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/
33 * Author: Peter Braam <braam@clusterfs.com>
34 * Author: Phil Schwan <phil@clusterfs.com>
35 * Author: Andreas Dilger <adilger@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <lustre_dlm.h>
40 #include <linux/pagemap.h>
41 #include <linux/file.h>
42 #include <linux/sched.h>
43 #include <linux/user_namespace.h>
44 #include <linux/uidgid.h>
45 #include <linux/falloc.h>
46 #include <linux/ktime.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <lustre_swab.h>
51 #include "cl_object.h"
52 #include "llite_internal.h"
53 #include "vvp_internal.h"
56 struct inode *sp_inode;
61 __u64 pa_data_version;
67 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
69 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
72 static struct ll_file_data *ll_file_data_get(void)
74 struct ll_file_data *fd;
76 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
80 fd->fd_write_failed = false;
81 pcc_file_init(&fd->fd_pcc_file);
86 static void ll_file_data_put(struct ll_file_data *fd)
89 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
93 * Packs all the attributes into @op_data for the CLOSE rpc.
95 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
96 struct obd_client_handle *och)
100 ll_prep_md_op_data(op_data, inode, NULL, NULL,
101 0, 0, LUSTRE_OPC_ANY, NULL);
103 op_data->op_attr.ia_mode = inode->i_mode;
104 op_data->op_attr.ia_atime = inode->i_atime;
105 op_data->op_attr.ia_mtime = inode->i_mtime;
106 op_data->op_attr.ia_ctime = inode->i_ctime;
107 /* In case of encrypted file without the key, visible size was rounded
108 * up to next LUSTRE_ENCRYPTION_UNIT_SIZE, and clear text size was
109 * stored into lli_lazysize in ll_merge_attr(), so set proper file size
110 * now that we are closing.
112 if (llcrypt_require_key(inode) == -ENOKEY &&
113 ll_i2info(inode)->lli_attr_valid & OBD_MD_FLLAZYSIZE)
114 op_data->op_attr.ia_size = ll_i2info(inode)->lli_lazysize;
116 op_data->op_attr.ia_size = i_size_read(inode);
117 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
118 ATTR_MTIME | ATTR_MTIME_SET |
120 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
121 op_data->op_attr_blocks = inode->i_blocks;
122 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
123 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
124 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
125 op_data->op_open_handle = och->och_open_handle;
127 if (och->och_flags & FMODE_WRITE &&
128 test_and_clear_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags))
129 /* For HSM: if inode data has been modified, pack it so that
130 * MDT can set data dirty flag in the archive. */
131 op_data->op_bias |= MDS_DATA_MODIFIED;
137 * Perform a close, possibly with a bias.
138 * The meaning of "data" depends on the value of "bias".
140 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
141 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
144 static int ll_close_inode_openhandle(struct inode *inode,
145 struct obd_client_handle *och,
146 enum mds_op_bias bias, void *data)
148 struct obd_export *md_exp = ll_i2mdexp(inode);
149 const struct ll_inode_info *lli = ll_i2info(inode);
150 struct md_op_data *op_data;
151 struct ptlrpc_request *req = NULL;
155 if (class_exp2obd(md_exp) == NULL) {
156 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
157 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
161 OBD_ALLOC_PTR(op_data);
162 /* We leak openhandle and request here on error, but not much to be
163 * done in OOM case since app won't retry close on error either. */
165 GOTO(out, rc = -ENOMEM);
167 ll_prepare_close(inode, op_data, och);
169 case MDS_CLOSE_LAYOUT_MERGE:
170 /* merge blocks from the victim inode */
171 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
172 op_data->op_attr.ia_valid |= ATTR_SIZE;
173 op_data->op_xvalid |= OP_XVALID_BLOCKS;
175 case MDS_CLOSE_LAYOUT_SPLIT:
176 case MDS_CLOSE_LAYOUT_SWAP: {
177 struct split_param *sp = data;
179 LASSERT(data != NULL);
180 op_data->op_bias |= bias;
181 op_data->op_data_version = 0;
182 op_data->op_lease_handle = och->och_lease_handle;
183 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
184 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
185 op_data->op_mirror_id = sp->sp_mirror_id;
187 op_data->op_fid2 = *ll_inode2fid(data);
192 case MDS_CLOSE_RESYNC_DONE: {
193 struct ll_ioc_lease *ioc = data;
195 LASSERT(data != NULL);
196 op_data->op_attr_blocks +=
197 ioc->lil_count * op_data->op_attr_blocks;
198 op_data->op_attr.ia_valid |= ATTR_SIZE;
199 op_data->op_xvalid |= OP_XVALID_BLOCKS;
200 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
202 op_data->op_lease_handle = och->och_lease_handle;
203 op_data->op_data = &ioc->lil_ids[0];
204 op_data->op_data_size =
205 ioc->lil_count * sizeof(ioc->lil_ids[0]);
209 case MDS_PCC_ATTACH: {
210 struct pcc_param *param = data;
212 LASSERT(data != NULL);
213 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
214 op_data->op_archive_id = param->pa_archive_id;
215 op_data->op_data_version = param->pa_data_version;
216 op_data->op_lease_handle = och->och_lease_handle;
220 case MDS_HSM_RELEASE:
221 LASSERT(data != NULL);
222 op_data->op_bias |= MDS_HSM_RELEASE;
223 op_data->op_data_version = *(__u64 *)data;
224 op_data->op_lease_handle = och->och_lease_handle;
225 op_data->op_attr.ia_valid |= ATTR_SIZE;
226 op_data->op_xvalid |= OP_XVALID_BLOCKS;
230 LASSERT(data == NULL);
234 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
235 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
236 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
237 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
239 rc = md_close(md_exp, op_data, och->och_mod, &req);
240 if (rc != 0 && rc != -EINTR)
241 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
242 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
244 if (rc == 0 && op_data->op_bias & bias) {
245 struct mdt_body *body;
247 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
248 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
251 if (bias & MDS_PCC_ATTACH) {
252 struct pcc_param *param = data;
254 param->pa_layout_gen = body->mbo_layout_gen;
258 ll_finish_md_op_data(op_data);
262 md_clear_open_replay_data(md_exp, och);
263 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
266 ptlrpc_req_finished(req); /* This is close request */
270 int ll_md_real_close(struct inode *inode, fmode_t fmode)
272 struct ll_inode_info *lli = ll_i2info(inode);
273 struct obd_client_handle **och_p;
274 struct obd_client_handle *och;
279 if (fmode & FMODE_WRITE) {
280 och_p = &lli->lli_mds_write_och;
281 och_usecount = &lli->lli_open_fd_write_count;
282 } else if (fmode & FMODE_EXEC) {
283 och_p = &lli->lli_mds_exec_och;
284 och_usecount = &lli->lli_open_fd_exec_count;
286 LASSERT(fmode & FMODE_READ);
287 och_p = &lli->lli_mds_read_och;
288 och_usecount = &lli->lli_open_fd_read_count;
291 mutex_lock(&lli->lli_och_mutex);
292 if (*och_usecount > 0) {
293 /* There are still users of this handle, so skip
295 mutex_unlock(&lli->lli_och_mutex);
301 mutex_unlock(&lli->lli_och_mutex);
304 /* There might be a race and this handle may already
306 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
312 static int ll_md_close(struct inode *inode, struct file *file)
314 union ldlm_policy_data policy = {
315 .l_inodebits = { MDS_INODELOCK_OPEN },
317 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
318 struct ll_file_data *fd = file->private_data;
319 struct ll_inode_info *lli = ll_i2info(inode);
320 struct lustre_handle lockh;
321 enum ldlm_mode lockmode;
325 /* clear group lock, if present */
326 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
327 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
329 mutex_lock(&lli->lli_och_mutex);
330 if (fd->fd_lease_och != NULL) {
332 struct obd_client_handle *lease_och;
334 lease_och = fd->fd_lease_och;
335 fd->fd_lease_och = NULL;
336 mutex_unlock(&lli->lli_och_mutex);
338 /* Usually the lease is not released when the
339 * application crashed, we need to release here. */
340 rc = ll_lease_close(lease_och, inode, &lease_broken);
342 mutex_lock(&lli->lli_och_mutex);
344 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
345 "Clean up lease "DFID" %d/%d\n",
346 PFID(&lli->lli_fid), rc, lease_broken);
349 if (fd->fd_och != NULL) {
350 struct obd_client_handle *och;
354 mutex_unlock(&lli->lli_och_mutex);
356 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
360 /* Let's see if we have good enough OPEN lock on the file and if
361 we can skip talking to MDS */
362 if (fd->fd_omode & FMODE_WRITE) {
364 LASSERT(lli->lli_open_fd_write_count);
365 lli->lli_open_fd_write_count--;
366 } else if (fd->fd_omode & FMODE_EXEC) {
368 LASSERT(lli->lli_open_fd_exec_count);
369 lli->lli_open_fd_exec_count--;
372 LASSERT(lli->lli_open_fd_read_count);
373 lli->lli_open_fd_read_count--;
375 mutex_unlock(&lli->lli_och_mutex);
377 /* LU-4398: do not cache write open lock if the file has exec bit */
378 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
379 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
380 LDLM_IBITS, &policy, lockmode, &lockh))
381 rc = ll_md_real_close(inode, fd->fd_omode);
384 file->private_data = NULL;
385 ll_file_data_put(fd);
390 /* While this returns an error code, fput() the caller does not, so we need
391 * to make every effort to clean up all of our state here. Also, applications
392 * rarely check close errors and even if an error is returned they will not
393 * re-try the close call.
395 int ll_file_release(struct inode *inode, struct file *file)
397 struct ll_file_data *fd;
398 struct ll_sb_info *sbi = ll_i2sbi(inode);
399 struct ll_inode_info *lli = ll_i2info(inode);
400 ktime_t kstart = ktime_get();
405 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
406 PFID(ll_inode2fid(inode)), inode);
408 fd = file->private_data;
411 /* The last ref on @file, maybe not the the owner pid of statahead,
412 * because parent and child process can share the same file handle. */
413 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
414 ll_deauthorize_statahead(inode, fd);
416 if (is_root_inode(inode)) {
417 file->private_data = NULL;
418 ll_file_data_put(fd);
422 pcc_file_release(inode, file);
424 if (!S_ISDIR(inode->i_mode)) {
425 if (lli->lli_clob != NULL)
426 lov_read_and_clear_async_rc(lli->lli_clob);
427 lli->lli_async_rc = 0;
430 lli->lli_close_fd_time = ktime_get();
432 rc = ll_md_close(inode, file);
434 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
435 libcfs_debug_dumplog();
438 if (!rc && !is_root_inode(inode))
439 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
440 ktime_us_delta(ktime_get(), kstart));
444 static inline int ll_dom_readpage(void *data, struct page *page)
446 /* since ll_dom_readpage is a page cache helper, it is safe to assume
447 * mapping and host pointers are set here
450 struct niobuf_local *lnb = data;
454 inode = page2inode(page);
456 kaddr = kmap_atomic(page);
457 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
458 if (lnb->lnb_len < PAGE_SIZE)
459 memset(kaddr + lnb->lnb_len, 0,
460 PAGE_SIZE - lnb->lnb_len);
461 kunmap_atomic(kaddr);
463 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
464 if (!llcrypt_has_encryption_key(inode)) {
465 CDEBUG(D_SEC, "no enc key for "DFID"\n",
466 PFID(ll_inode2fid(inode)));
469 unsigned int offs = 0;
471 while (offs < PAGE_SIZE) {
472 /* decrypt only if page is not empty */
473 if (memcmp(page_address(page) + offs,
474 page_address(ZERO_PAGE(0)),
475 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
478 rc = llcrypt_decrypt_pagecache_blocks(page,
479 LUSTRE_ENCRYPTION_UNIT_SIZE,
484 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
489 flush_dcache_page(page);
490 SetPageUptodate(page);
497 #ifdef HAVE_READ_CACHE_PAGE_WANTS_FILE
498 static inline int ll_dom_read_folio(struct file *file, struct folio *folio0)
500 return ll_dom_readpage(file->private_data, folio_page(folio0, 0));
503 #define ll_dom_read_folio ll_dom_readpage
506 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
510 struct ll_inode_info *lli = ll_i2info(inode);
511 struct cl_object *obj = lli->lli_clob;
512 struct address_space *mapping = inode->i_mapping;
514 struct niobuf_remote *rnb;
515 struct mdt_body *body;
517 unsigned long index, start;
518 struct niobuf_local lnb;
527 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
531 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
532 if (rnb == NULL || rnb->rnb_len == 0)
535 /* LU-11595: Server may return whole file and that is OK always or
536 * it may return just file tail and its offset must be aligned with
537 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
538 * smaller then offset may be not aligned and that data is just ignored.
540 if (rnb->rnb_offset & ~PAGE_MASK)
543 /* Server returns whole file or just file tail if it fills in reply
544 * buffer, in both cases total size should be equal to the file size.
546 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
547 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
548 !(inode && IS_ENCRYPTED(inode))) {
549 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
550 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
551 rnb->rnb_len, body->mbo_dom_size);
555 env = cl_env_get(&refcheck);
558 io = vvp_env_thread_io(env);
560 io->ci_ignore_layout = 1;
561 rc = cl_io_init(env, io, CIT_MISC, obj);
565 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
566 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
568 data = (char *)rnb + sizeof(*rnb);
570 lnb.lnb_file_offset = rnb->rnb_offset;
571 start = lnb.lnb_file_offset >> PAGE_SHIFT;
573 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
574 lnb.lnb_page_offset = 0;
576 struct cl_page *page;
578 lnb.lnb_data = data + (index << PAGE_SHIFT);
579 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
580 if (lnb.lnb_len > PAGE_SIZE)
581 lnb.lnb_len = PAGE_SIZE;
583 vmpage = ll_read_cache_page(mapping, index + start,
584 ll_dom_read_folio, &lnb);
585 if (IS_ERR(vmpage)) {
586 CWARN("%s: cannot fill page %lu for "DFID
587 " with data: rc = %li\n",
588 ll_i2sbi(inode)->ll_fsname, index + start,
589 PFID(lu_object_fid(&obj->co_lu)),
594 if (vmpage->mapping == NULL) {
597 /* page was truncated */
600 /* attach VM page to CL page cache */
601 page = cl_page_find(env, obj, vmpage->index, vmpage,
604 ClearPageUptodate(vmpage);
609 SetPageUptodate(vmpage);
610 cl_page_put(env, page);
614 } while (rnb->rnb_len > (index << PAGE_SHIFT));
618 cl_env_put(env, &refcheck);
623 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
624 struct lookup_intent *itp)
626 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
627 struct dentry *parent = de->d_parent;
630 struct md_op_data *op_data;
631 struct ptlrpc_request *req = NULL;
635 LASSERT(parent != NULL);
636 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
638 /* if server supports open-by-fid, or file name is invalid, don't pack
639 * name in open request */
640 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
641 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
643 len = de->d_name.len;
644 name = kmalloc(len + 1, GFP_NOFS);
649 spin_lock(&de->d_lock);
650 if (len != de->d_name.len) {
651 spin_unlock(&de->d_lock);
655 memcpy(name, de->d_name.name, len);
657 spin_unlock(&de->d_lock);
659 if (!lu_name_is_valid_2(name, len)) {
665 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
666 name, len, 0, LUSTRE_OPC_OPEN, NULL);
667 if (IS_ERR(op_data)) {
669 RETURN(PTR_ERR(op_data));
671 op_data->op_data = lmm;
672 op_data->op_data_size = lmmsize;
674 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
676 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
677 &ll_md_blocking_ast, 0);
679 ll_finish_md_op_data(op_data);
681 /* reason for keep own exit path - don`t flood log
682 * with messages with -ESTALE errors.
684 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
685 it_open_error(DISP_OPEN_OPEN, itp))
687 ll_release_openhandle(de, itp);
691 if (it_disposition(itp, DISP_LOOKUP_NEG))
692 GOTO(out, rc = -ENOENT);
694 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
695 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
696 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
700 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
702 if (!rc && itp->it_lock_mode) {
705 /* If we got a lock back and it has a LOOKUP bit set,
706 * make sure the dentry is marked as valid so we can find it.
707 * We don't need to care about actual hashing since other bits
708 * of kernel will deal with that later.
710 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
711 if (bits & MDS_INODELOCK_LOOKUP) {
712 d_lustre_revalidate(de);
713 ll_update_dir_depth(parent->d_inode, de->d_inode);
716 /* if DoM bit returned along with LAYOUT bit then there
717 * can be read-on-open data returned.
719 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
720 ll_dom_finish_open(de->d_inode, req);
724 ptlrpc_req_finished(req);
725 ll_intent_drop_lock(itp);
727 /* We did open by fid, but by the time we got to the server, the object
728 * disappeared. This is possible if the object was unlinked, but it's
729 * also possible if the object was unlinked by a rename. In the case
730 * of an object renamed over our existing one, we can't fail this open.
731 * O_CREAT also goes through this path if we had an existing dentry,
732 * and it's obviously wrong to return ENOENT for O_CREAT.
734 * Instead let's return -ESTALE, and the VFS will retry the open with
735 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
736 * revalidate, causing a lookup. This causes extra lookups in the case
737 * where we had a dentry in cache but the file is being unlinked and we
738 * lose the race with unlink, but this should be very rare.
746 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
747 struct obd_client_handle *och)
749 struct mdt_body *body;
751 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
752 och->och_open_handle = body->mbo_open_handle;
753 och->och_fid = body->mbo_fid1;
754 och->och_lease_handle.cookie = it->it_lock_handle;
755 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
756 och->och_flags = it->it_flags;
758 return md_set_open_replay_data(md_exp, och, it);
761 static int ll_local_open(struct file *file, struct lookup_intent *it,
762 struct ll_file_data *fd, struct obd_client_handle *och)
764 struct inode *inode = file_inode(file);
767 LASSERT(!file->private_data);
774 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
779 file->private_data = fd;
780 ll_readahead_init(inode, &fd->fd_ras);
781 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
786 void ll_track_file_opens(struct inode *inode)
788 struct ll_inode_info *lli = ll_i2info(inode);
789 struct ll_sb_info *sbi = ll_i2sbi(inode);
791 /* do not skew results with delays from never-opened inodes */
792 if (ktime_to_ns(lli->lli_close_fd_time))
793 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
794 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
796 if (ktime_after(ktime_get(),
797 ktime_add_ms(lli->lli_close_fd_time,
798 sbi->ll_oc_max_ms))) {
799 lli->lli_open_fd_count = 1;
800 lli->lli_close_fd_time = ns_to_ktime(0);
802 lli->lli_open_fd_count++;
805 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
806 lli->lli_open_fd_count);
809 /* Open a file, and (for the very first open) create objects on the OSTs at
810 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
811 * creation or open until ll_lov_setstripe() ioctl is called.
