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>
47 #ifdef HAVE_LINUX_FILELOCK_HEADER
48 #include <linux/filelock.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_swab.h>
54 #include "cl_object.h"
55 #include "llite_internal.h"
56 #include "vvp_internal.h"
59 struct inode *sp_inode;
64 __u64 pa_data_version;
70 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
72 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
75 static struct ll_file_data *ll_file_data_get(void)
77 struct ll_file_data *fd;
79 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
83 fd->fd_write_failed = false;
84 pcc_file_init(&fd->fd_pcc_file);
89 static void ll_file_data_put(struct ll_file_data *fd)
92 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
96 * Packs all the attributes into @op_data for the CLOSE rpc.
98 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
99 struct obd_client_handle *och)
103 ll_prep_md_op_data(op_data, inode, NULL, NULL,
104 0, 0, LUSTRE_OPC_ANY, NULL);
106 op_data->op_attr.ia_mode = inode->i_mode;
107 op_data->op_attr.ia_atime = inode->i_atime;
108 op_data->op_attr.ia_mtime = inode->i_mtime;
109 op_data->op_attr.ia_ctime = inode->i_ctime;
110 /* In case of encrypted file without the key, visible size was rounded
111 * up to next LUSTRE_ENCRYPTION_UNIT_SIZE, and clear text size was
112 * stored into lli_lazysize in ll_merge_attr(), so set proper file size
113 * now that we are closing.
115 if (llcrypt_require_key(inode) == -ENOKEY &&
116 ll_i2info(inode)->lli_attr_valid & OBD_MD_FLLAZYSIZE)
117 op_data->op_attr.ia_size = ll_i2info(inode)->lli_lazysize;
119 op_data->op_attr.ia_size = i_size_read(inode);
120 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
121 ATTR_MTIME | ATTR_MTIME_SET |
123 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
124 op_data->op_attr_blocks = inode->i_blocks;
125 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
126 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
127 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
128 op_data->op_open_handle = och->och_open_handle;
130 if (och->och_flags & FMODE_WRITE &&
131 test_and_clear_bit(LLIF_DATA_MODIFIED,
132 &ll_i2info(inode)->lli_flags))
133 /* For HSM: if inode data has been modified, pack it so that
134 * MDT can set data dirty flag in the archive. */
135 op_data->op_bias |= MDS_DATA_MODIFIED;
141 * Perform a close, possibly with a bias.
142 * The meaning of "data" depends on the value of "bias".
144 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
145 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
148 static int ll_close_inode_openhandle(struct inode *inode,
149 struct obd_client_handle *och,
150 enum mds_op_bias bias, void *data)
152 struct obd_export *md_exp = ll_i2mdexp(inode);
153 const struct ll_inode_info *lli = ll_i2info(inode);
154 struct md_op_data *op_data;
155 struct ptlrpc_request *req = NULL;
159 if (class_exp2obd(md_exp) == NULL) {
160 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
161 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
165 OBD_ALLOC_PTR(op_data);
166 /* We leak openhandle and request here on error, but not much to be
167 * done in OOM case since app won't retry close on error either. */
169 GOTO(out, rc = -ENOMEM);
171 ll_prepare_close(inode, op_data, och);
173 case MDS_CLOSE_LAYOUT_MERGE:
174 /* merge blocks from the victim inode */
175 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
176 op_data->op_attr.ia_valid |= ATTR_SIZE;
177 op_data->op_xvalid |= OP_XVALID_BLOCKS;
179 case MDS_CLOSE_LAYOUT_SPLIT:
180 case MDS_CLOSE_LAYOUT_SWAP: {
181 struct split_param *sp = data;
183 LASSERT(data != NULL);
184 op_data->op_bias |= bias;
185 op_data->op_data_version = 0;
186 op_data->op_lease_handle = och->och_lease_handle;
187 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
188 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
189 op_data->op_mirror_id = sp->sp_mirror_id;
191 op_data->op_fid2 = *ll_inode2fid(data);
196 case MDS_CLOSE_RESYNC_DONE: {
197 struct ll_ioc_lease *ioc = data;
199 LASSERT(data != NULL);
200 op_data->op_attr_blocks +=
201 ioc->lil_count * op_data->op_attr_blocks;
