsm_user_request) +
5739 sizeof(struct hsm_user_item);
5740 OBD_ALLOC(hur, len);
5744 hur->hur_request.hr_action = HUA_RESTORE;
5745 hur->hur_request.hr_archive_id = 0;
5746 hur->hur_request.hr_flags = 0;
5747 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
5748 sizeof(hur->hur_user_item[0].hui_fid));
5749 hur->hur_user_item[0].hui_extent.offset = offset;
5750 hur->hur_user_item[0].hui_extent.length = length;
5751 hur->hur_request.hr_itemcount = 1;
5752 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,