813 * If we already have the stripe MD locally then we don't request it in
814 * md_open(), by passing a lmm_size = 0.
816 * It is up to the application to ensure no other processes open this file
817 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
818 * used. We might be able to avoid races of that sort by getting lli_open_sem
819 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
820 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
822 int ll_file_open(struct inode *inode, struct file *file)
824 struct ll_inode_info *lli = ll_i2info(inode);
825 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
826 .it_flags = file->f_flags };
827 struct obd_client_handle **och_p = NULL;
828 __u64 *och_usecount = NULL;
829 struct ll_file_data *fd;
830 ktime_t kstart = ktime_get();
834 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
835 PFID(ll_inode2fid(inode)), inode, file->f_flags);
837 it = file->private_data; /* XXX: compat macro */
838 file->private_data = NULL; /* prevent ll_local_open assertion */
840 if (S_ISREG(inode->i_mode)) {
841 rc = ll_file_open_encrypt(inode, file);
843 if (it && it->it_disposition)
844 ll_release_openhandle(file_dentry(file), it);
845 GOTO(out_nofiledata, rc);
849 fd = ll_file_data_get();
851 GOTO(out_nofiledata, rc = -ENOMEM);
854 if (S_ISDIR(inode->i_mode))
855 ll_authorize_statahead(inode, fd);
857 ll_track_file_opens(inode);
858 if (is_root_inode(inode)) {
859 file->private_data = fd;
863 if (!it || !it->it_disposition) {
864 /* Convert f_flags into access mode. We cannot use file->f_mode,
865 * because everything but O_ACCMODE mask was stripped from
867 if ((oit.it_flags + 1) & O_ACCMODE)
869 if (file->f_flags & O_TRUNC)
870 oit.it_flags |= FMODE_WRITE;
872 /* kernel only call f_op->open in dentry_open. filp_open calls
873 * dentry_open after call to open_namei that checks permissions.
874 * Only nfsd_open call dentry_open directly without checking
875 * permissions and because of that this code below is safe.
877 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
878 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
880 /* We do not want O_EXCL here, presumably we opened the file
881 * already? XXX - NFS implications? */
882 oit.it_flags &= ~O_EXCL;
884 /* bug20584, if "it_flags" contains O_CREAT, the file will be
885 * created if necessary, then "IT_CREAT" should be set to keep
886 * consistent with it */
887 if (oit.it_flags & O_CREAT)
888 oit.it_op |= IT_CREAT;
894 /* Let's see if we have file open on MDS already. */
895 if (it->it_flags & FMODE_WRITE) {
896 och_p = &lli->lli_mds_write_och;
897 och_usecount = &lli->lli_open_fd_write_count;
898 } else if (it->it_flags & FMODE_EXEC) {
899 och_p = &lli->lli_mds_exec_och;
900 och_usecount = &lli->lli_open_fd_exec_count;
902 och_p = &lli->lli_mds_read_och;
903 och_usecount = &lli->lli_open_fd_read_count;
906 mutex_lock(&lli->lli_och_mutex);
907 if (*och_p) { /* Open handle is present */
908 if (it_disposition(it, DISP_OPEN_OPEN)) {
909 /* Well, there's extra open request that we do not need,
910 * let's close it somehow. This will decref request. */
911 rc = it_open_error(DISP_OPEN_OPEN, it);
913 mutex_unlock(&lli->lli_och_mutex);
914 GOTO(out_openerr, rc);
917 ll_release_openhandle(file_dentry(file), it);
921 rc = ll_local_open(file, it, fd, NULL);
924 mutex_unlock(&lli->lli_och_mutex);
925 GOTO(out_openerr, rc);
928 LASSERT(*och_usecount == 0);
929 if (!it->it_disposition) {
930 struct dentry *dentry = file_dentry(file);
931 struct ll_sb_info *sbi = ll_i2sbi(inode);
932 int open_threshold = sbi->ll_oc_thrsh_count;
934 /* We cannot just request lock handle now, new ELC code
935 * means that one of other OPEN locks for this file
936 * could be cancelled, and since blocking ast handler
937 * would attempt to grab och_mutex as well, that would
938 * result in a deadlock
940 mutex_unlock(&lli->lli_och_mutex);
942 * Normally called under two situations:
943 * 1. fhandle / NFS export.
944 * 2. A race/condition on MDS resulting in no open
945 * handle to be returned from LOOKUP|OPEN request,
946 * for example if the target entry was a symlink.
948 * For NFSv3 we need to always cache the open lock
949 * for pre 5.5 Linux kernels.
951 * After reaching number of opens of this inode
952 * we always ask for an open lock on it to handle
953 * bad userspace actors that open and close files
954 * in a loop for absolutely no good reason
956 /* fhandle / NFS path. */
957 if (lli->lli_open_thrsh_count != UINT_MAX)
958 open_threshold = lli->lli_open_thrsh_count;
960 if (filename_is_volatile(dentry->d_name.name,
963 /* There really is nothing here, but this
964 * make this more readable I think.
965 * We do not want openlock for volatile
966 * files under any circumstances
968 } else if (open_threshold > 0) {
969 /* Take MDS_OPEN_LOCK with many opens */
970 if (lli->lli_open_fd_count >= open_threshold)
971 it->it_flags |= MDS_OPEN_LOCK;
973 /* If this is open after we just closed */
974 else if (ktime_before(ktime_get(),
975 ktime_add_ms(lli->lli_close_fd_time,
976 sbi->ll_oc_thrsh_ms)))
977 it->it_flags |= MDS_OPEN_LOCK;
981 * Always specify MDS_OPEN_BY_FID because we don't want
982 * to get file with different fid.
984 it->it_flags |= MDS_OPEN_BY_FID;
985 rc = ll_intent_file_open(dentry, NULL, 0, it);
987 GOTO(out_openerr, rc);
991 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
993 GOTO(out_och_free, rc = -ENOMEM);
997 /* md_intent_lock() didn't get a request ref if there was an
998 * open error, so don't do cleanup on the request here
1000 /* XXX (green): Should not we bail out on any error here, not
1001 * just open error? */
1002 rc = it_open_error(DISP_OPEN_OPEN, it);
1004 GOTO(out_och_free, rc);
1006 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
1007 "inode %p: disposition %x, status %d\n", inode,
1008 it_disposition(it, ~0), it->it_status);
1010 rc = ll_local_open(file, it, fd, *och_p);
1012 GOTO(out_och_free, rc);
1015 rc = pcc_file_open(inode, file);
1017 GOTO(out_och_free, rc);
1019 mutex_unlock(&lli->lli_och_mutex);
1023 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1024 different kind of OPEN lock for this same inode gets cancelled
1025 by ldlm_cancel_lru */
1026 if (!S_ISREG(inode->i_mode))
1027 GOTO(out_och_free, rc);
1028 cl_lov_delay_create_clear(&file->f_flags);
1029 GOTO(out_och_free, rc);
1033 if (och_p && *och_p) {
1034 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1035 *och_p = NULL; /* OBD_FREE writes some magic there */
1038 mutex_unlock(&lli->lli_och_mutex);
1041 if (lli->lli_opendir_key == fd)
1042 ll_deauthorize_statahead(inode, fd);
1045 ll_file_data_put(fd);
1047 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1048 ktime_us_delta(ktime_get(), kstart));
1052 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1053 ptlrpc_req_finished(it->it_request);
1054 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1060 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1061 struct ldlm_lock_desc *desc, void *data, int flag)
1064 struct lustre_handle lockh;
1068 case LDLM_CB_BLOCKING:
1069 ldlm_lock2handle(lock, &lockh);
1070 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1072 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1076 case LDLM_CB_CANCELING:
1084 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1085 * and save it as fd->fd_och so as to force client to reopen the file even
1086 * if it has an open lock in cache already.
1088 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1089 struct lustre_handle *old_open_handle)
1091 struct ll_inode_info *lli = ll_i2info(inode);
1092 struct ll_file_data *fd = file->private_data;
1093 struct obd_client_handle **och_p;
1094 __u64 *och_usecount;
1098 /* Get the openhandle of the file */
1099 mutex_lock(&lli->lli_och_mutex);
1100 if (fd->fd_lease_och != NULL)
1101 GOTO(out_unlock, rc = -EBUSY);
1103 if (fd->fd_och == NULL) {
1104 if (file->f_mode & FMODE_WRITE) {
1105 LASSERT(lli->lli_mds_write_och != NULL);
1106 och_p = &lli->lli_mds_write_och;
1107 och_usecount = &lli->lli_open_fd_write_count;
1109 LASSERT(lli->lli_mds_read_och != NULL);
1110 och_p = &lli->lli_mds_read_och;
1111 och_usecount = &lli->lli_open_fd_read_count;
1114 if (*och_usecount > 1)
1115 GOTO(out_unlock, rc = -EBUSY);
1117 fd->fd_och = *och_p;
1122 *old_open_handle = fd->fd_och->och_open_handle;
1126 mutex_unlock(&lli->lli_och_mutex);
1131 * Release ownership on lli_mds_*_och when putting back a file lease.
1133 static int ll_lease_och_release(struct inode *inode, struct file *file)
1135 struct ll_inode_info *lli = ll_i2info(inode);
1136 struct ll_file_data *fd = file->private_data;
1137 struct obd_client_handle **och_p;
1138 struct obd_client_handle *old_och = NULL;
1139 __u64 *och_usecount;
1143 mutex_lock(&lli->lli_och_mutex);
1144 if (file->f_mode & FMODE_WRITE) {
1145 och_p = &lli->lli_mds_write_och;
1146 och_usecount = &lli->lli_open_fd_write_count;
1148 och_p = &lli->lli_mds_read_och;
1149 och_usecount = &lli->lli_open_fd_read_count;
1152 /* The file may have been open by another process (broken lease) so
1153 * *och_p is not NULL. In this case we should simply increase usecount
1156 if (*och_p != NULL) {
1157 old_och = fd->fd_och;
1160 *och_p = fd->fd_och;
1164 mutex_unlock(&lli->lli_och_mutex);
1166 if (old_och != NULL)
1167 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1173 * Acquire a lease and open the file.
1175 static struct obd_client_handle *
1176 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1179 struct lookup_intent it = { .it_op = IT_OPEN };
1180 struct ll_sb_info *sbi = ll_i2sbi(inode);
1181 struct md_op_data *op_data;
1182 struct ptlrpc_request *req = NULL;
1183 struct lustre_handle old_open_handle = { 0 };
1184 struct obd_client_handle *och = NULL;
1189 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1190 RETURN(ERR_PTR(-EINVAL));
1193 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1194 RETURN(ERR_PTR(-EPERM));
1196 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1198 RETURN(ERR_PTR(rc));
1203 RETURN(ERR_PTR(-ENOMEM));
1205 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1206 LUSTRE_OPC_ANY, NULL);
1207 if (IS_ERR(op_data))
1208 GOTO(out, rc = PTR_ERR(op_data));
1210 /* To tell the MDT this openhandle is from the same owner */
1211 op_data->op_open_handle = old_open_handle;
1213 it.it_flags = fmode | open_flags;
1214 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1215 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1216 &ll_md_blocking_lease_ast,
1217 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1218 * it can be cancelled which may mislead applications that the lease is
1220 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1221 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1222 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1223 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1224 ll_finish_md_op_data(op_data);
1225 ptlrpc_req_finished(req);
1227 GOTO(out_release_it, rc);
1229 if (it_disposition(&it, DISP_LOOKUP_NEG))
1230 GOTO(out_release_it, rc = -ENOENT);
1232 rc = it_open_error(DISP_OPEN_OPEN, &it);
1234 GOTO(out_release_it, rc);
1236 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1237 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1239 GOTO(out_release_it, rc);
1241 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1242 GOTO(out_close, rc = -EOPNOTSUPP);
1244 /* already get lease, handle lease lock */
1245 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1246 if (!it.it_lock_mode ||
1247 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1248 /* open lock must return for lease */
1249 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1250 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1252 GOTO(out_close, rc = -EPROTO);
1255 ll_intent_release(&it);
1259 /* Cancel open lock */
1260 if (it.it_lock_mode != 0) {
1261 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1263 it.it_lock_mode = 0;
1264 och->och_lease_handle.cookie = 0ULL;
1266 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1268 CERROR("%s: error closing file "DFID": %d\n",
1269 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1270 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1272 ll_intent_release(&it);
1276 RETURN(ERR_PTR(rc));
1280 * Check whether a layout swap can be done between two inodes.
1282 * \param[in] inode1 First inode to check
1283 * \param[in] inode2 Second inode to check
1285 * \retval 0 on success, layout swap can be performed between both inodes
1286 * \retval negative error code if requirements are not met
1288 static int ll_check_swap_layouts_validity(struct inode *inode1,
1289 struct inode *inode2)
1291 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1294 if (inode_permission(&init_user_ns, inode1, MAY_WRITE) ||
1295 inode_permission(&init_user_ns, inode2, MAY_WRITE))
1298 if (inode1->i_sb != inode2->i_sb)
1304 static int ll_swap_layouts_close(struct obd_client_handle *och,
1305 struct inode *inode, struct inode *inode2)
1307 const struct lu_fid *fid1 = ll_inode2fid(inode);
1308 const struct lu_fid *fid2;
1312 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1313 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1315 rc = ll_check_swap_layouts_validity(inode, inode2);
1317 GOTO(out_free_och, rc);
1319 /* We now know that inode2 is a lustre inode */
1320 fid2 = ll_inode2fid(inode2);
1322 rc = lu_fid_cmp(fid1, fid2);
1324 GOTO(out_free_och, rc = -EINVAL);
1326 /* Close the file and {swap,merge} layouts between inode & inode2.
1327 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1328 * because we still need it to pack l_remote_handle to MDT. */
1329 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1332 och = NULL; /* freed in ll_close_inode_openhandle() */
1342 * Release lease and close the file.
1343 * It will check if the lease has ever broken.
1345 static int ll_lease_close_intent(struct obd_client_handle *och,
1346 struct inode *inode,
1347 bool *lease_broken, enum mds_op_bias bias,
1350 struct ldlm_lock *lock;
1351 bool cancelled = true;
1355 lock = ldlm_handle2lock(&och->och_lease_handle);
1357 lock_res_and_lock(lock);
1358 cancelled = ldlm_is_cancel(lock);
1359 unlock_res_and_lock(lock);
1360 LDLM_LOCK_PUT(lock);
1363 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1364 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1366 if (lease_broken != NULL)
1367 *lease_broken = cancelled;
1369 if (!cancelled && !bias)
1370 ldlm_cli_cancel(&och->och_lease_handle, 0);
1372 if (cancelled) { /* no need to excute intent */
1377 rc = ll_close_inode_openhandle(inode, och, bias, data);
1381 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1384 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1388 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1390 static int ll_lease_file_resync(struct obd_client_handle *och,
1391 struct inode *inode, void __user *uarg)
1393 struct ll_sb_info *sbi = ll_i2sbi(inode);
1394 struct md_op_data *op_data;
1395 struct ll_ioc_lease_id ioc;
1396 __u64 data_version_unused;
1400 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1401 LUSTRE_OPC_ANY, NULL);
1402 if (IS_ERR(op_data))
1403 RETURN(PTR_ERR(op_data));
1405 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
1408 /* before starting file resync, it's necessary to clean up page cache
1409 * in client memory, otherwise once the layout version is increased,
1410 * writing back cached data will be denied the OSTs. */
1411 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1415 op_data->op_lease_handle = och->och_lease_handle;
1416 op_data->op_mirror_id = ioc.lil_mirror_id;
1417 rc = md_file_resync(sbi->ll_md_exp, op_data);
1423 ll_finish_md_op_data(op_data);
1427 static int ll_merge_attr_nolock(const struct lu_env *env, struct inode *inode)
1429 struct ll_inode_info *lli = ll_i2info(inode);
1430 struct cl_object *obj = lli->lli_clob;
1431 struct cl_attr *attr = vvp_env_thread_attr(env);
1439 /* Merge timestamps the most recently obtained from MDS with
1440 * timestamps obtained from OSTs.
1442 * Do not overwrite atime of inode because it may be refreshed
1443 * by file_accessed() function. If the read was served by cache
1444 * data, there is no RPC to be sent so that atime may not be
1445 * transferred to OSTs at all. MDT only updates atime at close time
1446 * if it's at least 'mdd.*.atime_diff' older.
1447 * All in all, the atime in Lustre does not strictly comply with
1448 * POSIX. Solving this problem needs to send an RPC to MDT for each
1449 * read, this will hurt performance.
1451 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1452 inode->i_atime.tv_sec < lli->lli_atime)
1453 inode->i_atime.tv_sec = lli->lli_atime;
1455 inode->i_mtime.tv_sec = lli->lli_mtime;
1456 inode->i_ctime.tv_sec = lli->lli_ctime;
1458 mtime = inode->i_mtime.tv_sec;
1459 atime = inode->i_atime.tv_sec;
1460 ctime = inode->i_ctime.tv_sec;
1462 cl_object_attr_lock(obj);
1463 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1466 rc = cl_object_attr_get(env, obj, attr);
1467 cl_object_attr_unlock(obj);
1470 GOTO(out, rc = (rc == -ENODATA ? 0 : rc));
1472 if (atime < attr->cat_atime)
1473 atime = attr->cat_atime;
1475 if (ctime < attr->cat_ctime)
1476 ctime = attr->cat_ctime;
1478 if (mtime < attr->cat_mtime)
1479 mtime = attr->cat_mtime;
1481 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu i_blocks %llu\n",
1482 PFID(&lli->lli_fid), attr->cat_size, attr->cat_blocks);
1484 if (llcrypt_require_key(inode) == -ENOKEY) {
1485 /* Without the key, round up encrypted file size to next
1486 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1487 * lli_lazysize for proper file size setting at close time.
1489 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1490 lli->lli_lazysize = attr->cat_size;
1491 attr->cat_size = round_up(attr->cat_size,
1492 LUSTRE_ENCRYPTION_UNIT_SIZE);
1494 i_size_write(inode, attr->cat_size);
1495 inode->i_blocks = attr->cat_blocks;
1497 inode->i_mtime.tv_sec = mtime;
1498 inode->i_atime.tv_sec = atime;
1499 inode->i_ctime.tv_sec = ctime;
1506 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1510 ll_inode_size_lock(inode);
1511 rc = ll_merge_attr_nolock(env, inode);
1512 ll_inode_size_unlock(inode);
1517 /* Use to update size and blocks on inode for LSOM if there is no contention */
1518 int ll_merge_attr_try(const struct lu_env *env, struct inode *inode)
1522 if (ll_inode_size_trylock(inode)) {
1523 rc = ll_merge_attr_nolock(env, inode);
1524 ll_inode_size_unlock(inode);
1531 * Set designated mirror for I/O.