202 op_data->op_attr.ia_valid |= ATTR_SIZE;
203 op_data->op_xvalid |= OP_XVALID_BLOCKS;
204 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
206 op_data->op_lease_handle = och->och_lease_handle;
207 op_data->op_data = &ioc->lil_ids[0];
208 op_data->op_data_size =
209 ioc->lil_count * sizeof(ioc->lil_ids[0]);
213 case MDS_PCC_ATTACH: {
214 struct pcc_param *param = data;
216 LASSERT(data != NULL);
217 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
218 op_data->op_archive_id = param->pa_archive_id;
219 op_data->op_data_version = param->pa_data_version;
220 op_data->op_lease_handle = och->och_lease_handle;
224 case MDS_HSM_RELEASE:
225 LASSERT(data != NULL);
226 op_data->op_bias |= MDS_HSM_RELEASE;
227 op_data->op_data_version = *(__u64 *)data;
228 op_data->op_lease_handle = och->och_lease_handle;
229 op_data->op_attr.ia_valid |= ATTR_SIZE;
230 op_data->op_xvalid |= OP_XVALID_BLOCKS;
234 LASSERT(data == NULL);
238 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
239 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
240 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
241 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
243 rc = md_close(md_exp, op_data, och->och_mod, &req);
244 if (rc != 0 && rc != -EINTR)
245 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
246 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
248 if (rc == 0 && op_data->op_bias & bias) {
249 struct mdt_body *body;
251 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
252 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
255 if (bias & MDS_PCC_ATTACH) {
256 struct pcc_param *param = data;
258 param->pa_layout_gen = body->mbo_layout_gen;
262 ll_finish_md_op_data(op_data);
266 md_clear_open_replay_data(md_exp, och);
267 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
270 ptlrpc_req_finished(req); /* This is close request */
274 int ll_md_real_close(struct inode *inode, fmode_t fmode)
276 struct ll_inode_info *lli = ll_i2info(inode);
277 struct obd_client_handle **och_p;
278 struct obd_client_handle *och;
283 if (fmode & FMODE_WRITE) {
284 och_p = &lli->lli_mds_write_och;
285 och_usecount = &lli->lli_open_fd_write_count;
286 } else if (fmode & FMODE_EXEC) {
287 och_p = &lli->lli_mds_exec_och;
288 och_usecount = &lli->lli_open_fd_exec_count;
290 LASSERT(fmode & FMODE_READ);
291 och_p = &lli->lli_mds_read_och;
292 och_usecount = &lli->lli_open_fd_read_count;
295 mutex_lock(&lli->lli_och_mutex);
296 if (*och_usecount > 0) {
297 /* There are still users of this handle, so skip
299 mutex_unlock(&lli->lli_och_mutex);
305 mutex_unlock(&lli->lli_och_mutex);
308 /* There might be a race and this handle may already
310 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
316 static int ll_md_close(struct inode *inode, struct file *file)
318 union ldlm_policy_data policy = {
319 .l_inodebits = { MDS_INODELOCK_OPEN },
321 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
322 struct ll_file_data *fd = file->private_data;
323 struct ll_inode_info *lli = ll_i2info(inode);
324 struct lustre_handle lockh;
325 enum ldlm_mode lockmode;
329 /* clear group lock, if present */
330 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
331 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
333 mutex_lock(&lli->lli_och_mutex);
334 if (fd->fd_lease_och != NULL) {
336 struct obd_client_handle *lease_och;
338 lease_och = fd->fd_lease_och;
339 fd->fd_lease_och = NULL;
340 mutex_unlock(&lli->lli_och_mutex);
342 /* Usually the lease is not released when the
343 * application crashed, we need to release here. */
344 rc = ll_lease_close(lease_och, inode, &lease_broken);
346 mutex_lock(&lli->lli_och_mutex);
348 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
349 "Clean up lease "DFID" %d/%d\n",
350 PFID(&lli->lli_fid), rc, lease_broken);
353 if (fd->fd_och != NULL) {
354 struct obd_client_handle *och;
358 mutex_unlock(&lli->lli_och_mutex);
360 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
364 /* Let's see if we have good enough OPEN lock on the file and if
365 we can skip talking to MDS */
366 if (fd->fd_omode & FMODE_WRITE) {
368 LASSERT(lli->lli_open_fd_write_count);