1533 * So far only read, write, and truncated can support to issue I/O to
1534 * designated mirror.
1536 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1538 struct ll_file_data *fd = file->private_data;
1540 /* clear layout version for generic(non-resync) I/O in case it carries
1541 * stale layout version due to I/O restart */
1542 io->ci_layout_version = 0;
1544 /* FLR: disable non-delay for designated mirror I/O because obviously
1545 * only one mirror is available */
1546 if (fd->fd_designated_mirror > 0) {
1548 io->ci_designated_mirror = fd->fd_designated_mirror;
1549 io->ci_layout_version = fd->fd_layout_version;
1552 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1553 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1557 * This is relatime_need_update() from Linux 5.17, which is not exported.
1559 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1560 struct timespec64 now)
1563 if (!(mnt->mnt_flags & MNT_RELATIME))
1566 * Is mtime younger than atime? If yes, update atime:
1568 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1571 * Is ctime younger than atime? If yes, update atime:
1573 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1577 * Is the previous atime value older than a day? If yes,
1580 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1583 * Good, we can skip the atime update:
1589 * Very similar to kernel function: !__atime_needs_update()
1591 static bool file_is_noatime(const struct file *file)
1593 struct vfsmount *mnt = file->f_path.mnt;
1594 struct inode *inode = file_inode((struct file *)file);
1595 struct timespec64 now;
1597 if (file->f_flags & O_NOATIME)
1600 if (inode->i_flags & S_NOATIME)
1603 if (IS_NOATIME(inode))
1606 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1609 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1612 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1615 now = current_time(inode);
1617 if (!relatime_need_update(mnt, inode, now))
1623 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1624 struct vvp_io_args *args)
1626 struct inode *inode = file_inode(file);
1627 struct ll_file_data *fd = file->private_data;
1629 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1630 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1632 if (iot == CIT_WRITE) {
1633 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1634 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1635 file->f_flags & O_DIRECT ||
1637 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1638 io->u.ci_wr.wr_sync |= !!(args &&
1639 (args->u.normal.via_iocb->ki_flags &
1645 io->ci_iocb_nowait = !!(args &&
1646 (args->u.normal.via_iocb->ki_flags &
1650 io->ci_obj = ll_i2info(inode)->lli_clob;
1651 io->ci_lockreq = CILR_MAYBE;
1652 if (ll_file_nolock(file)) {
1653 io->ci_lockreq = CILR_NEVER;
1654 io->ci_no_srvlock = 1;
1655 } else if (file->f_flags & O_APPEND) {
1656 io->ci_lockreq = CILR_MANDATORY;
1658 io->ci_noatime = file_is_noatime(file);
1659 io->ci_async_readahead = false;
1661 /* FLR: only use non-delay I/O for read as there is only one
1662 * avaliable mirror for write. */
1663 io->ci_ndelay = !(iot == CIT_WRITE);
1665 ll_io_set_mirror(io, file);
1668 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1671 struct ll_inode_info *lli = ll_i2info(inode);
1672 struct ll_sb_info *sbi = ll_i2sbi(inode);
1673 enum obd_heat_type sample_type;
1674 enum obd_heat_type iobyte_type;
1675 __u64 now = ktime_get_real_seconds();
1677 if (!ll_sbi_has_file_heat(sbi) ||
1678 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1681 if (iot == CIT_READ) {
1682 sample_type = OBD_HEAT_READSAMPLE;
1683 iobyte_type = OBD_HEAT_READBYTE;
1684 } else if (iot == CIT_WRITE) {
1685 sample_type = OBD_HEAT_WRITESAMPLE;
1686 iobyte_type = OBD_HEAT_WRITEBYTE;
1691 spin_lock(&lli->lli_heat_lock);
1692 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1693 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1694 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1695 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1696 spin_unlock(&lli->lli_heat_lock);
1700 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1701 struct file *file, enum cl_io_type iot,
1702 loff_t *ppos, size_t count)
1704 struct vvp_io *vio = vvp_env_io(env);
1705 struct inode *inode = file_inode(file);
1706 struct ll_inode_info *lli = ll_i2info(inode);
1707 struct ll_sb_info *sbi = ll_i2sbi(inode);
1708 struct ll_file_data *fd = file->private_data;
1709 struct range_lock range;
1710 bool range_locked = false;
1716 unsigned int retried = 0, dio_lock = 0;
1717 bool is_aio = false;
1718 bool is_parallel_dio = false;
1719 struct cl_dio_aio *ci_dio_aio = NULL;
1721 bool partial_io = false;
1722 size_t max_io_pages, max_cached_pages;
1726 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1727 file_dentry(file)->d_name.name,
1728 iot == CIT_READ ? "read" : "write", *ppos, count);
1730 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1731 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1732 if (max_io_pages > (max_cached_pages >> 2))
1733 max_io_pages = max_cached_pages >> 2;
1735 io = vvp_env_thread_io(env);
1736 if (file->f_flags & O_DIRECT) {
1737 if (file->f_flags & O_APPEND)
1739 if (!is_sync_kiocb(args->u.normal.via_iocb))
1742 /* the kernel does not support AIO on pipes, and parallel DIO
1743 * uses part of the AIO path, so we must not do parallel dio
1746 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1749 if (!ll_sbi_has_parallel_dio(sbi))
1750 is_parallel_dio = false;
1752 ci_dio_aio = cl_dio_aio_alloc(args->u.normal.via_iocb,
1753 ll_i2info(inode)->lli_clob, is_aio);
1755 GOTO(out, rc = -ENOMEM);
1760 * IO block size need be aware of cached page limit, otherwise
1761 * if we have small max_cached_mb but large block IO issued, io
1762 * could not be finished and blocked whole client.
1764 if (file->f_flags & O_DIRECT)
1767 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1768 partial_io = per_bytes < count;
1769 io = vvp_env_thread_io(env);
1770 ll_io_init(io, file, iot, args);
1771 io->ci_dio_aio = ci_dio_aio;
1772 io->ci_dio_lock = dio_lock;
1773 io->ci_ndelay_tried = retried;
1774 io->ci_parallel_dio = is_parallel_dio;
1776 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1777 if (file->f_flags & O_APPEND)
1778 range_lock_init(&range, 0, LUSTRE_EOF);
1780 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1782 vio->vui_fd = file->private_data;
1783 vio->vui_iter = args->u.normal.via_iter;
1784 vio->vui_iocb = args->u.normal.via_iocb;
1785 /* Direct IO reads must also take range lock,
1786 * or multiple reads will try to work on the same pages
1787 * See LU-6227 for details.
1789 if (((iot == CIT_WRITE) ||
1790 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1791 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1792 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1794 rc = range_lock(&lli->lli_write_tree, &range);
1798 range_locked = true;
1801 ll_cl_add(inode, env, io, LCC_RW);
1802 rc = cl_io_loop(env, io);
1803 ll_cl_remove(inode, env);
1805 /* cl_io_rw_init() handled IO */
1809 if (io->ci_dio_aio && !is_aio) {
1810 struct cl_sync_io *anchor = &io->ci_dio_aio->cda_sync;
1812 /* for dio, EIOCBQUEUED is an implementation detail,
1813 * and we don't return it to userspace
1815 if (rc == -EIOCBQUEUED)
1818 /* N/B: parallel DIO may be disabled during i/o submission;
1819 * if that occurs, I/O shifts to sync, so it's all resolved
1820 * before we get here, and this wait call completes
1823 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1829 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1831 range_unlock(&lli->lli_write_tree, &range);
1832 range_locked = false;
1836 * In order to move forward AIO, ci_nob was increased,
1837 * but that doesn't mean io have been finished, it just
1838 * means io have been submited, we will always return
1839 * EIOCBQUEUED to the caller, So we could only return
1840 * number of bytes in non-AIO case.
1842 if (io->ci_nob > 0) {
1845 result += io->ci_nob;
1846 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1851 count -= io->ci_nob;
1853 /* prepare IO restart */
1855 args->u.normal.via_iter = vio->vui_iter;
1859 * Reexpand iov count because it was zero
1862 iov_iter_reexpand(vio->vui_iter, count);
1863 if (per_bytes == io->ci_nob)
1864 io->ci_need_restart = 1;
1868 cl_io_fini(env, io);
1871 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1872 file->f_path.dentry->d_name.name,
1873 iot, rc, result, io->ci_need_restart);
1875 if ((!rc || rc == -ENODATA || rc == -ENOLCK || rc == -EIOCBQUEUED) &&
1876 count > 0 && io->ci_need_restart && retries-- > 0) {
1878 "%s: restart %s from ppos=%lld count=%zu retries=%u ret=%zd: rc = %d\n",
1879 file_dentry(file)->d_name.name,
1880 iot == CIT_READ ? "read" : "write",
1881 *ppos, count, retries, result, rc);
1882 /* preserve the tried count for FLR */
1883 retried = io->ci_ndelay_tried;
1884 dio_lock = io->ci_dio_lock;
1888 if (io->ci_dio_aio) {
1890 * VFS will call aio_complete() if no -EIOCBQUEUED
1891 * is returned for AIO, so we can not call aio_complete()
1894 * NB: This is safe because the atomic_dec_and_lock in
1895 * cl_sync_io_init has implicit memory barriers, so this will
1896 * be seen by whichever thread completes the DIO/AIO, even if
1899 if (rc != -EIOCBQUEUED)
1900 io->ci_dio_aio->cda_no_aio_complete = 1;
1902 * Drop one extra reference so that end_io() could be
1903 * called for this IO context, we could call it after
1904 * we make sure all AIO requests have been proceed.
1906 cl_sync_io_note(env, &io->ci_dio_aio->cda_sync,
1907 rc == -EIOCBQUEUED ? 0 : rc);
1909 LASSERT(io->ci_dio_aio->cda_creator_free);
1910 cl_dio_aio_free(env, io->ci_dio_aio);
1911 io->ci_dio_aio = NULL;
1915 if (iot == CIT_READ) {
1917 ll_stats_ops_tally(ll_i2sbi(inode),
1918 LPROC_LL_READ_BYTES, result);
1919 } else if (iot == CIT_WRITE) {
1921 ll_stats_ops_tally(ll_i2sbi(inode),
1922 LPROC_LL_WRITE_BYTES, result);
1923 fd->fd_write_failed = false;
1924 } else if (result == 0 && rc == 0) {
1927 fd->fd_write_failed = true;
1929 fd->fd_write_failed = false;
1930 } else if (rc != -ERESTARTSYS) {
1931 fd->fd_write_failed = true;
1935 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1937 ll_heat_add(inode, iot, result);
1939 RETURN(result > 0 ? result : rc);
1943 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1944 * especially for small I/O.
1946 * To serve a read request, CLIO has to create and initialize a cl_io and
1947 * then request DLM lock. This has turned out to have siginificant overhead
1948 * and affects the performance of small I/O dramatically.
1950 * It's not necessary to create a cl_io for each I/O. Under the help of read
1951 * ahead, most of the pages being read are already in memory cache and we can
1952 * read those pages directly because if the pages exist, the corresponding DLM
1953 * lock must exist so that page content must be valid.
1955 * In fast read implementation, the llite speculatively finds and reads pages
1956 * in memory cache. There are three scenarios for fast read:
1957 * - If the page exists and is uptodate, kernel VM will provide the data and
1958 * CLIO won't be intervened;
1959 * - If the page was brought into memory by read ahead, it will be exported
1960 * and read ahead parameters will be updated;
1961 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1962 * it will go back and invoke normal read, i.e., a cl_io will be created
1963 * and DLM lock will be requested.
1965 * POSIX compliance: posix standard states that read is intended to be atomic.
1966 * Lustre read implementation is in line with Linux kernel read implementation
1967 * and neither of them complies with POSIX standard in this matter. Fast read
1968 * doesn't make the situation worse on single node but it may interleave write
1969 * results from multiple nodes due to short read handling in ll_file_aio_read().
1971 * \param env - lu_env
1972 * \param iocb - kiocb from kernel
1973 * \param iter - user space buffers where the data will be copied
1975 * \retval - number of bytes have been read, or error code if error occurred.
1978 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1982 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1985 /* NB: we can't do direct IO for fast read because it will need a lock
1986 * to make IO engine happy. */
1987 if (iocb->ki_filp->f_flags & O_DIRECT)
1990 result = generic_file_read_iter(iocb, iter);
1992 /* If the first page is not in cache, generic_file_aio_read() will be
1993 * returned with -ENODATA. Fall back to full read path.
1994 * See corresponding code in ll_readpage().
1996 * if we raced with page deletion, we might get EIO. Rather than add
1997 * locking to the fast path for this rare case, fall back to the full
1998 * read path. (See vvp_io_read_start() for rest of handling.
2000 if (result == -ENODATA || result == -EIO)
2004 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
2005 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
2006 LPROC_LL_READ_BYTES, result);
2013 * Confine read iter lest read beyond the EOF
2015 * \param iocb [in] kernel iocb
2016 * \param to [in] reader iov_iter
2018 * \retval <0 failure
2020 * \retval >0 @iocb->ki_pos has passed the EOF
2022 static int file_read_confine_iter(struct lu_env *env, struct kiocb *iocb,
2023 struct iov_iter *to)
2025 struct cl_attr *attr = vvp_env_thread_attr(env);
2026 struct file *file = iocb->ki_filp;
2027 struct inode *inode = file_inode(file);
2028 struct ll_inode_info *lli = ll_i2info(inode);
2029 loff_t read_end = iocb->ki_pos + iov_iter_count(to);
2034 cl_object_attr_lock(lli->lli_clob);
2035 rc = cl_object_attr_get(env, lli->lli_clob, attr);
2036 cl_object_attr_unlock(lli->lli_clob);
2040 kms = attr->cat_kms;
2041 /* if read beyond end-of-file, adjust read count */
2042 if (kms > 0 && (iocb->ki_pos >= kms || read_end > kms)) {
2043 rc = ll_glimpse_size(inode);
2047 size = i_size_read(inode);
2048 if (iocb->ki_pos >= size || read_end > size) {
2050 "%s: read [%llu, %llu] over eof, kms %llu, file_size %llu.\n",
2051 file_dentry(file)->d_name.name,
2052 iocb->ki_pos, read_end, kms, size);
2054 if (iocb->ki_pos >= size)
2057 if (read_end > size)
2058 iov_iter_truncate(to, size - iocb->ki_pos);
2066 * Read from a file (through the page cache).
2068 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2071 struct vvp_io_args *args;
2072 struct file *file = iocb->ki_filp;
2076 ktime_t kstart = ktime_get();
2078 bool stale_data = false;
2082 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2083 file_dentry(file)->d_name.name,
2084 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2085 iov_iter_count(to));
2087 if (!iov_iter_count(to))
2090 env = cl_env_get(&refcheck);
2092 RETURN(PTR_ERR(env));
2094 result = file_read_confine_iter(env, iocb, to);
2097 else if (result > 0)
2101 * Currently when PCC read failed, we do not fall back to the
2102 * normal read path, just return the error.
2103 * The resaon is that: for RW-PCC, the file data may be modified
2104 * in the PCC and inconsistent with the data on OSTs (or file
2105 * data has been removed from the Lustre file system), at this
2106 * time, fallback to the normal read path may read the wrong
2108 * TODO: for RO-PCC (readonly PCC), fall back to normal read
2109 * path: read data from data copy on OSTs.
2111 result = pcc_file_read_iter(iocb, to, &cached);
2115 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
2117 result = ll_do_fast_read(iocb, to);
2118 if (result < 0 || iov_iter_count(to) == 0)
2121 args = ll_env_args(env);
2122 args->u.normal.via_iter = to;
2123 args->u.normal.via_iocb = iocb;
2125 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
2126 &iocb->ki_pos, iov_iter_count(to));
2129 else if (result == 0)
2133 cl_env_put(env, &refcheck);
2135 if (stale_data && result > 0) {
2137 * we've reached EOF before the read, the data read are cached
2140 iov_iter_truncate(to, 0);
2145 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2146 file->private_data, iocb->ki_pos, result,
2148 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2149 ktime_us_delta(ktime_get(), kstart));
2153 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2154 file_dentry(file)->d_name.name,
2155 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2156 iov_iter_count(to));
2162 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2163 * If a page is already in the page cache and dirty (and some other things -
2164 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2165 * write to it without doing a full I/O, because Lustre already knows about it
2166 * and will write it out. This saves a lot of processing time.
2168 * All writes here are within one page, so exclusion is handled by the page
2169 * lock on the vm page. We do not do tiny writes for writes which touch
2170 * multiple pages because it's very unlikely multiple sequential pages are
2171 * are already dirty.
2173 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2174 * and are unlikely to be to already dirty pages.
2176 * Attribute updates are important here, we do them in ll_tiny_write_end.
2178 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2180 ssize_t count = iov_iter_count(iter);
2181 struct file *file = iocb->ki_filp;
2182 struct inode *inode = file_inode(file);
2183 bool lock_inode = !IS_NOSEC(inode);
2188 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2189 * of function for why.
2191 if (count >= PAGE_SIZE ||
2192 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2195 if (unlikely(lock_inode))
2197 result = __generic_file_write_iter(iocb, iter);
2199 if (unlikely(lock_inode))
2200 inode_unlock(inode);
2202 /* If the page is not already dirty, ll_tiny_write_begin returns
2203 * -ENODATA. We continue on to normal write.
2205 if (result == -ENODATA)
2209 ll_heat_add(inode, CIT_WRITE, result);
2210 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2213 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2219 * Write to a file (through the page cache).
2221 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2223 struct vvp_io_args *args;
2225 ssize_t rc_tiny = 0, rc_normal;
2226 struct file *file = iocb->ki_filp;
2229 ktime_t kstart = ktime_get();
2234 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2235 file_dentry(file)->d_name.name,
2236 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2237 iov_iter_count(from));
2239 if (!iov_iter_count(from))
2240 GOTO(out, rc_normal = 0);
2243 * When PCC write failed, we usually do not fall back to the normal
2244 * write path, just return the error. But there is a special case when
2245 * returned error code is -ENOSPC due to running out of space on PCC HSM
2246 * bakcend. At this time, it will fall back to normal I/O path and
2247 * retry the I/O. As the file is in HSM released state, it will restore
2248 * the file data to OSTs first and redo the write again. And the
2249 * restore process will revoke the layout lock and detach the file
2250 * from PCC cache automatically.
2252 result = pcc_file_write_iter(iocb, from, &cached);
2253 if (cached && result != -ENOSPC && result != -EDQUOT)
2254 GOTO(out, rc_normal = result);
2256 /* NB: we can't do direct IO for tiny writes because they use the page
2257 * cache, we can't do sync writes because tiny writes can't flush
2258 * pages, and we can't do append writes because we can't guarantee the
2259 * required DLM locks are held to protect file size.
2261 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2262 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2263 rc_tiny = ll_do_tiny_write(iocb, from);
2265 /* In case of error, go on and try normal write - Only stop if tiny
2266 * write completed I/O.
2268 if (iov_iter_count(from) == 0)
2269 GOTO(out, rc_normal = rc_tiny);
2271 env = cl_env_get(&refcheck);
2273 RETURN(PTR_ERR(env));
2275 args = ll_env_args(env);
2276 args->u.normal.via_iter = from;
2277 args->u.normal.via_iocb = iocb;
2279 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2280 &iocb->ki_pos, iov_iter_count(from));
2282 /* On success, combine bytes written. */
2283 if (rc_tiny >= 0 && rc_normal > 0)
2284 rc_normal += rc_tiny;
2285 /* On error, only return error from normal write if tiny write did not
2286 * write any bytes. Otherwise return bytes written by tiny write.
2288 else if (rc_tiny > 0)
2289 rc_normal = rc_tiny;
2291 cl_env_put(env, &refcheck);
2293 if (rc_normal > 0) {
2294 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2295 file->private_data, iocb->ki_pos,
2297 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2298 ktime_us_delta(ktime_get(), kstart));
2302 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2303 file_dentry(file)->d_name.name,
2304 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2305 iov_iter_count(from));
2310 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2312 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2314 static int ll_file_get_iov_count(const struct iovec *iov,
2315 unsigned long *nr_segs, size_t *count,
2321 for (seg = 0; seg < *nr_segs; seg++) {
2322 const struct iovec *iv = &iov[seg];
2325 * If any segment has a negative length, or the cumulative
2326 * length ever wraps negative then return -EINVAL.
2329 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2331 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2336 cnt -= iv->iov_len; /* This segment is no good */
2343 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2344 unsigned long nr_segs, loff_t pos)
2351 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2358 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2359 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2360 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2361 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2362 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2364 result = ll_file_read_iter(iocb, &to);
2369 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2372 struct iovec iov = { .iov_base = buf, .iov_len = count };
2381 init_sync_kiocb(&kiocb, file);
2382 kiocb.ki_pos = *ppos;
2383 #ifdef HAVE_KIOCB_KI_LEFT
2384 kiocb.ki_left = count;
2385 #elif defined(HAVE_KI_NBYTES)
2386 kiocb.i_nbytes = count;
2389 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2390 *ppos = kiocb.ki_pos;
2396 * Write to a file (through the page cache).
2399 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2400 unsigned long nr_segs, loff_t pos)
2402 struct iov_iter from;
2407 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2414 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2415 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2416 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2417 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2418 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2420 result = ll_file_write_iter(iocb, &from);
2425 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2426 size_t count, loff_t *ppos)
2428 struct iovec iov = { .iov_base = (void __user *)buf,
2438 init_sync_kiocb(&kiocb, file);
2439 kiocb.ki_pos = *ppos;
2440 #ifdef HAVE_KIOCB_KI_LEFT
2441 kiocb.ki_left = count;
2442 #elif defined(HAVE_KI_NBYTES)
2443 kiocb.ki_nbytes = count;
2446 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2447 *ppos = kiocb.ki_pos;
2451 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2453 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2454 __u64 flags, struct lov_user_md *lum, int lum_size)
2456 struct lookup_intent oit = {
2458 .it_flags = flags | MDS_OPEN_BY_FID,
2463 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2464 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2465 /* this code will only exist for big-endian systems */
2466 lustre_swab_lov_user_md(lum, 0);
2469 ll_inode_size_lock(inode);
2470 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2472 GOTO(out_unlock, rc);
2474 ll_release_openhandle(dentry, &oit);
2477 ll_inode_size_unlock(inode);
2478 ll_intent_release(&oit);
2483 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2484 struct lov_mds_md **lmmp, int *lmm_size,
2485 struct ptlrpc_request **request)
2487 struct ll_sb_info *sbi = ll_i2sbi(inode);
2488 struct mdt_body *body;
2489 struct lov_mds_md *lmm = NULL;
2490 struct ptlrpc_request *req = NULL;
2491 struct md_op_data *op_data;
2496 rc = ll_get_default_mdsize(sbi, &lmmsize);
2500 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2501 strlen(filename), lmmsize,
2502 LUSTRE_OPC_ANY, NULL);
2503 if (IS_ERR(op_data))
2504 RETURN(PTR_ERR(op_data));
2506 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2507 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2508 ll_finish_md_op_data(op_data);
2510 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2515 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2516 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2518 lmmsize = body->mbo_eadatasize;
2520 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2522 GOTO(out, rc = -ENODATA);
2524 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2525 LASSERT(lmm != NULL);
2527 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2528 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2529 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2530 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2531 GOTO(out, rc = -EPROTO);
2534 * This is coming from the MDS, so is probably in
2535 * little endian. We convert it to host endian before
2536 * passing it to userspace.
2538 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2539 int stripe_count = 0;
2541 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2542 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2543 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2544 if (le32_to_cpu(lmm->lmm_pattern) &
2545 LOV_PATTERN_F_RELEASED)
2547 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2549 /* if function called for directory - we should
2550 * avoid swab not existent lsm objects
2552 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2553 S_ISREG(body->mbo_mode))
2554 lustre_swab_lov_user_md_objects(
2555 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2557 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2558 S_ISREG(body->mbo_mode))
2559 lustre_swab_lov_user_md_objects(
2560 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2562 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2563 lustre_swab_lov_comp_md_v1(
2564 (struct lov_comp_md_v1 *)lmm);
2568 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2569 struct lov_comp_md_v1 *comp_v1 = NULL;
2570 struct lov_comp_md_entry_v1 *ent;
2571 struct lov_user_md_v1 *v1;
2575 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2576 /* Dump the striping information */
2577 for (; i < comp_v1->lcm_entry_count; i++) {
2578 ent = &comp_v1->lcm_entries[i];
2579 off = ent->lcme_offset;
2580 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2582 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2583 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2587 * Return valid stripe_count and stripe_size instead of 0 for
2588 * DoM files to avoid divide-by-zero for older userspace that
2589 * calls this ioctl, e.g. lustre ADIO driver.
2591 if (lmm->lmm_stripe_count == 0)
2592 lmm->lmm_stripe_count = 1;
2593 if (lmm->lmm_stripe_size == 0) {
2594 /* Since the first component of the file data is placed
2595 * on the MDT for faster access, the stripe_size of the
2596 * second one is always that applications which are
2599 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2600 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2602 i = comp_v1->lcm_entry_count > 1 ?
2603 comp_v1->lcm_entry_count - 1 : 0;
2604 ent = &comp_v1->lcm_entries[i];
2605 off = ent->lcme_offset;
2606 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2607 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2612 *lmm_size = lmmsize;
2617 static int ll_lov_setea(struct inode *inode, struct file *file,
2620 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2621 struct lov_user_md *lump;
2622 int lum_size = sizeof(*lump) + sizeof(struct lov_user_ost_data);
2626 if (!capable(CAP_SYS_ADMIN))
2629 OBD_ALLOC_LARGE(lump, lum_size);
2633 if (copy_from_user(lump, arg, lum_size))
2634 GOTO(out_lump, rc = -EFAULT);
2636 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2638 cl_lov_delay_create_clear(&file->f_flags);
2641 OBD_FREE_LARGE(lump, lum_size);
2645 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2652 /* exit before doing any work if pointer is bad */
2653 if (unlikely(!ll_access_ok(lum, sizeof(struct lov_user_md))))
2656 env = cl_env_get(&refcheck);
2658 RETURN(PTR_ERR(env));
2660 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2661 cl_env_put(env, &refcheck);
2665 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2668 struct lov_user_md __user *lum = arg;
2669 struct lov_user_md *klum;
2671 __u64 flags = FMODE_WRITE;
2674 rc = ll_copy_user_md(lum, &klum);
2679 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2684 rc = put_user(0, &lum->lmm_stripe_count);
2688 rc = ll_layout_refresh(inode, &gen);
2692 rc = ll_file_getstripe(inode, arg, lum_size);
2693 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2694 ll_i2info(inode)->lli_clob) {
2695 struct iattr attr = { 0 };
2697 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2698 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2701 cl_lov_delay_create_clear(&file->f_flags);
2704 OBD_FREE_LARGE(klum, lum_size);
2710 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2712 struct ll_inode_info *lli = ll_i2info(inode);
2713 struct cl_object *obj = lli->lli_clob;
2714 struct ll_file_data *fd = file->private_data;
2715 struct ll_grouplock grouplock;
2720 CWARN("group id for group lock must not be 0\n");
2724 if (ll_file_nolock(file))
2725 RETURN(-EOPNOTSUPP);
2727 if (file->f_flags & O_NONBLOCK) {
2728 if (!mutex_trylock(&lli->lli_group_mutex))
2731 mutex_lock(&lli->lli_group_mutex);
2734 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2735 CWARN("group lock already existed with gid %lu\n",
2736 fd->fd_grouplock.lg_gid);
2737 GOTO(out, rc = -EINVAL);
2739 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2740 if (file->f_flags & O_NONBLOCK)
2741 GOTO(out, rc = -EAGAIN);
2742 mutex_unlock(&lli->lli_group_mutex);
2743 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2744 GOTO(retry, rc = 0);
2746 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2749 * XXX: group lock needs to protect all OST objects while PFL
2750 * can add new OST objects during the IO, so we'd instantiate
2751 * all OST objects before getting its group lock.
2756 struct cl_layout cl = {
2757 .cl_is_composite = false,
2759 struct lu_extent ext = {
2761 .e_end = OBD_OBJECT_EOF,
2764 env = cl_env_get(&refcheck);
2766 GOTO(out, rc = PTR_ERR(env));
2768 rc = cl_object_layout_get(env, obj, &cl);
2769 if (rc >= 0 && cl.cl_is_composite)
2770 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2773 cl_env_put(env, &refcheck);
2778 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2779 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2784 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2785 fd->fd_grouplock = grouplock;
2786 if (lli->lli_group_users == 0)
2787 lli->lli_group_gid = grouplock.lg_gid;
2788 lli->lli_group_users++;
2790 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2792 mutex_unlock(&lli->lli_group_mutex);
2797 static int ll_put_grouplock(struct inode *inode, struct file *file,
2800 struct ll_inode_info *lli = ll_i2info(inode);
2801 struct ll_file_data *fd = file->private_data;
2802 struct ll_grouplock grouplock;
2806 mutex_lock(&lli->lli_group_mutex);
2807 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2808 CWARN("no group lock held\n");
2809 GOTO(out, rc = -EINVAL);
2812 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2814 if (fd->fd_grouplock.lg_gid != arg) {
2815 CWARN("group lock %lu doesn't match current id %lu\n",
2816 arg, fd->fd_grouplock.lg_gid);
2817 GOTO(out, rc = -EINVAL);
2820 grouplock = fd->fd_grouplock;
2821 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2822 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2824 cl_put_grouplock(&grouplock);
2826 lli->lli_group_users--;
2827 if (lli->lli_group_users == 0) {
2828 lli->lli_group_gid = 0;
2829 wake_up_var(&lli->lli_group_users);
2831 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2834 mutex_unlock(&lli->lli_group_mutex);
2840 * Close inode open handle
2842 * \param dentry [in] dentry which contains the inode
2843 * \param it [in,out] intent which contains open info and result
2846 * \retval <0 failure
2848 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2850 struct inode *inode = dentry->d_inode;
2851 struct obd_client_handle *och;
2857 /* Root ? Do nothing. */
2858 if (is_root_inode(inode))
2861 /* No open handle to close? Move away */
2862 if (!it_disposition(it, DISP_OPEN_OPEN))
2865 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2867 OBD_ALLOC(och, sizeof(*och));
2869 GOTO(out, rc = -ENOMEM);
2871 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2875 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2877 /* this one is in place of ll_file_open */
2878 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2879 ptlrpc_req_finished(it->it_request);
2880 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2886 * Get size for inode for which FIEMAP mapping is requested.
2887 * Make the FIEMAP get_info call and returns the result.
2888 * \param fiemap kernel buffer to hold extens
2889 * \param num_bytes kernel buffer size
2891 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2897 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2900 /* Checks for fiemap flags */
2901 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2902 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2906 /* Check for FIEMAP_FLAG_SYNC */
2907 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2908 rc = filemap_fdatawrite(inode->i_mapping);
2913 env = cl_env_get(&refcheck);
2915 RETURN(PTR_ERR(env));
2917 if (i_size_read(inode) == 0) {
2918 rc = ll_glimpse_size(inode);
2923 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLPROJID;
2924 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2925 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2927 /* If filesize is 0, then there would be no objects for mapping */
2928 if (fmkey.lfik_oa.o_size == 0) {
2929 fiemap->fm_mapped_extents = 0;
2933 fmkey.lfik_fiemap = *fiemap;
2935 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2936 &fmkey, fiemap, &num_bytes);
2938 cl_env_put(env, &refcheck);
2942 static int fid2path_for_enc_file(struct inode *parent, char *gfpath,
2945 struct dentry *de = NULL, *de_parent = d_find_any_alias(parent);
2946 struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
2947 struct llcrypt_str de_name;
2948 char *p, *ptr = gfpath;
2949 size_t len = 0, len_orig = 0;
2950 int enckey = -1, nameenc = -1;
2954 while ((p = strsep(&gfpath, "/")) != NULL) {
2962 len_orig = strlen(p);
2964 rc = sscanf(p, "["SFID"]", RFID(&fid));
2966 p = strchr(p, ']') + 1;
2972 if (!IS_ENCRYPTED(parent)) {
2973 if (gfpathlen < len + 1) {
2978 memmove(ptr, p, len);
2982 gfpathlen -= len + 1;
2986 /* From here, we know parent is encrypted */
2989 rc = llcrypt_prepare_readdir(parent);
2990 if (rc && rc != -ENOKEY) {
2997 if (llcrypt_has_encryption_key(parent))
3003 llcrypt_policy_has_filename_enc(parent);
3006 /* Even if names are not encrypted, we still need to call
3007 * ll_fname_disk_to_usr in order to decode names as they are
3008 * coming from the wire.
3010 rc = llcrypt_fname_alloc_buffer(parent, NAME_MAX + 1, &lltr);
3018 rc = ll_fname_disk_to_usr(parent, 0, 0, &de_name,
3021 llcrypt_fname_free_buffer(&lltr);
3025 lltr.name[lltr.len] = '\0';
3027 if (lltr.len <= len_orig && gfpathlen >= lltr.len + 1) {
3028 memcpy(ptr, lltr.name, lltr.len);
3033 gfpathlen -= lltr.len + 1;
3037 llcrypt_fname_free_buffer(&lltr);
3039 if (rc == -EOVERFLOW) {
3046 /* We reached the end of the string, which means
3047 * we are dealing with the last component in the path.
3048 * So save a useless lookup and exit.
3054 if (enckey == 0 || nameenc == 0)
3058 de = lookup_one_len(p, de_parent, len);
3059 inode_unlock(parent);
3060 if (IS_ERR_OR_NULL(de) || !de->d_inode) {
3066 parent = de->d_inode;
3073 if (!IS_ERR_OR_NULL(de))
3078 int __ll_fid2path(struct inode *inode, struct getinfo_fid2path *gfout,
3079 size_t outsize, __u32 pathlen_orig)
3081 struct obd_export *exp = ll_i2mdexp(inode);
3084 /* Append root FID after gfout to let MDT know the root FID so that
3085 * it can lookup the correct path, this is mainly for fileset.
3086 * old server without fileset mount support will ignore this.
3088 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
3090 /* Call mdc_iocontrol */
3091 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
3093 if (!rc && gfout->gf_pathlen && gfout->gf_u.gf_path[0] == '/') {
3094 /* by convention, server side (mdt_path_current()) puts
3095 * a leading '/' to tell client that we are dealing with
3098 rc = fid2path_for_enc_file(inode, gfout->gf_u.gf_path,
3100 if (!rc && strlen(gfout->gf_u.gf_path) > pathlen_orig)
3107 int ll_fid2path(struct inode *inode, void __user *arg)
3109 const struct getinfo_fid2path __user *gfin = arg;
3110 __u32 pathlen, pathlen_orig;
3111 struct getinfo_fid2path *gfout;
3117 if (!capable(CAP_DAC_READ_SEARCH) &&
3118 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3121 /* Only need to get the buflen */
3122 if (get_user(pathlen, &gfin->gf_pathlen))
3125 if (pathlen > PATH_MAX)
3127 pathlen_orig = pathlen;
3130 outsize = sizeof(*gfout) + pathlen;
3131 OBD_ALLOC(gfout, outsize);
3135 if (copy_from_user(gfout, arg, sizeof(*gfout)))
3136 GOTO(gf_free, rc = -EFAULT);
3138 gfout->gf_pathlen = pathlen;
3139 rc = __ll_fid2path(inode, gfout, outsize, pathlen_orig);
3143 if (copy_to_user(arg, gfout, sizeof(*gfout) + pathlen_orig))
3147 OBD_FREE(gfout, outsize);
3148 if (rc == -ENAMETOOLONG) {
3149 pathlen += PATH_MAX;
3156 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
3158 struct cl_object *obj = ll_i2info(inode)->lli_clob;
3166 ioc->idv_version = 0;
3167 ioc->idv_layout_version = UINT_MAX;
3169 /* If no file object initialized, we consider its version is 0. */
3173 env = cl_env_get(&refcheck);
3175 RETURN(PTR_ERR(env));
3177 io = vvp_env_thread_io(env);
3179 io->u.ci_data_version.dv_data_version = 0;
3180 io->u.ci_data_version.dv_layout_version = UINT_MAX;
3181 io->u.ci_data_version.dv_flags = ioc->idv_flags;
3184 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
3185 result = cl_io_loop(env, io);
3187 result = io->ci_result;
3189 ioc->idv_version = io->u.ci_data_version.dv_data_version;
3190 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
3192 cl_io_fini(env, io);
3194 if (unlikely(io->ci_need_restart))
3197 cl_env_put(env, &refcheck);
3203 * Read the data_version for inode.
3205 * This value is computed using stripe object version on OST.
3206 * Version is computed using server side locking.
3208 * @param flags if do sync on the OST side;
3210 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
3211 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
3213 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
3215 struct ioc_data_version ioc = { .idv_flags = flags };
3218 rc = ll_ioc_data_version(inode, &ioc);
3220 *data_version = ioc.idv_version;
3226 * Trigger a HSM release request for the provided inode.
3228 int ll_hsm_release(struct inode *inode)
3231 struct obd_client_handle *och = NULL;
3232 __u64 data_version = 0;
3238 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
3239 ll_i2sbi(inode)->ll_fsname,
3240 PFID(&ll_i2info(inode)->lli_fid));
3242 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
3244 GOTO(out, rc = PTR_ERR(och));
3246 /* Grab latest data_version and [am]time values */
3247 rc = ll_data_version(inode, &data_version,
3248 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
3252 env = cl_env_get(&refcheck);
3254 GOTO(out, rc = PTR_ERR(env));
3256 rc = ll_merge_attr(env, inode);
3257 cl_env_put(env, &refcheck);
3259 /* If error happen, we have the wrong size for a file.