369 lli->lli_open_fd_write_count--;
370 } else if (fd->fd_omode & FMODE_EXEC) {
372 LASSERT(lli->lli_open_fd_exec_count);
373 lli->lli_open_fd_exec_count--;
376 LASSERT(lli->lli_open_fd_read_count);
377 lli->lli_open_fd_read_count--;
379 mutex_unlock(&lli->lli_och_mutex);
381 /* LU-4398: do not cache write open lock if the file has exec bit */
382 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
383 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
384 LDLM_IBITS, &policy, lockmode, &lockh))
385 rc = ll_md_real_close(inode, fd->fd_omode);
388 file->private_data = NULL;
389 ll_file_data_put(fd);
394 /* While this returns an error code, fput() the caller does not, so we need
395 * to make every effort to clean up all of our state here. Also, applications
396 * rarely check close errors and even if an error is returned they will not
397 * re-try the close call.
399 int ll_file_release(struct inode *inode, struct file *file)
401 struct ll_file_data *fd;
402 struct ll_sb_info *sbi = ll_i2sbi(inode);
403 struct ll_inode_info *lli = ll_i2info(inode);
404 ktime_t kstart = ktime_get();
409 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
410 PFID(ll_inode2fid(inode)), inode);
412 fd = file->private_data;
415 /* The last ref on @file, maybe not the the owner pid of statahead,
416 * because parent and child process can share the same file handle. */
417 if (S_ISDIR(inode->i_mode) &&
418 (lli->lli_opendir_key == fd || fd->fd_sai))
419 ll_deauthorize_statahead(inode, fd);
421 if (is_root_inode(inode)) {
422 file->private_data = NULL;
423 ll_file_data_put(fd);
427 pcc_file_release(inode, file);
429 if (!S_ISDIR(inode->i_mode)) {
430 if (lli->lli_clob != NULL)
431 lov_read_and_clear_async_rc(lli->lli_clob);
432 lli->lli_async_rc = 0;
435 lli->lli_close_fd_time = ktime_get();
437 rc = ll_md_close(inode, file);
439 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
440 libcfs_debug_dumplog();
443 if (!rc && !is_root_inode(inode))
444 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
445 ktime_us_delta(ktime_get(), kstart));
449 static inline int ll_dom_readpage(void *data, struct page *page)
451 /* since ll_dom_readpage is a page cache helper, it is safe to assume
452 * mapping and host pointers are set here
455 struct niobuf_local *lnb = data;
459 inode = page2inode(page);
461 kaddr = kmap_atomic(page);
462 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
463 if (lnb->lnb_len < PAGE_SIZE)
464 memset(kaddr + lnb->lnb_len, 0,
465 PAGE_SIZE - lnb->lnb_len);
466 kunmap_atomic(kaddr);
468 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
469 if (!llcrypt_has_encryption_key(inode)) {
470 CDEBUG(D_SEC, "no enc key for "DFID"\n",
471 PFID(ll_inode2fid(inode)));
474 unsigned int offs = 0;
476 while (offs < PAGE_SIZE) {
477 /* decrypt only if page is not empty */
478 if (memcmp(page_address(page) + offs,
479 page_address(ZERO_PAGE(0)),
480 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
483 rc = llcrypt_decrypt_pagecache_blocks(page,
484 LUSTRE_ENCRYPTION_UNIT_SIZE,
489 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
494 flush_dcache_page(page);
495 SetPageUptodate(page);
502 #ifdef HAVE_READ_CACHE_PAGE_WANTS_FILE
503 static inline int ll_dom_read_folio(struct file *file, struct folio *folio0)
505 return ll_dom_readpage(file->private_data, folio_page(folio0, 0));
508 #define ll_dom_read_folio ll_dom_readpage
511 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
515 struct ll_inode_info *lli = ll_i2info(inode);
516 struct cl_object *obj = lli->lli_clob;
517 struct address_space *mapping = inode->i_mapping;
519 struct niobuf_remote *rnb;
520 struct mdt_body *body;
522 unsigned long index, start;
523 struct niobuf_local lnb;
532 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
536 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
537 if (rnb == NULL || rnb->rnb_len == 0)
540 /* LU-11595: Server may return whole file and that is OK always or
541 * it may return just file tail and its offset must be aligned with
542 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
543 * smaller then offset may be not aligned and that data is just ignored.