3265 /* Release the file.
3266 * NB: lease lock handle is released in mdc_hsm_release_pack() because
3267 * we still need it to pack l_remote_handle to MDT. */
3268 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
3274 if (och != NULL && !IS_ERR(och)) /* close the file */
3275 ll_lease_close(och, inode, NULL);
3280 struct ll_swap_stack {
3283 struct inode *inode1;
3284 struct inode *inode2;
3289 static int ll_swap_layouts(struct file *file1, struct file *file2,
3290 struct lustre_swap_layouts *lsl)
3292 struct mdc_swap_layouts msl;
3293 struct md_op_data *op_data;
3296 struct ll_swap_stack *llss = NULL;
3299 OBD_ALLOC_PTR(llss);
3303 llss->inode1 = file_inode(file1);
3304 llss->inode2 = file_inode(file2);
3306 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
3310 /* we use 2 bool because it is easier to swap than 2 bits */
3311 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
3312 llss->check_dv1 = true;
3314 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
3315 llss->check_dv2 = true;
3317 /* we cannot use lsl->sl_dvX directly because we may swap them */
3318 llss->dv1 = lsl->sl_dv1;
3319 llss->dv2 = lsl->sl_dv2;
3321 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
3322 if (rc == 0) /* same file, done! */
3325 if (rc < 0) { /* sequentialize it */
3326 swap(llss->inode1, llss->inode2);
3328 swap(llss->dv1, llss->dv2);
3329 swap(llss->check_dv1, llss->check_dv2);
3333 if (gid != 0) { /* application asks to flush dirty cache */
3334 rc = ll_get_grouplock(llss->inode1, file1, gid);
3338 rc = ll_get_grouplock(llss->inode2, file2, gid);
3340 ll_put_grouplock(llss->inode1, file1, gid);
3345 /* ultimate check, before swaping the layouts we check if
3346 * dataversion has changed (if requested) */
3347 if (llss->check_dv1) {
3348 rc = ll_data_version(llss->inode1, &dv, 0);
3351 if (dv != llss->dv1)
3352 GOTO(putgl, rc = -EAGAIN);
3355 if (llss->check_dv2) {
3356 rc = ll_data_version(llss->inode2, &dv, 0);
3359 if (dv != llss->dv2)
3360 GOTO(putgl, rc = -EAGAIN);
3363 /* struct md_op_data is used to send the swap args to the mdt
3364 * only flags is missing, so we use struct mdc_swap_layouts
3365 * through the md_op_data->op_data */
3366 /* flags from user space have to be converted before they are send to
3367 * server, no flag is sent today, they are only used on the client */
3370 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3371 0, LUSTRE_OPC_ANY, &msl);
3372 if (IS_ERR(op_data))
3373 GOTO(free, rc = PTR_ERR(op_data));
3375 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3376 sizeof(*op_data), op_data, NULL);
3377 ll_finish_md_op_data(op_data);
3384 ll_put_grouplock(llss->inode2, file2, gid);
3385 ll_put_grouplock(llss->inode1, file1, gid);
3395 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3397 struct obd_export *exp = ll_i2mdexp(inode);
3398 struct md_op_data *op_data;
3402 /* Detect out-of range masks */
3403 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3406 /* Non-root users are forbidden to set or clear flags which are
3407 * NOT defined in HSM_USER_MASK. */
3408 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3409 !capable(CAP_SYS_ADMIN))
3412 if (!exp_connect_archive_id_array(exp)) {
3413 /* Detect out-of range archive id */
3414 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3415 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3419 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3420 LUSTRE_OPC_ANY, hss);
3421 if (IS_ERR(op_data))
3422 RETURN(PTR_ERR(op_data));
3424 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3427 ll_finish_md_op_data(op_data);
3432 static int ll_hsm_import(struct inode *inode, struct file *file,
3433 struct hsm_user_import *hui)
3435 struct hsm_state_set *hss = NULL;
3436 struct iattr *attr = NULL;
3440 if (!S_ISREG(inode->i_mode))
3446 GOTO(out, rc = -ENOMEM);
3448 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3449 hss->hss_archive_id = hui->hui_archive_id;
3450 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3451 rc = ll_hsm_state_set(inode, hss);
3455 OBD_ALLOC_PTR(attr);
3457 GOTO(out, rc = -ENOMEM);
3459 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3460 attr->ia_mode |= S_IFREG;
3461 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3462 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3463 attr->ia_size = hui->hui_size;
3464 attr->ia_mtime.tv_sec = hui->hui_mtime;
3465 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3466 attr->ia_atime.tv_sec = hui->hui_atime;
3467 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3469 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3470 ATTR_UID | ATTR_GID |
3471 ATTR_MTIME | ATTR_MTIME_SET |
3472 ATTR_ATIME | ATTR_ATIME_SET;
3476 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3480 inode_unlock(inode);
3492 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3494 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3495 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3498 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3500 struct inode *inode = file_inode(file);
3502 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3503 ATTR_MTIME | ATTR_MTIME_SET |
3506 .tv_sec = lfu->lfu_atime_sec,
3507 .tv_nsec = lfu->lfu_atime_nsec,
3510 .tv_sec = lfu->lfu_mtime_sec,
3511 .tv_nsec = lfu->lfu_mtime_nsec,
3514 .tv_sec = lfu->lfu_ctime_sec,
3515 .tv_nsec = lfu->lfu_ctime_nsec,
3521 if (!capable(CAP_SYS_ADMIN))
3524 if (!S_ISREG(inode->i_mode))
3528 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3530 inode_unlock(inode);
3535 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3538 case MODE_READ_USER:
3540 case MODE_WRITE_USER:
3547 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3549 /* Used to allow the upper layers of the client to request an LDLM lock
3550 * without doing an actual read or write.
3552 * Used for ladvise lockahead to manually request specific locks.
3554 * \param[in] file file this ladvise lock request is on
3555 * \param[in] ladvise ladvise struct describing this lock request
3557 * \retval 0 success, no detailed result available (sync requests
3558 * and requests sent to the server [not handled locally]
3559 * cannot return detailed results)
3560 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3561 * see definitions for details.
3562 * \retval negative negative errno on error
3564 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3566 struct lu_env *env = NULL;
3567 struct cl_io *io = NULL;
3568 struct cl_lock *lock = NULL;
3569 struct cl_lock_descr *descr = NULL;
3570 struct dentry *dentry = file->f_path.dentry;
3571 struct inode *inode = dentry->d_inode;
3572 enum cl_lock_mode cl_mode;
3573 off_t start = ladvise->lla_start;
3574 off_t end = ladvise->lla_end;
3581 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3582 dentry, dentry->d_inode,
3583 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3586 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3588 GOTO(out, result = cl_mode);
3590 /* Get IO environment */
3591 result = cl_io_get(inode, &env, &io, &refcheck);
3595 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3598 * nothing to do for this io. This currently happens when
3599 * stripe sub-object's are not yet created.
3601 result = io->ci_result;
3602 } else if (result == 0) {
3603 lock = vvp_env_lock(env);
3604 descr = &lock->cll_descr;
3606 descr->cld_obj = io->ci_obj;
3607 /* Convert byte offsets to pages */
3608 descr->cld_start = cl_index(io->ci_obj, start);
3609 descr->cld_end = cl_index(io->ci_obj, end);
3610 descr->cld_mode = cl_mode;
3611 /* CEF_MUST is used because we do not want to convert a
3612 * lockahead request to a lockless lock */
3613 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3615 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3616 descr->cld_enq_flags |= CEF_SPECULATIVE;
3618 result = cl_lock_request(env, io, lock);
3620 /* On success, we need to release the lock */
3622 cl_lock_release(env, lock);
3624 cl_io_fini(env, io);
3625 cl_env_put(env, &refcheck);
3627 /* -ECANCELED indicates a matching lock with a different extent
3628 * was already present, and -EEXIST indicates a matching lock
3629 * on exactly the same extent was already present.
3630 * We convert them to positive values for userspace to make
3631 * recognizing true errors easier.
3632 * Note we can only return these detailed results on async requests,
3633 * as sync requests look the same as i/o requests for locking. */
3634 if (result == -ECANCELED)
3635 result = LLA_RESULT_DIFFERENT;
3636 else if (result == -EEXIST)
3637 result = LLA_RESULT_SAME;
3642 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3644 static int ll_ladvise_sanity(struct inode *inode,
3645 struct llapi_lu_ladvise *ladvise)
3647 struct ll_sb_info *sbi = ll_i2sbi(inode);
3648 enum lu_ladvise_type advice = ladvise->lla_advice;
3649 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3650 * be in the first 32 bits of enum ladvise_flags */
3651 __u32 flags = ladvise->lla_peradvice_flags;
3652 /* 3 lines at 80 characters per line, should be plenty */
3655 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3658 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3659 sbi->ll_fsname, advice,
3660 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3664 /* Per-advice checks */
3666 case LU_LADVISE_LOCKNOEXPAND:
3667 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3669 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3670 "rc = %d\n", sbi->ll_fsname, flags,
3671 ladvise_names[advice], rc);
3675 case LU_LADVISE_LOCKAHEAD:
3676 /* Currently only READ and WRITE modes can be requested */
3677 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3678 ladvise->lla_lockahead_mode == 0) {
3680 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3681 "rc = %d\n", sbi->ll_fsname,
3682 ladvise->lla_lockahead_mode,
3683 ladvise_names[advice], rc);
3687 case LU_LADVISE_WILLREAD:
3688 case LU_LADVISE_DONTNEED:
3690 /* Note fall through above - These checks apply to all advices
3691 * except LOCKNOEXPAND */
3692 if (flags & ~LF_DEFAULT_MASK) {
3694 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3695 "rc = %d\n", sbi->ll_fsname, flags,
3696 ladvise_names[advice], rc);
3699 if (ladvise->lla_start >= ladvise->lla_end) {
3701 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3702 "for %s: rc = %d\n", sbi->ll_fsname,
3703 ladvise->lla_start, ladvise->lla_end,
3704 ladvise_names[advice], rc);
3716 * Give file access advices
3718 * The ladvise interface is similar to Linux fadvise() system call, except it
3719 * forwards the advices directly from Lustre client to server. The server side
3720 * codes will apply appropriate read-ahead and caching techniques for the
3721 * corresponding files.
3723 * A typical workload for ladvise is e.g. a bunch of different clients are
3724 * doing small random reads of a file, so prefetching pages into OSS cache
3725 * with big linear reads before the random IO is a net benefit. Fetching
3726 * all that data into each client cache with fadvise() may not be, due to
3727 * much more data being sent to the client.
3729 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3730 struct llapi_lu_ladvise *ladvise)
3734 struct cl_ladvise_io *lio;
3739 env = cl_env_get(&refcheck);
3741 RETURN(PTR_ERR(env));
3743 io = vvp_env_thread_io(env);
3744 io->ci_obj = ll_i2info(inode)->lli_clob;
3746 /* initialize parameters for ladvise */
3747 lio = &io->u.ci_ladvise;
3748 lio->li_start = ladvise->lla_start;
3749 lio->li_end = ladvise->lla_end;
3750 lio->li_fid = ll_inode2fid(inode);
3751 lio->li_advice = ladvise->lla_advice;
3752 lio->li_flags = flags;
3754 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3755 rc = cl_io_loop(env, io);
3759 cl_io_fini(env, io);
3760 cl_env_put(env, &refcheck);
3764 static int ll_lock_noexpand(struct file *file, int flags)
3766 struct ll_file_data *fd = file->private_data;
3768 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3773 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3776 struct fsxattr fsxattr;
3778 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3781 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3782 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3783 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3784 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3785 if (copy_to_user(uarg, &fsxattr, sizeof(fsxattr)))
3791 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3795 * Project Quota ID state is only allowed to change from within the init
3796 * namespace. Enforce that restriction only if we are trying to change
3797 * the quota ID state. Everything else is allowed in user namespaces.
3799 if (current_user_ns() == &init_user_ns) {
3801 * Caller is allowed to change the project ID. if it is being
3802 * changed, make sure that the new value is valid.
3804 if (ll_i2info(inode)->lli_projid != projid &&
3805 !projid_valid(make_kprojid(&init_user_ns, projid)))
3811 if (ll_i2info(inode)->lli_projid != projid)
3814 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3815 if (!(xflags & FS_XFLAG_PROJINHERIT))
3818 if (xflags & FS_XFLAG_PROJINHERIT)
3825 static int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3827 struct ptlrpc_request *req = NULL;
3828 struct md_op_data *op_data;
3829 struct cl_object *obj;
3830 unsigned int inode_flags;
3833 CDEBUG(D_QUOTA, DFID" xflags=%x projid=%u\n",
3834 PFID(ll_inode2fid(inode)), xflags, projid);
3835 rc = ll_ioctl_check_project(inode, xflags, projid);
3839 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3840 LUSTRE_OPC_ANY, NULL);
3841 if (IS_ERR(op_data))
3842 RETURN(PTR_ERR(op_data));
3844 inode_flags = ll_xflags_to_inode_flags(xflags);
3845 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3846 if (xflags & FS_XFLAG_PROJINHERIT)
3847 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3849 /* pass projid to md_op_data */
3850 op_data->op_projid = projid;
3852 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3853 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3854 ptlrpc_req_finished(req);
3856 GOTO(out_fsxattr, rc);
3857 ll_update_inode_flags(inode, op_data->op_attr_flags);
3859 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3860 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3861 GOTO(out_fsxattr, rc);
3863 obj = ll_i2info(inode)->lli_clob;
3865 struct iattr attr = { 0 };
3867 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3871 ll_finish_md_op_data(op_data);
3875 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3878 struct fsxattr fsxattr;
3882 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3885 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3886 fsxattr.fsx_projid));
3889 int ll_ioctl_project(struct file *file, unsigned int cmd, void __user *uarg)
3891 struct lu_project lu_project;
3892 struct dentry *dentry = file_dentry(file);
3893 struct inode *inode = file_inode(file);
3894 struct dentry *child_dentry = NULL;
3895 int rc = 0, name_len;
3897 if (copy_from_user(&lu_project, uarg, sizeof(lu_project)))
3900 /* apply child dentry if name is valid */
3901 name_len = strnlen(lu_project.project_name, NAME_MAX);
3902 if (name_len > 0 && name_len <= NAME_MAX) {
3904 child_dentry = lookup_one_len(lu_project.project_name,
3906 inode_unlock(inode);
3907 if (IS_ERR(child_dentry)) {
3908 rc = PTR_ERR(child_dentry);
3911 inode = child_dentry->d_inode;
3916 } else if (name_len > NAME_MAX) {
3921 switch (lu_project.project_type) {
3922 case LU_PROJECT_SET:
3923 rc = ll_set_project(inode, lu_project.project_xflags,
3924 lu_project.project_id);
3926 case LU_PROJECT_GET:
3927 lu_project.project_xflags =
3928 ll_inode_flags_to_xflags(inode->i_flags);
3929 if (test_bit(LLIF_PROJECT_INHERIT,
3930 &ll_i2info(inode)->lli_flags))
3931 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3932 lu_project.project_id = ll_i2info(inode)->lli_projid;
3933 if (copy_to_user(uarg, &lu_project, sizeof(lu_project))) {
3943 if (!IS_ERR_OR_NULL(child_dentry))
3948 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3951 struct inode *inode = file_inode(file);
3952 struct ll_file_data *fd = file->private_data;
3953 struct ll_inode_info *lli = ll_i2info(inode);
3954 struct obd_client_handle *och = NULL;
3955 struct split_param sp;
3956 struct pcc_param param;
3957 bool lease_broken = false;
3959 enum mds_op_bias bias = 0;
3961 struct file *layout_file = NULL;
3963 size_t data_size = 0;
3964 bool attached = false;
3969 mutex_lock(&lli->lli_och_mutex);
3970 if (fd->fd_lease_och != NULL) {
3971 och = fd->fd_lease_och;
3972 fd->fd_lease_och = NULL;
3974 mutex_unlock(&lli->lli_och_mutex);
3979 fmode = och->och_flags;
3981 switch (ioc->lil_flags) {
3982 case LL_LEASE_RESYNC_DONE:
3983 if (ioc->lil_count > IOC_IDS_MAX)
3984 GOTO(out_lease_close, rc = -EINVAL);
3986 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3987 OBD_ALLOC(data, data_size);
3989 GOTO(out_lease_close, rc = -ENOMEM);
3991 if (copy_from_user(data, uarg, data_size))
3992 GOTO(out_lease_close, rc = -EFAULT);
3994 bias = MDS_CLOSE_RESYNC_DONE;
3996 case LL_LEASE_LAYOUT_MERGE:
3997 if (ioc->lil_count != 1)
3998 GOTO(out_lease_close, rc = -EINVAL);
4000 uarg += sizeof(*ioc);
4001 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4002 GOTO(out_lease_close, rc = -EFAULT);
4004 layout_file = fget(fdv);
4006 GOTO(out_lease_close, rc = -EBADF);
4008 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
4009 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
4010 GOTO(out_lease_close, rc = -EPERM);
4012 data = file_inode(layout_file);
4013 bias = MDS_CLOSE_LAYOUT_MERGE;
4015 case LL_LEASE_LAYOUT_SPLIT: {
4018 if (ioc->lil_count != 2)
4019 GOTO(out_lease_close, rc = -EINVAL);
4021 uarg += sizeof(*ioc);
4022 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4023 GOTO(out_lease_close, rc = -EFAULT);
4025 uarg += sizeof(fdv);
4026 if (copy_from_user(&mirror_id, uarg, sizeof(mirror_id)))
4027 GOTO(out_lease_close, rc = -EFAULT);
4028 if (mirror_id >= MIRROR_ID_NEG)
4029 GOTO(out_lease_close, rc = -EINVAL);
4031 layout_file = fget(fdv);
4033 GOTO(out_lease_close, rc = -EBADF);
4035 /* if layout_file == file, it means to destroy the mirror */
4036 sp.sp_inode = file_inode(layout_file);
4037 sp.sp_mirror_id = (__u16)mirror_id;
4039 bias = MDS_CLOSE_LAYOUT_SPLIT;
4042 case LL_LEASE_PCC_ATTACH:
4043 if (ioc->lil_count != 1)
4046 if (IS_ENCRYPTED(inode))
4047 RETURN(-EOPNOTSUPP);
4049 uarg += sizeof(*ioc);
4050 if (copy_from_user(¶m.pa_archive_id, uarg, sizeof(__u32)))
4051 GOTO(out_lease_close, rc2 = -EFAULT);
4053 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
4055 GOTO(out_lease_close, rc2);
4058 /* Grab latest data version */
4059 rc2 = ll_data_version(inode, ¶m.pa_data_version,
4062 GOTO(out_lease_close, rc2);
4065 bias = MDS_PCC_ATTACH;
4068 /* without close intent */
4073 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
4077 rc = ll_lease_och_release(inode, file);
4086 if (ioc->lil_flags == LL_LEASE_RESYNC_DONE && data)
4087 OBD_FREE(data, data_size);
4092 if (ioc->lil_flags == LL_LEASE_PCC_ATTACH) {
4095 rc = pcc_readwrite_attach_fini(file, inode,
4096 param.pa_layout_gen,
4101 ll_layout_refresh(inode, &fd->fd_layout_version);
4104 rc = ll_lease_type_from_fmode(fmode);
4108 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
4111 struct inode *inode = file_inode(file);
4112 struct ll_inode_info *lli = ll_i2info(inode);
4113 struct ll_file_data *fd = file->private_data;
4114 struct obd_client_handle *och = NULL;
4115 __u64 open_flags = 0;
4121 switch (ioc->lil_mode) {
4122 case LL_LEASE_WRLCK:
4123 if (!(file->f_mode & FMODE_WRITE))
4125 fmode = FMODE_WRITE;
4127 case LL_LEASE_RDLCK:
4128 if (!(file->f_mode & FMODE_READ))
4132 case LL_LEASE_UNLCK:
4133 RETURN(ll_file_unlock_lease(file, ioc, uarg));
4138 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
4140 /* apply for lease */
4141 if (ioc->lil_flags & LL_LEASE_RESYNC)
4142 open_flags = MDS_OPEN_RESYNC;
4143 och = ll_lease_open(inode, file, fmode, open_flags);
4145 RETURN(PTR_ERR(och));
4147 if (ioc->lil_flags & LL_LEASE_RESYNC) {
4148 rc = ll_lease_file_resync(och, inode, uarg);
4150 ll_lease_close(och, inode, NULL);
4153 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
4155 ll_lease_close(och, inode, NULL);
4161 mutex_lock(&lli->lli_och_mutex);
4162 if (fd->fd_lease_och == NULL) {
4163 fd->fd_lease_och = och;
4166 mutex_unlock(&lli->lli_och_mutex);
4168 /* impossible now that only excl is supported for now */
4169 ll_lease_close(och, inode, &lease_broken);
4175 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
4177 struct ll_inode_info *lli = ll_i2info(inode);
4178 struct ll_sb_info *sbi = ll_i2sbi(inode);
4179 __u64 now = ktime_get_real_seconds();
4182 spin_lock(&lli->lli_heat_lock);
4183 heat->lh_flags = lli->lli_heat_flags;
4184 for (i = 0; i < heat->lh_count; i++)
4185 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
4186 now, sbi->ll_heat_decay_weight,
4187 sbi->ll_heat_period_second);
4188 spin_unlock(&lli->lli_heat_lock);
4191 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
4193 struct ll_inode_info *lli = ll_i2info(inode);
4196 spin_lock(&lli->lli_heat_lock);
4197 if (flags & LU_HEAT_FLAG_CLEAR)
4198 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
4200 if (flags & LU_HEAT_FLAG_OFF)
4201 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
4203 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
4205 spin_unlock(&lli->lli_heat_lock);
4211 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4213 struct inode *inode = file_inode(file);
4214 struct ll_file_data *fd = file->private_data;
4215 void __user *uarg = (void __user *)arg;
4219 CDEBUG(D_VFSTRACE|D_IOCTL, "VFS Op:inode="DFID"(%pK) cmd=%x arg=%lx\n",
4220 PFID(ll_inode2fid(inode)), inode, cmd, arg);
4221 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
4223 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
4224 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
4227 /* can't do a generic karg == NULL check here, since it is too noisy and
4228 * we need to return -ENOTTY for unsupported ioctls instead of -EINVAL.
4231 case LL_IOC_GETFLAGS:
4232 /* Get the current value of the file flags */
4233 return put_user(fd->fd_flags, (int __user *)arg);
4234 case LL_IOC_SETFLAGS:
4235 case LL_IOC_CLRFLAGS:
4236 /* Set or clear specific file flags */
4237 /* XXX This probably needs checks to ensure the flags are
4238 * not abused, and to handle any flag side effects.
4240 if (get_user(flags, (int __user *)arg))
4243 if (cmd == LL_IOC_SETFLAGS) {
4244 if ((flags & LL_FILE_IGNORE_LOCK) &&
4245 !(file->f_flags & O_DIRECT)) {
4247 CERROR("%s: unable to disable locking on non-O_DIRECT file "DFID": rc = %d\n",
4248 current->comm, PFID(ll_inode2fid(inode)),
4253 fd->fd_flags |= flags;
4255 fd->fd_flags &= ~flags;
4258 case LL_IOC_LOV_SETSTRIPE:
4259 case LL_IOC_LOV_SETSTRIPE_NEW:
4260 RETURN(ll_lov_setstripe(inode, file, uarg));
4261 case LL_IOC_LOV_SETEA:
4262 RETURN(ll_lov_setea(inode, file, uarg));
4263 case LL_IOC_LOV_SWAP_LAYOUTS: {
4265 struct lustre_swap_layouts lsl;
4267 if (copy_from_user(&lsl, uarg, sizeof(lsl)))
4270 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
4273 file2 = fget(lsl.sl_fd);
4277 /* O_WRONLY or O_RDWR */
4278 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
4279 GOTO(out, rc = -EPERM);
4281 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
4282 struct obd_client_handle *och = NULL;
4283 struct ll_inode_info *lli;
4284 struct inode *inode2;
4286 lli = ll_i2info(inode);
4287 mutex_lock(&lli->lli_och_mutex);
4288 if (fd->fd_lease_och != NULL) {
4289 och = fd->fd_lease_och;
4290 fd->fd_lease_och = NULL;
4292 mutex_unlock(&lli->lli_och_mutex);
4294 GOTO(out, rc = -ENOLCK);
4295 inode2 = file_inode(file2);
4296 rc = ll_swap_layouts_close(och, inode, inode2);
4298 rc = ll_swap_layouts(file, file2, &lsl);
4304 case LL_IOC_LOV_GETSTRIPE:
4305 case LL_IOC_LOV_GETSTRIPE_NEW:
4306 RETURN(ll_file_getstripe(inode, uarg, 0));
4307 case LL_IOC_GROUP_LOCK:
4308 RETURN(ll_get_grouplock(inode, file, arg));
4309 case LL_IOC_GROUP_UNLOCK:
4310 RETURN(ll_put_grouplock(inode, file, arg));
4311 case LL_IOC_DATA_VERSION: {
4312 struct ioc_data_version idv;
4315 if (copy_from_user(&idv, uarg, sizeof(idv)))
4318 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4319 rc = ll_ioc_data_version(inode, &idv);
4321 if (rc == 0 && copy_to_user(uarg, &idv, sizeof(idv)))
4326 case LL_IOC_HSM_STATE_GET: {
4327 struct md_op_data *op_data;
4328 struct hsm_user_state *hus;
4331 if (!ll_access_ok(uarg, sizeof(*hus)))
4338 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4339 LUSTRE_OPC_ANY, hus);
4340 if (IS_ERR(op_data)) {
4341 rc = PTR_ERR(op_data);
4343 rc = obd_iocontrol(cmd, ll_i2mdexp(inode),
4344 sizeof(*op_data), op_data, NULL);
4346 if (copy_to_user(uarg, hus, sizeof(*hus)))
4349 ll_finish_md_op_data(op_data);
4354 case LL_IOC_HSM_STATE_SET: {
4355 struct hsm_state_set *hss;
4362 if (copy_from_user(hss, uarg, sizeof(*hss)))
4365 rc = ll_hsm_state_set(inode, hss);
4370 case LL_IOC_HSM_ACTION: {
4371 struct md_op_data *op_data;
4372 struct hsm_current_action *hca;
4376 if (!ll_access_ok(uarg, sizeof(*hca)))
4383 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4384 LUSTRE_OPC_ANY, hca);
4385 if (IS_ERR(op_data)) {
4387 RETURN(PTR_ERR(op_data));
4390 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4393 GOTO(skip_copy, rc);
4395 /* The hsm_current_action retreived from the server could
4396 * contain corrupt information. If it is incorrect data collect
4397 * debug information. We still send the data even if incorrect
4398 * to user land to handle.
4400 action = hsm_user_action2name(hca->hca_action);
4401 if (strcmp(action, "UNKNOWN") == 0 ||
4402 hca->hca_state > HPS_DONE) {
4404 "HSM current state %s action %s, offset = %llu, length %llu\n",
4405 hsm_progress_state2name(hca->hca_state), action,
4406 hca->hca_location.offset, hca->hca_location.length);
4409 if (copy_to_user(uarg, hca, sizeof(*hca)))
4412 ll_finish_md_op_data(op_data);
4416 case LL_IOC_SET_LEASE_OLD: {
4417 struct ll_ioc_lease ioc = { .lil_mode = arg };
4419 RETURN(ll_file_set_lease(file, &ioc, 0));
4421 case LL_IOC_SET_LEASE: {
4422 struct ll_ioc_lease ioc;
4424 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
4427 RETURN(ll_file_set_lease(file, &ioc, uarg));
4429 case LL_IOC_GET_LEASE: {
4430 struct ll_inode_info *lli = ll_i2info(inode);
4431 struct ldlm_lock *lock = NULL;
4434 mutex_lock(&lli->lli_och_mutex);
4435 if (fd->fd_lease_och != NULL) {
4436 struct obd_client_handle *och = fd->fd_lease_och;
4438 lock = ldlm_handle2lock(&och->och_lease_handle);
4440 lock_res_and_lock(lock);
4441 if (!ldlm_is_cancel(lock))
4442 fmode = och->och_flags;
4444 unlock_res_and_lock(lock);
4445 LDLM_LOCK_PUT(lock);
4448 mutex_unlock(&lli->lli_och_mutex);
4450 RETURN(ll_lease_type_from_fmode(fmode));
4452 case LL_IOC_HSM_IMPORT: {
4453 struct hsm_user_import *hui;
4459 if (copy_from_user(hui, uarg, sizeof(*hui)))
4462 rc = ll_hsm_import(inode, file, hui);
4467 case LL_IOC_FUTIMES_3: {
4468 struct ll_futimes_3 lfu;
4470 if (copy_from_user(&lfu, uarg, sizeof(lfu)))
4473 RETURN(ll_file_futimes_3(file, &lfu));
4475 case LL_IOC_LADVISE: {
4476 struct llapi_ladvise_hdr *k_ladvise_hdr;
4477 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4480 int alloc_size = sizeof(*k_ladvise_hdr);
4483 u_ladvise_hdr = uarg;
4484 OBD_ALLOC_PTR(k_ladvise_hdr);
4485 if (k_ladvise_hdr == NULL)
4488 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4489 GOTO(out_ladvise, rc = -EFAULT);
4491 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4492 k_ladvise_hdr->lah_count < 1)
4493 GOTO(out_ladvise, rc = -EINVAL);
4495 num_advise = k_ladvise_hdr->lah_count;
4496 if (num_advise >= LAH_COUNT_MAX)
4497 GOTO(out_ladvise, rc = -EFBIG);
4499 OBD_FREE_PTR(k_ladvise_hdr);
4500 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4501 lah_advise[num_advise]);
4502 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4503 if (k_ladvise_hdr == NULL)
4507 * TODO: submit multiple advices to one server in a single RPC
4509 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4510 GOTO(out_ladvise, rc = -EFAULT);
4512 for (i = 0; i < num_advise; i++) {
4513 struct llapi_lu_ladvise *k_ladvise =
4514 &k_ladvise_hdr->lah_advise[i];
4515 struct llapi_lu_ladvise __user *u_ladvise =
4516 &u_ladvise_hdr->lah_advise[i];
4518 rc = ll_ladvise_sanity(inode, k_ladvise);
4520 GOTO(out_ladvise, rc);
4522 switch (k_ladvise->lla_advice) {
4523 case LU_LADVISE_LOCKNOEXPAND:
4524 rc = ll_lock_noexpand(file,
4525 k_ladvise->lla_peradvice_flags);
4526 GOTO(out_ladvise, rc);
4527 case LU_LADVISE_LOCKAHEAD:
4529 rc = ll_file_lock_ahead(file, k_ladvise);
4532 GOTO(out_ladvise, rc);
4535 &u_ladvise->lla_lockahead_result))
4536 GOTO(out_ladvise, rc = -EFAULT);
4539 rc = ll_ladvise(inode, file,
4540 k_ladvise_hdr->lah_flags,
4543 GOTO(out_ladvise, rc);
4550 OBD_FREE(k_ladvise_hdr, alloc_size);
4553 case LL_IOC_FLR_SET_MIRROR: {
4554 /* mirror I/O must be direct to avoid polluting page cache
4556 if (!(file->f_flags & O_DIRECT))
4559 fd->fd_designated_mirror = arg;
4562 case LL_IOC_HEAT_GET: {
4563 struct lu_heat uheat;
4564 struct lu_heat *heat;
4567 if (copy_from_user(&uheat, uarg, sizeof(uheat)))
4570 if (uheat.lh_count > OBD_HEAT_COUNT)
4571 uheat.lh_count = OBD_HEAT_COUNT;
4573 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4574 OBD_ALLOC(heat, size);
4578 heat->lh_count = uheat.lh_count;
4579 ll_heat_get(inode, heat);
4580 rc = copy_to_user(uarg, heat, size);
4581 OBD_FREE(heat, size);
4582 RETURN(rc ? -EFAULT : 0);
4584 case LL_IOC_HEAT_SET: {
4587 if (copy_from_user(&flags, uarg, sizeof(flags)))
4590 rc = ll_heat_set(inode, flags);
4593 case LL_IOC_PCC_DETACH: {
4594 struct lu_pcc_detach *detach;
4596 OBD_ALLOC_PTR(detach);
4600 if (copy_from_user(detach, uarg, sizeof(*detach)))
4601 GOTO(out_detach_free, rc = -EFAULT);
4603 if (!S_ISREG(inode->i_mode))
4604 GOTO(out_detach_free, rc = -EINVAL);
4606 if (!inode_owner_or_capable(&init_user_ns, inode))
4607 GOTO(out_detach_free, rc = -EPERM);
4609 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4611 OBD_FREE_PTR(detach);
4614 case LL_IOC_PCC_STATE: {
4615 struct lu_pcc_state __user *ustate = uarg;
4616 struct lu_pcc_state *state;
4618 OBD_ALLOC_PTR(state);
4622 if (copy_from_user(state, ustate, sizeof(*state)))
4623 GOTO(out_state, rc = -EFAULT);
4625 rc = pcc_ioctl_state(file, inode, state);
4627 GOTO(out_state, rc);
4629 if (copy_to_user(ustate, state, sizeof(*state)))
4630 GOTO(out_state, rc = -EFAULT);
4633 OBD_FREE_PTR(state);
4637 rc = ll_iocontrol(inode, file, cmd, uarg);
4640 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, uarg));
4644 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4646 struct inode *inode = file_inode(file);
4649 struct cl_lseek_io *lsio;
4656 env = cl_env_get(&refcheck);
4658 RETURN(PTR_ERR(env));
4660 io = vvp_env_thread_io(env);
4661 io->ci_obj = ll_i2info(inode)->lli_clob;
4662 ll_io_set_mirror(io, file);
4664 lsio = &io->u.ci_lseek;
4665 lsio->ls_start = offset;
4666 lsio->ls_whence = whence;
4667 lsio->ls_result = -ENXIO;
4670 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4672 struct vvp_io *vio = vvp_env_io(env);
4674 vio->vui_fd = file->private_data;
4675 rc = cl_io_loop(env, io);
4679 retval = rc ? : lsio->ls_result;
4680 cl_io_fini(env, io);
4681 } while (unlikely(io->ci_need_restart));
4683 cl_env_put(env, &refcheck);
4685 /* Without the key, SEEK_HOLE return value has to be
4686 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4688 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4689 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4694 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4696 struct inode *inode = file_inode(file);
4697 loff_t retval = offset, eof = 0;
4698 ktime_t kstart = ktime_get();
4702 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4703 PFID(ll_inode2fid(inode)), inode, retval, retval,
4706 if (origin == SEEK_END) {
4707 retval = ll_glimpse_size(inode);
4710 eof = i_size_read(inode);
4713 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4717 /* flush local cache first if any */
4718 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4721 retval = ll_lseek(file, offset, origin);
4724 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4726 retval = generic_file_llseek_size(file, offset, origin,
4727 ll_file_maxbytes(inode), eof);
4730 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4731 ktime_us_delta(ktime_get(), kstart));
4735 static int ll_flush(struct file *file, fl_owner_t id)
4737 struct inode *inode = file_inode(file);
4738 struct ll_inode_info *lli = ll_i2info(inode);
4739 struct ll_file_data *fd = file->private_data;
4742 LASSERT(!S_ISDIR(inode->i_mode));
4744 /* catch async errors that were recorded back when async writeback
4745 * failed for pages in this mapping. */
4746 rc = lli->lli_async_rc;
4747 lli->lli_async_rc = 0;
4748 if (lli->lli_clob != NULL) {
4749 err = lov_read_and_clear_async_rc(lli->lli_clob);
4754 /* The application has been told write failure already.
4755 * Do not report failure again. */
4756 if (fd->fd_write_failed)
4758 return rc ? -EIO : 0;
4762 * Called to make sure a portion of file has been written out.
4763 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4765 * Return how many pages have been written.
4767 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4768 enum cl_fsync_mode mode, int ignore_layout)
4772 struct cl_fsync_io *fio;
4777 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4778 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4781 env = cl_env_get(&refcheck);
4783 RETURN(PTR_ERR(env));
4785 io = vvp_env_thread_io(env);
4786 io->ci_obj = ll_i2info(inode)->lli_clob;
4787 io->ci_ignore_layout = ignore_layout;
4789 /* initialize parameters for sync */
4790 fio = &io->u.ci_fsync;
4791 fio->fi_start = start;
4793 fio->fi_fid = ll_inode2fid(inode);
4794 fio->fi_mode = mode;
4795 fio->fi_nr_written = 0;
4797 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4798 result = cl_io_loop(env, io);
4800 result = io->ci_result;
4802 result = fio->fi_nr_written;
4803 cl_io_fini(env, io);
4804 cl_env_put(env, &refcheck);
4810 * When dentry is provided (the 'else' case), file_dentry() may be
4811 * null and dentry must be used directly rather than pulled from
4812 * file_dentry() as is done otherwise.