545 if (rnb->rnb_offset & ~PAGE_MASK)
548 /* Server returns whole file or just file tail if it fills in reply
549 * buffer, in both cases total size should be equal to the file size.
551 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
552 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
553 !(inode && IS_ENCRYPTED(inode))) {
554 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
555 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
556 rnb->rnb_len, body->mbo_dom_size);
560 env = cl_env_get(&refcheck);
563 io = vvp_env_thread_io(env);
565 io->ci_ignore_layout = 1;
566 rc = cl_io_init(env, io, CIT_MISC, obj);
570 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
571 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
573 data = (char *)rnb + sizeof(*rnb);
575 lnb.lnb_file_offset = rnb->rnb_offset;
576 start = lnb.lnb_file_offset >> PAGE_SHIFT;
578 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
579 lnb.lnb_page_offset = 0;
581 struct cl_page *page;
583 lnb.lnb_data = data + (index << PAGE_SHIFT);
584 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
585 if (lnb.lnb_len > PAGE_SIZE)
586 lnb.lnb_len = PAGE_SIZE;
588 vmpage = ll_read_cache_page(mapping, index + start,
589 ll_dom_read_folio, &lnb);
590 if (IS_ERR(vmpage)) {
591 CWARN("%s: cannot fill page %lu for "DFID
592 " with data: rc = %li\n",
593 ll_i2sbi(inode)->ll_fsname, index + start,
594 PFID(lu_object_fid(&obj->co_lu)),
599 if (vmpage->mapping == NULL) {
602 /* page was truncated */
605 /* attach VM page to CL page cache */
606 page = cl_page_find(env, obj, vmpage->index, vmpage,
609 ClearPageUptodate(vmpage);
614 SetPageUptodate(vmpage);
615 cl_page_put(env, page);
619 } while (rnb->rnb_len > (index << PAGE_SHIFT));
623 cl_env_put(env, &refcheck);
628 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
629 struct lookup_intent *itp)
631 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
632 struct dentry *parent = de->d_parent;
635 struct md_op_data *op_data;
636 struct ptlrpc_request *req = NULL;
640 LASSERT(parent != NULL);
641 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
643 /* if server supports open-by-fid, or file name is invalid, don't pack
644 * name in open request */
645 if (CFS_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
646 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
648 len = de->d_name.len;
649 name = kmalloc(len + 1, GFP_NOFS);
654 spin_lock(&de->d_lock);
655 if (len != de->d_name.len) {
656 spin_unlock(&de->d_lock);
660 memcpy(name, de->d_name.name, len);
662 spin_unlock(&de->d_lock);
664 if (!lu_name_is_valid_2(name, len)) {
670 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
671 name, len, 0, LUSTRE_OPC_OPEN, NULL);
672 if (IS_ERR(op_data)) {
674 RETURN(PTR_ERR(op_data));
676 op_data->op_data = lmm;
677 op_data->op_data_size = lmmsize;
679 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
681 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
682 &ll_md_blocking_ast, 0);
684 ll_finish_md_op_data(op_data);
686 /* reason for keep own exit path - don`t flood log
687 * with messages with -ESTALE errors.