4815 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4817 struct dentry *dentry = file_dentry(file);
4818 struct inode *inode = dentry->d_inode;
4819 struct ll_inode_info *lli = ll_i2info(inode);
4820 struct ptlrpc_request *req;
4821 ktime_t kstart = ktime_get();
4827 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4828 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4830 /* fsync's caller has already called _fdata{sync,write}, we want
4831 * that IO to finish before calling the osc and mdc sync methods */
4832 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4834 /* catch async errors that were recorded back when async writeback
4835 * failed for pages in this mapping. */
4836 if (!S_ISDIR(inode->i_mode)) {
4837 err = lli->lli_async_rc;
4838 lli->lli_async_rc = 0;
4841 if (lli->lli_clob != NULL) {
4842 err = lov_read_and_clear_async_rc(lli->lli_clob);
4848 if (S_ISREG(inode->i_mode) && !lli->lli_synced_to_mds) {
4850 * only the first sync on MDS makes sense,
4851 * everything else is stored on OSTs
4853 err = md_fsync(ll_i2sbi(inode)->ll_md_exp,
4854 ll_inode2fid(inode), &req);
4858 lli->lli_synced_to_mds = true;
4859 ptlrpc_req_finished(req);
4863 if (S_ISREG(inode->i_mode)) {
4864 struct ll_file_data *fd = file->private_data;
4867 /* Sync metadata on MDT first, and then sync the cached data
4870 err = pcc_fsync(file, start, end, datasync, &cached);
4872 err = cl_sync_file_range(inode, start, end,
4874 if (rc == 0 && err < 0)
4877 fd->fd_write_failed = true;
4879 fd->fd_write_failed = false;
4883 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4884 ktime_us_delta(ktime_get(), kstart));
4889 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4891 struct inode *inode = file_inode(file);
4892 struct ll_sb_info *sbi = ll_i2sbi(inode);
4893 struct ldlm_enqueue_info einfo = {
4894 .ei_type = LDLM_FLOCK,
4895 .ei_cb_cp = ldlm_flock_completion_ast,
4896 .ei_cbdata = file_lock,
4898 struct md_op_data *op_data;
4899 struct lustre_handle lockh = { 0 };
4900 union ldlm_policy_data flock = { { 0 } };
4901 int fl_type = file_lock->fl_type;
4902 ktime_t kstart = ktime_get();
4908 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4909 PFID(ll_inode2fid(inode)), file_lock);
4911 if (file_lock->fl_flags & FL_FLOCK) {
4912 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4913 /* flocks are whole-file locks */
4914 flock.l_flock.end = OFFSET_MAX;
4915 /* For flocks owner is determined by the local file desctiptor*/
4916 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4917 } else if (file_lock->fl_flags & FL_POSIX) {
4918 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4919 flock.l_flock.start = file_lock->fl_start;
4920 flock.l_flock.end = file_lock->fl_end;
4924 flock.l_flock.pid = file_lock->fl_pid;
4926 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4927 /* Somewhat ugly workaround for svc lockd.
4928 * lockd installs custom fl_lmops->lm_compare_owner that checks
4929 * for the fl_owner to be the same (which it always is on local node
4930 * I guess between lockd processes) and then compares pid.
4931 * As such we assign pid to the owner field to make it all work,
4932 * conflict with normal locks is unlikely since pid space and
4933 * pointer space for current->files are not intersecting */
4934 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4935 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4940 einfo.ei_mode = LCK_PR;
4943 /* An unlock request may or may not have any relation to
4944 * existing locks so we may not be able to pass a lock handle
4945 * via a normal ldlm_lock_cancel() request. The request may even
4946 * unlock a byte range in the middle of an existing lock. In
4947 * order to process an unlock request we need all of the same
4948 * information that is given with a normal read or write record
4949 * lock request. To avoid creating another ldlm unlock (cancel)
4950 * message we'll treat a LCK_NL flock request as an unlock. */
4951 einfo.ei_mode = LCK_NL;
4954 einfo.ei_mode = LCK_PW;
4958 CERROR("%s: fcntl from '%s' unknown lock type=%d: rc = %d\n",
4959 sbi->ll_fsname, current->comm, fl_type, rc);
4974 flags = LDLM_FL_BLOCK_NOWAIT;
4980 flags = LDLM_FL_TEST_LOCK;
4984 CERROR("%s: fcntl from '%s' unknown lock command=%d: rc = %d\n",
4985 sbi->ll_fsname, current->comm, cmd, rc);
4989 /* Save the old mode so that if the mode in the lock changes we
4990 * can decrement the appropriate reader or writer refcount. */
4991 file_lock->fl_type = einfo.ei_mode;
4993 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4994 LUSTRE_OPC_ANY, NULL);
4995 if (IS_ERR(op_data))
4996 RETURN(PTR_ERR(op_data));
4998 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4999 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
5000 flock.l_flock.pid, flags, einfo.ei_mode,
5001 flock.l_flock.start, flock.l_flock.end);
5003 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
5006 /* Restore the file lock type if not TEST lock. */
5007 if (!(flags & LDLM_FL_TEST_LOCK))
5008 file_lock->fl_type = fl_type;
5010 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
5011 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
5012 !(flags & LDLM_FL_TEST_LOCK))
5013 rc2 = locks_lock_file_wait(file, file_lock);
5015 if ((file_lock->fl_flags & FL_FLOCK) &&
5016 (rc == 0 || file_lock->fl_type == F_UNLCK))
5017 rc2 = flock_lock_file_wait(file, file_lock);
5018 if ((file_lock->fl_flags & FL_POSIX) &&
5019 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
5020 !(flags & LDLM_FL_TEST_LOCK))
5021 rc2 = posix_lock_file_wait(file, file_lock);
5022 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
5024 if (rc2 && file_lock->fl_type != F_UNLCK) {
5025 einfo.ei_mode = LCK_NL;
5026 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
5031 ll_finish_md_op_data(op_data);
5034 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
5035 ktime_us_delta(ktime_get(), kstart));
5039 int ll_get_fid_by_name(struct inode *parent, const char *name,
5040 int namelen, struct lu_fid *fid,
5041 struct inode **inode)
5043 struct md_op_data *op_data = NULL;
5044 struct mdt_body *body;
5045 struct ptlrpc_request *req;
5049 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
5050 LUSTRE_OPC_ANY, NULL);
5051 if (IS_ERR(op_data))
5052 RETURN(PTR_ERR(op_data));
5054 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
5055 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
5056 ll_finish_md_op_data(op_data);
5060 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
5062 GOTO(out_req, rc = -EFAULT);
5064 *fid = body->mbo_fid1;
5067 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
5069 ptlrpc_req_finished(req);
5073 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
5074 const char *name, __u32 flags)
5076 struct dentry *dchild = NULL;
5077 struct inode *child_inode = NULL;
5078 struct md_op_data *op_data;
5079 struct ptlrpc_request *request = NULL;
5080 struct obd_client_handle *och = NULL;
5082 struct mdt_body *body;
5083 __u64 data_version = 0;
5084 size_t namelen = strlen(name);
5085 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
5089 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
5090 PFID(ll_inode2fid(parent)), name,
5091 lum->lum_stripe_offset, lum->lum_stripe_count);
5093 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
5094 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
5095 lustre_swab_lmv_user_md(lum);
5097 /* Get child FID first */
5098 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
5101 dchild = d_lookup(file_dentry(file), &qstr);
5103 if (dchild->d_inode)
5104 child_inode = igrab(dchild->d_inode);
5109 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
5118 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
5119 OBD_CONNECT2_DIR_MIGRATE)) {
5120 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
5121 ll_dir_striped(child_inode)) {
5122 CERROR("%s: MDT doesn't support stripe directory "
5123 "migration!\n", ll_i2sbi(parent)->ll_fsname);
5124 GOTO(out_iput, rc = -EOPNOTSUPP);
5129 * lfs migrate command needs to be blocked on the client
5130 * by checking the migrate FID against the FID of the
5133 if (is_root_inode(child_inode))
5134 GOTO(out_iput, rc = -EINVAL);
5136 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
5137 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
5138 if (IS_ERR(op_data))
5139 GOTO(out_iput, rc = PTR_ERR(op_data));
5141 inode_lock(child_inode);
5142 op_data->op_fid3 = *ll_inode2fid(child_inode);
5143 if (!fid_is_sane(&op_data->op_fid3)) {
5144 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
5145 ll_i2sbi(parent)->ll_fsname, name,
5146 PFID(&op_data->op_fid3));
5147 GOTO(out_unlock, rc = -EINVAL);
5150 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
5151 op_data->op_data = lum;
5152 op_data->op_data_size = lumlen;
5154 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
5155 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
5156 lmv_dir_layout_changing(ll_i2info(parent)->lli_lsm_md))
5157 op_data->op_bias |= MDS_MIGRATE_NSONLY;
5160 if (S_ISREG(child_inode->i_mode)) {
5161 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
5165 GOTO(out_unlock, rc);
5168 rc = ll_data_version(child_inode, &data_version,
5171 GOTO(out_close, rc);
5173 op_data->op_open_handle = och->och_open_handle;
5174 op_data->op_data_version = data_version;
5175 op_data->op_lease_handle = och->och_lease_handle;
5176 op_data->op_bias |= MDS_CLOSE_MIGRATE;
5178 spin_lock(&och->och_mod->mod_open_req->rq_lock);
5179 och->och_mod->mod_open_req->rq_replay = 0;
5180 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
5183 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
5184 op_data->op_name, op_data->op_namelen,
5185 op_data->op_name, op_data->op_namelen, &request);
5187 LASSERT(request != NULL);
5188 ll_update_times(request, parent);
5191 if (rc == 0 || rc == -EAGAIN) {
5192 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
5193 LASSERT(body != NULL);
5195 /* If the server does release layout lock, then we cleanup
5196 * the client och here, otherwise release it in out_close: */
5197 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
5198 obd_mod_put(och->och_mod);
5199 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
5201 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
5207 if (request != NULL) {
5208 ptlrpc_req_finished(request);
5212 /* Try again if the lease has cancelled. */
5213 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
5218 ll_lease_close(och, child_inode, NULL);
5220 clear_nlink(child_inode);
5222 inode_unlock(child_inode);
5223 ll_finish_md_op_data(op_data);
5230 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
5232 struct ll_file_data *fd = file->private_data;
5236 * In order to avoid flood of warning messages, only print one message
5237 * for one file. And the entire message rate on the client is limited
5238 * by CDEBUG_LIMIT too.
5240 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5241 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5242 CDEBUG_LIMIT(D_CONSOLE,
5243 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5249 * test if some locks matching bits and l_req_mode are acquired
5250 * - bits can be in different locks
5251 * - if found clear the common lock bits in *bits
5252 * - the bits not found, are kept in *bits
5254 * \param bits [IN] searched lock bits [IN]
5255 * \param l_req_mode [IN] searched lock mode
5256 * \retval boolean, true iff all bits are found
5258 int ll_have_md_lock(struct obd_export *exp, struct inode *inode, __u64 *bits,
5259 enum ldlm_mode l_req_mode)
5261 struct lustre_handle lockh;
5262 union ldlm_policy_data policy;
5263 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5264 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5273 fid = &ll_i2info(inode)->lli_fid;
5274 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5275 ldlm_lockname[mode]);
5277 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5278 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5279 policy.l_inodebits.bits = *bits & BIT(i);
5280 if (policy.l_inodebits.bits == 0)
5283 if (md_lock_match(exp, flags, fid, LDLM_IBITS, &policy, mode,
5285 struct ldlm_lock *lock;
5287 lock = ldlm_handle2lock(&lockh);
5290 ~(lock->l_policy_data.l_inodebits.bits);
5291 LDLM_LOCK_PUT(lock);
5293 *bits &= ~policy.l_inodebits.bits;
5300 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5301 struct lustre_handle *lockh, __u64 flags,
5302 enum ldlm_mode mode)
5304 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5309 fid = &ll_i2info(inode)->lli_fid;
5310 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5312 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5313 fid, LDLM_IBITS, &policy, mode, lockh);
5318 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5320 /* Already unlinked. Just update nlink and return success */
5321 if (rc == -ENOENT) {
5323 /* If it is striped directory, and there is bad stripe
5324 * Let's revalidate the dentry again, instead of returning
5326 if (ll_dir_striped(inode))
5329 /* This path cannot be hit for regular files unless in
5330 * case of obscure races, so no need to to validate
5332 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5334 } else if (rc != 0) {
5335 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5336 "%s: revalidate FID "DFID" error: rc = %d\n",
5337 ll_i2sbi(inode)->ll_fsname,
5338 PFID(ll_inode2fid(inode)), rc);
5344 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5346 struct inode *parent;
5347 struct inode *inode = dentry->d_inode;
5348 struct obd_export *exp = ll_i2mdexp(inode);
5349 struct lookup_intent oit = {
5352 struct ptlrpc_request *req = NULL;
5353 struct md_op_data *op_data;
5354 const char *name = NULL;
5359 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5360 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5362 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5363 parent = dentry->d_parent->d_inode;
5364 name = dentry->d_name.name;
5365 namelen = dentry->d_name.len;
5370 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5371 LUSTRE_OPC_ANY, NULL);
5372 if (IS_ERR(op_data))
5373 RETURN(PTR_ERR(op_data));
5375 /* Call getattr by fid */
5376 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5377 op_data->op_flags = MF_GETATTR_BY_FID;
5378 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5379 ll_finish_md_op_data(op_data);
5381 rc = ll_inode_revalidate_fini(inode, rc);
5385 rc = ll_revalidate_it_finish(req, &oit, dentry);
5387 ll_intent_release(&oit);
5391 /* Unlinked? Unhash dentry, so it is not picked up later by
5392 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5393 * here to preserve get_cwd functionality on 2.6.
5395 if (!dentry->d_inode->i_nlink)
5396 d_lustre_invalidate(dentry);
5398 ll_lookup_finish_locks(&oit, dentry);
5400 ptlrpc_req_finished(req);
5405 static int ll_merge_md_attr(struct inode *inode)
5407 struct ll_inode_info *lli = ll_i2info(inode);
5408 struct cl_attr attr = { 0 };
5411 if (!lli->lli_lsm_md)
5414 down_read(&lli->lli_lsm_sem);
5415 if (!lmv_dir_striped(lli->lli_lsm_md)) {
5416 up_read(&lli->lli_lsm_sem);
5419 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
5420 &attr, ll_md_blocking_ast);
5421 up_read(&lli->lli_lsm_sem);
5425 spin_lock(&inode->i_lock);
5426 set_nlink(inode, attr.cat_nlink);
5427 spin_unlock(&inode->i_lock);
5429 inode->i_blocks = attr.cat_blocks;
5430 i_size_write(inode, attr.cat_size);
5432 ll_i2info(inode)->lli_atime = attr.cat_atime;
5433 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5434 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5439 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5440 unsigned int flags, bool foreign)
5442 struct inode *inode = de->d_inode;
5443 struct ll_sb_info *sbi = ll_i2sbi(inode);
5444 struct ll_inode_info *lli = ll_i2info(inode);
5445 struct inode *dir = de->d_parent->d_inode;
5446 bool need_glimpse = true;
5447 ktime_t kstart = ktime_get();
5450 /* The OST object(s) determine the file size, blocks and mtime. */
5451 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5452 request_mask & STATX_MTIME))
5453 need_glimpse = false;
5455 if (dentry_may_statahead(dir, de))
5456 ll_start_statahead(dir, de, need_glimpse &&
5457 !(flags & AT_STATX_DONT_SYNC));
5459 if (flags & AT_STATX_DONT_SYNC)
5460 GOTO(fill_attr, rc = 0);
5462 rc = ll_inode_revalidate(de, IT_GETATTR);
5466 /* foreign file/dir are always of zero length, so don't
5467 * need to validate size.
5469 if (S_ISREG(inode->i_mode) && !foreign) {
5473 GOTO(fill_attr, rc);
5475 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5476 if (cached && rc < 0)
5480 GOTO(fill_attr, rc);
5483 * If the returned attr is masked with OBD_MD_FLSIZE &
5484 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5485 * or blocks obtained from MDT is strictly correct, and the file
5486 * is usually not being modified by clients, and the [a|m|c]time
5487 * got from MDT is also strictly correct.
5488 * Under this circumstance, it does not need to send glimpse
5489 * RPCs to OSTs for file attributes such as the size and blocks.
5491 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5492 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5493 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5494 inode->i_mtime.tv_sec = lli->lli_mtime;
5495 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5496 inode->i_atime.tv_sec = lli->lli_atime;
5497 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5498 inode->i_ctime.tv_sec = lli->lli_ctime;
5499 GOTO(fill_attr, rc);
5502 /* In case of restore, the MDT has the right size and has
5503 * already send it back without granting the layout lock,
5504 * inode is up-to-date so glimpse is useless.
5505 * Also to glimpse we need the layout, in case of a running
5506 * restore the MDT holds the layout lock so the glimpse will
5507 * block up to the end of restore (getattr will block)
5509 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5510 rc = ll_glimpse_size(inode);
5515 /* If object isn't regular a file then don't validate size. */
5516 /* foreign dir is not striped dir */
5517 if (ll_dir_striped(inode) && !foreign) {
5518 rc = ll_merge_md_attr(inode);
5523 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5524 inode->i_atime.tv_sec = lli->lli_atime;
5525 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5526 inode->i_mtime.tv_sec = lli->lli_mtime;
5527 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5528 inode->i_ctime.tv_sec = lli->lli_ctime;
5532 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5534 if (ll_need_32bit_api(sbi)) {
5535 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5536 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5537 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5539 stat->ino = inode->i_ino;
5540 stat->dev = inode->i_sb->s_dev;
5541 stat->rdev = inode->i_rdev;
5544 /* foreign symlink to be exposed as a real symlink */
5546 stat->mode = inode->i_mode;
5548 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5550 stat->uid = inode->i_uid;
5551 stat->gid = inode->i_gid;
5552 stat->atime = inode->i_atime;
5553 stat->mtime = inode->i_mtime;
5554 stat->ctime = inode->i_ctime;
5555 /* stat->blksize is used to tell about preferred IO size */
5556 if (sbi->ll_stat_blksize)
5557 stat->blksize = sbi->ll_stat_blksize;
5558 else if (S_ISREG(inode->i_mode))
5559 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5560 LL_MAX_BLKSIZE_BITS);
5562 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5564 stat->nlink = inode->i_nlink;
5565 stat->size = i_size_read(inode);
5566 stat->blocks = inode->i_blocks;
5568 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5569 if (flags & AT_STATX_DONT_SYNC) {
5570 if (stat->size == 0 &&
5571 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5572 stat->size = lli->lli_lazysize;
5573 if (stat->blocks == 0 &&
5574 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5575 stat->blocks = lli->lli_lazyblocks;
5578 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5579 stat->result_mask |= STATX_BTIME;
5580 stat->btime.tv_sec = lli->lli_btime;
5583 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5584 #ifdef HAVE_LUSTRE_CRYPTO
5585 stat->attributes_mask |= STATX_ATTR_ENCRYPTED;
5587 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5588 /* if Lustre specific LUSTRE_ENCRYPT_FL flag is set, also set
5589 * ext4 equivalent to please statx
5591 if (stat->attributes & LUSTRE_ENCRYPT_FL)
5592 stat->attributes |= STATX_ATTR_ENCRYPTED;
5593 stat->result_mask &= request_mask;
5596 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5597 ktime_us_delta(ktime_get(), kstart));
5602 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5603 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
5604 struct kstat *stat, u32 request_mask, unsigned int flags)
5606 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5610 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5612 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5613 AT_STATX_SYNC_AS_STAT, false);
5617 int cl_falloc(struct file *file, struct inode *inode, int mode, loff_t offset,
5620 loff_t size = i_size_read(inode);
5628 env = cl_env_get(&refcheck);
5630 RETURN(PTR_ERR(env));
5632 io = vvp_env_thread_io(env);
5633 io->ci_obj = ll_i2info(inode)->lli_clob;
5634 ll_io_set_mirror(io, file);
5636 io->ci_verify_layout = 1;
5637 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5638 io->u.ci_setattr.sa_falloc_mode = mode;
5639 io->u.ci_setattr.sa_falloc_offset = offset;
5640 io->u.ci_setattr.sa_falloc_end = offset + len;
5641 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5643 CDEBUG(D_INODE, "UID %u GID %u PRJID %u\n",
5644 from_kuid(&init_user_ns, inode->i_uid),
5645 from_kgid(&init_user_ns, inode->i_gid),
5646 ll_i2info(inode)->lli_projid);
5648 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5649 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5650 io->u.ci_setattr.sa_falloc_projid = ll_i2info(inode)->lli_projid;
5652 if (io->u.ci_setattr.sa_falloc_end > size) {
5653 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5655 /* Check new size against VFS/VM file size limit and rlimit */
5656 rc = inode_newsize_ok(inode, newsize);
5659 if (newsize > ll_file_maxbytes(inode)) {
5660 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5661 (unsigned long long)newsize,
5662 ll_file_maxbytes(inode));
5669 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5671 rc = cl_io_loop(env, io);
5674 cl_io_fini(env, io);
5675 } while (unlikely(io->ci_need_restart));
5678 cl_env_put(env, &refcheck);
5682 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5684 struct inode *inode = file_inode(filp);
5687 if (offset < 0 || len <= 0)
5690 * Encrypted inodes can't handle collapse range or zero range or insert
5691 * range since we would need to re-encrypt blocks with a different IV or
5692 * XTS tweak (which are based on the logical block number).
5693 * Similar to what ext4 does.
5695 if (IS_ENCRYPTED(inode) &&
5696 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5697 FALLOC_FL_ZERO_RANGE)))
5698 RETURN(-EOPNOTSUPP);
5701 * mode == 0 (which is standard prealloc) and PUNCH is supported
5702 * Rest of mode options are not supported yet.
5704 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5705 RETURN(-EOPNOTSUPP);
5707 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5709 rc = cl_falloc(filp, inode, mode, offset, len);
5711 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5712 * used by Lustre in several places. Retuning it here would
5713 * confuse applications that explicity test for EOPNOTSUPP
5714 * (95) and fall back to ftruncate().
5716 if (rc == -ENOTSUPP)
5722 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5723 __u64 start, __u64 len)
5727 struct fiemap *fiemap;
5728 unsigned int extent_count = fieinfo->fi_extents_max;
5730 num_bytes = sizeof(*fiemap) + (extent_count *
5731 sizeof(struct fiemap_extent));
5732 OBD_ALLOC_LARGE(fiemap, num_bytes);
5737 fiemap->fm_flags = fieinfo->fi_flags;
5738 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5739 fiemap->fm_start = start;
5740 fiemap->fm_length = len;
5741 if (extent_count > 0 &&
5742 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5743 sizeof(struct fiemap_extent)) != 0)
5744 GOTO(out, rc = -EFAULT);
5746 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5748 if (IS_ENCRYPTED(inode)) {
5751 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5752 fiemap->fm_extents[i].fe_flags |=
5753 FIEMAP_EXTENT_DATA_ENCRYPTED |
5754 FIEMAP_EXTENT_ENCODED;
5757 fieinfo->fi_flags = fiemap->fm_flags;
5758 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5759 if (extent_count > 0 &&
5760 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5761 fiemap->fm_mapped_extents *
5762 sizeof(struct fiemap_extent)) != 0)
5763 GOTO(out, rc = -EFAULT);
5765 OBD_FREE_LARGE(fiemap, num_bytes);
5769 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
5773 struct ll_sb_info *sbi;
5774 struct root_squash_info *squash;
5775 struct cred *cred = NULL;
5776 const struct cred *old_cred = NULL;
5777 bool squash_id = false;
5778 ktime_t kstart = ktime_get();
5782 if (mask & MAY_NOT_BLOCK)
5786 * as root inode are NOT getting validated in lookup operation,
5787 * need to revalidate PERM before permission check.
5789 if (is_root_inode(inode)) {
5790 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_GETATTR);
5795 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5796 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5798 /* squash fsuid/fsgid if needed */
5799 sbi = ll_i2sbi(inode);
5800 squash = &sbi->ll_squash;
5801 if (unlikely(squash->rsi_uid != 0 &&
5802 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5803 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5807 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5808 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5809 squash->rsi_uid, squash->rsi_gid);
5811 /* update current process's credentials
5812 * and FS capability */
5813 cred = prepare_creds();
5817 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5818 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5819 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5820 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5822 old_cred = override_creds(cred);
5825 rc = generic_permission(mnt_userns, inode, mask);
5826 /* restore current process's credentials and FS capability */
5828 revert_creds(old_cred);
5833 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5834 ktime_us_delta(ktime_get(), kstart));
5839 /* -o localflock - only provides locally consistent flock locks */
5840 static const struct file_operations ll_file_operations = {
5841 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5842 # ifdef HAVE_SYNC_READ_WRITE
5843 .read = new_sync_read,
5844 .write = new_sync_write,
5846 .read_iter = ll_file_read_iter,
5847 .write_iter = ll_file_write_iter,
5848 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5849 .read = ll_file_read,
5850 .aio_read = ll_file_aio_read,
5851 .write = ll_file_write,
5852 .aio_write = ll_file_aio_write,
5853 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5854 .unlocked_ioctl = ll_file_ioctl,
5855 .open = ll_file_open,
5856 .release = ll_file_release,
5857 .mmap = ll_file_mmap,
5858 .llseek = ll_file_seek,
5859 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5860 .splice_read = generic_file_splice_read,
5862 .splice_read = pcc_file_splice_read,
5864 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5865 .splice_write = iter_file_splice_write,
5869 .fallocate = ll_fallocate,
5872 static const struct file_operations ll_file_operations_flock = {
5873 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5874 # ifdef HAVE_SYNC_READ_WRITE
5875 .read = new_sync_read,
5876 .write = new_sync_write,
5877 # endif /* HAVE_SYNC_READ_WRITE */
5878 .read_iter = ll_file_read_iter,
5879 .write_iter = ll_file_write_iter,
5880 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5881 .read = ll_file_read,
5882 .aio_read = ll_file_aio_read,
5883 .write = ll_file_write,
5884 .aio_write = ll_file_aio_write,
5885 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5886 .unlocked_ioctl = ll_file_ioctl,
5887 .open = ll_file_open,
5888 .release = ll_file_release,
5889 .mmap = ll_file_mmap,
5890 .llseek = ll_file_seek,
5891 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5892 .splice_read = generic_file_splice_read,
5894 .splice_read = pcc_file_splice_read,
5896 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5897 .splice_write = iter_file_splice_write,
5901 .flock = ll_file_flock,
5902 .lock = ll_file_flock,
5903 .fallocate = ll_fallocate,
5906 /* These are for -o noflock - to return ENOSYS on flock calls */
5907 static const struct file_operations ll_file_operations_noflock = {
5908 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5909 # ifdef HAVE_SYNC_READ_WRITE
5910 .read = new_sync_read,
5911 .write = new_sync_write,
5912 # endif /* HAVE_SYNC_READ_WRITE */
5913 .read_iter = ll_file_read_iter,
5914 .write_iter = ll_file_write_iter,
5915 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5916 .read = ll_file_read,
5917 .aio_read = ll_file_aio_read,
5918 .write = ll_file_write,
5919 .aio_write = ll_file_aio_write,
5920 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5921 .unlocked_ioctl = ll_file_ioctl,
5922 .open = ll_file_open,
5923 .release = ll_file_release,
5924 .mmap = ll_file_mmap,
5925 .llseek = ll_file_seek,
5926 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5927 .splice_read = generic_file_splice_read,
5929 .splice_read = pcc_file_splice_read,
5931 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5932 .splice_write = iter_file_splice_write,
5936 .flock = ll_file_noflock,
5937 .lock = ll_file_noflock,
5938 .fallocate = ll_fallocate,
5941 const struct inode_operations ll_file_inode_operations = {
5942 .setattr = ll_setattr,
5943 .getattr = ll_getattr,
5944 .permission = ll_inode_permission,
5945 #ifdef HAVE_IOP_XATTR
5946 .setxattr = ll_setxattr,
5947 .getxattr = ll_getxattr,
5948 .removexattr = ll_removexattr,
5950 .listxattr = ll_listxattr,
5951 .fiemap = ll_fiemap,
5952 .get_acl = ll_get_acl,
5953 #ifdef HAVE_IOP_SET_ACL
5954 .set_acl = ll_set_acl,
5958 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5960 const struct file_operations *fops = &ll_file_operations_noflock;
5962 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
5963 fops = &ll_file_operations_flock;
5964 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
5965 fops = &ll_file_operations;
5970 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5972 struct ll_inode_info *lli = ll_i2info(inode);
5973 struct cl_object *obj = lli->lli_clob;
5982 env = cl_env_get(&refcheck);
5984 RETURN(PTR_ERR(env));
5986 rc = cl_conf_set(env, lli->lli_clob, conf);
5990 if (conf->coc_opc == OBJECT_CONF_SET) {
5991 struct ldlm_lock *lock = conf->coc_lock;
5992 struct cl_layout cl = {
5996 LASSERT(lock != NULL);
5997 LASSERT(ldlm_has_layout(lock));
5999 /* it can only be allowed to match after layout is
6000 * applied to inode otherwise false layout would be
6001 * seen. Applying layout shoud happen before dropping
6002 * the intent lock. */
6003 ldlm_lock_allow_match(lock);
6005 rc = cl_object_layout_get(env, obj, &cl);
6010 DFID": layout version change: %u -> %u\n",
6011 PFID(&lli->lli_fid), ll_layout_version_get(lli),
6013 ll_layout_version_set(lli, cl.cl_layout_gen);
6017 cl_env_put(env, &refcheck);
6019 RETURN(rc < 0 ? rc : 0);
6022 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
6023 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
6026 struct ll_sb_info *sbi = ll_i2sbi(inode);
6027 struct ptlrpc_request *req;
6034 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
6035 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
6036 lock->l_lvb_data, lock->l_lvb_len);
6038 if (lock->l_lvb_data != NULL)
6041 /* if layout lock was granted right away, the layout is returned
6042 * within DLM_LVB of dlm reply; otherwise if the lock was ever
6043 * blocked and then granted via completion ast, we have to fetch
6044 * layout here. Please note that we can't use the LVB buffer in
6045 * completion AST because it doesn't have a large enough buffer */
6046 rc = ll_get_default_mdsize(sbi, &lmmsize);
6050 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
6051 XATTR_NAME_LOV, lmmsize, &req);
6054 GOTO(out, rc = 0); /* empty layout */
6061 if (lmmsize == 0) /* empty layout */
6064 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
6066 GOTO(out, rc = -EFAULT);
6068 OBD_ALLOC_LARGE(lvbdata, lmmsize);
6069 if (lvbdata == NULL)
6070 GOTO(out, rc = -ENOMEM);
6072 memcpy(lvbdata, lmm, lmmsize);
6073 lock_res_and_lock(lock);
6074 if (unlikely(lock->l_lvb_data == NULL)) {
6075 lock->l_lvb_type = LVB_T_LAYOUT;
6076 lock->l_lvb_data = lvbdata;
6077 lock->l_lvb_len = lmmsize;
6080 unlock_res_and_lock(lock);
6083 OBD_FREE_LARGE(lvbdata, lmmsize);
6088 ptlrpc_req_finished(req);
6093 * Apply the layout to the inode. Layout lock is held and will be released
6096 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
6097 struct inode *inode)
6099 struct ll_inode_info *lli = ll_i2info(inode);
6100 struct ll_sb_info *sbi = ll_i2sbi(inode);
6101 struct ldlm_lock *lock;
6102 struct cl_object_conf conf;
6105 bool wait_layout = false;
6108 LASSERT(lustre_handle_is_used(lockh));
6110 lock = ldlm_handle2lock(lockh);
6111 LASSERT(lock != NULL);
6113 if (!ldlm_has_layout(lock))
6114 GOTO(out, rc = -EAGAIN);
6116 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
6117 PFID(&lli->lli_fid), inode);
6119 /* in case this is a caching lock and reinstate with new inode */
6120 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
6122 lock_res_and_lock(lock);
6123 lvb_ready = ldlm_is_lvb_ready(lock);
6124 unlock_res_and_lock(lock);
6126 /* checking lvb_ready is racy but this is okay. The worst case is
6127 * that multi processes may configure the file on the same time. */
6131 rc = ll_layout_fetch(inode, lock);
6135 /* for layout lock, lmm is stored in lock's lvb.
6136 * lvb_data is immutable if the lock is held so it's safe to access it
6139 * set layout to file. Unlikely this will fail as old layout was
6140 * surely eliminated */
6141 memset(&conf, 0, sizeof conf);
6142 conf.coc_opc = OBJECT_CONF_SET;
6143 conf.coc_inode = inode;
6144 conf.coc_lock = lock;
6145 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
6146 conf.u.coc_layout.lb_len = lock->l_lvb_len;
6147 rc = ll_layout_conf(inode, &conf);
6149 /* refresh layout failed, need to wait */
6150 wait_layout = rc == -EBUSY;
6153 LDLM_LOCK_PUT(lock);
6154 ldlm_lock_decref(lockh, mode);
6156 /* wait for IO to complete if it's still being used. */
6158 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
6159 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6161 memset(&conf, 0, sizeof conf);
6162 conf.coc_opc = OBJECT_CONF_WAIT;
6163 conf.coc_inode = inode;
6164 rc = ll_layout_conf(inode, &conf);
6168 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
6169 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
6175 * Issue layout intent RPC to MDS.
6176 * \param inode [in] file inode
6177 * \param intent [in] layout intent
6179 * \retval 0 on success
6180 * \retval < 0 error code
6182 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
6184 struct ll_inode_info *lli = ll_i2info(inode);
6185 struct ll_sb_info *sbi = ll_i2sbi(inode);
6186 struct md_op_data *op_data;
6187 struct lookup_intent it;
6188 struct ptlrpc_request *req;
6192 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
6193 0, 0, LUSTRE_OPC_ANY, NULL);
6194 if (IS_ERR(op_data))
6195 RETURN(PTR_ERR(op_data));
6197 op_data->op_data = intent;
6198 op_data->op_data_size = sizeof(*intent);
6200 memset(&it, 0, sizeof(it));
6201 it.it_op = IT_LAYOUT;
6202 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
6203 intent->li_opc == LAYOUT_INTENT_TRUNC)
6204 it.it_flags = FMODE_WRITE;
6206 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
6207 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6209 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
6210 &ll_md_blocking_ast, 0);
6211 if (it.it_request != NULL)
6212 ptlrpc_req_finished(it.it_request);
6213 it.it_request = NULL;
6215 ll_finish_md_op_data(op_data);
6217 /* set lock data in case this is a new lock */
6219 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
6221 ll_intent_drop_lock(&it);
6227 * This function checks if there exists a LAYOUT lock on the client side,
6228 * or enqueues it if it doesn't have one in cache.
6230 * This function will not hold layout lock so it may be revoked any time after
6231 * this function returns. Any operations depend on layout should be redone
6234 * This function should be called before lov_io_init() to get an uptodate
6235 * layout version, the caller should save the version number and after IO
6236 * is finished, this function should be called again to verify that layout
6237 * is not changed during IO time.
6239 int ll_layout_refresh(struct inode *inode, __u32 *gen)
6241 struct ll_inode_info *lli = ll_i2info(inode);
6242 struct ll_sb_info *sbi = ll_i2sbi(inode);
6243 struct lustre_handle lockh;
6244 struct layout_intent intent = {
6245 .li_opc = LAYOUT_INTENT_ACCESS,
6247 enum ldlm_mode mode;
6251 *gen = ll_layout_version_get(lli);
6252 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6253 *gen != CL_LAYOUT_GEN_NONE)
6257 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6258 LASSERT(S_ISREG(inode->i_mode));
6260 /* take layout lock mutex to enqueue layout lock exclusively. */
6261 mutex_lock(&lli->lli_layout_mutex);
6264 /* mostly layout lock is caching on the local side, so try to
6265 * match it before grabbing layout lock mutex. */
6266 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6267 LCK_CR | LCK_CW | LCK_PR |
6269 if (mode != 0) { /* hit cached lock */
6270 rc = ll_layout_lock_set(&lockh, mode, inode);
6276 rc = ll_layout_intent(inode, &intent);
6282 *gen = ll_layout_version_get(lli);
6283 mutex_unlock(&lli->lli_layout_mutex);
6289 * Issue layout intent RPC indicating where in a file an IO is about to write.
6291 * \param[in] inode file inode.
6292 * \param[in] ext write range with start offset of fille in bytes where
6293 * an IO is about to write, and exclusive end offset in
6296 * \retval 0 on success
6297 * \retval < 0 error code
6299 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6300 struct lu_extent *ext)
6302 struct layout_intent intent = {
6304 .li_extent.e_start = ext->e_start,
6305 .li_extent.e_end = ext->e_end,
6310 rc = ll_layout_intent(inode, &intent);
6316 * This function send a restore request to the MDT
6318 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6320 struct hsm_user_request *hur;
6324 len = sizeof(struct hsm_user_request) +
6325 sizeof(struct hsm_user_item);
6326 OBD_ALLOC(hur, len);
6330 hur->hur_request.hr_action = HUA_RESTORE;
6331 hur->hur_request.hr_archive_id = 0;
6332 hur->hur_request.hr_flags = 0;
6333 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6334 sizeof(hur->hur_user_item[0].hui_fid));
6335 hur->hur_user_item[0].hui_extent.offset = offset;
6336 hur->hur_user_item[0].hui_extent.length = length;
6337 hur->hur_request.hr_itemcount = 1;
6338 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,