689 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
690 it_open_error(DISP_OPEN_OPEN, itp))
692 ll_release_openhandle(de, itp);
696 if (it_disposition(itp, DISP_LOOKUP_NEG))
697 GOTO(out, rc = -ENOENT);
699 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
700 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
701 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
705 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
707 if (!rc && itp->it_lock_mode) {
710 /* If we got a lock back and it has a LOOKUP bit set,
711 * make sure the dentry is marked as valid so we can find it.
712 * We don't need to care about actual hashing since other bits
713 * of kernel will deal with that later.
715 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
716 if (bits & MDS_INODELOCK_LOOKUP)
717 d_lustre_revalidate(de);
719 /* if DoM bit returned along with LAYOUT bit then there
720 * can be read-on-open data returned.
722 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
723 ll_dom_finish_open(de->d_inode, req);
725 /* open may not fetch LOOKUP lock, update dir depth and default LMV
728 if (!rc && S_ISDIR(de->d_inode->i_mode))
729 ll_update_dir_depth_dmv(parent->d_inode, de);
732 ptlrpc_req_finished(req);
733 ll_intent_drop_lock(itp);
735 /* We did open by fid, but by the time we got to the server, the object
736 * disappeared. This is possible if the object was unlinked, but it's
737 * also possible if the object was unlinked by a rename. In the case
738 * of an object renamed over our existing one, we can't fail this open.
739 * O_CREAT also goes through this path if we had an existing dentry,
740 * and it's obviously wrong to return ENOENT for O_CREAT.
742 * Instead let's return -ESTALE, and the VFS will retry the open with
743 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
744 * revalidate, causing a lookup. This causes extra lookups in the case
745 * where we had a dentry in cache but the file is being unlinked and we
746 * lose the race with unlink, but this should be very rare.
754 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
755 struct obd_client_handle *och)
757 struct mdt_body *body;
759 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
760 och->och_open_handle = body->mbo_open_handle;
761 och->och_fid = body->mbo_fid1;
762 och->och_lease_handle.cookie = it->it_lock_handle;
763 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
764 och->och_flags = it->it_flags;
766 return md_set_open_replay_data(md_exp, och, it);
769 static int ll_local_open(struct file *file, struct lookup_intent *it,
770 struct ll_file_data *fd, struct obd_client_handle *och)
772 struct inode *inode = file_inode(file);
775 LASSERT(!file->private_data);
782 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
787 file->private_data = fd;
788 ll_readahead_init(inode, &fd->fd_ras);
789 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
794 void ll_track_file_opens(struct inode *inode)
796 struct ll_inode_info *lli = ll_i2info(inode);
797 struct ll_sb_info *sbi = ll_i2sbi(inode);
799 /* do not skew results with delays from never-opened inodes */
800 if (ktime_to_ns(lli->lli_close_fd_time))
801 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
802 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
804 if (ktime_after(ktime_get(),
805 ktime_add_ms(lli->lli_close_fd_time,
806 sbi->ll_oc_max_ms))) {
807 lli->lli_open_fd_count = 1;
808 lli->lli_close_fd_time = ns_to_ktime(0);
810 lli->lli_open_fd_count++;
813 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
814 lli->lli_open_fd_count);
817 /* Open a file, and (for the very first open) create objects on the OSTs at
818 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
819 * creation or open until ll_lov_setstripe() ioctl is called.
821 * If we already have the stripe MD locally then we don't request it in
822 * md_open(), by passing a lmm_size = 0.
824 * It is up to the application to ensure no other processes open this file
825 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
826 * used. We might be able to avoid races of that sort by getting lli_open_sem
827 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
828 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
830 int ll_file_open(struct inode *inode, struct file *file)
832 struct ll_inode_info *lli = ll_i2info(inode);
833 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
834 .it_flags = file->f_flags };
835 struct obd_client_handle **och_p = NULL;
836 __u64 *och_usecount = NULL;
837 struct ll_file_data *fd;
838 ktime_t kstart = ktime_get();
842 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
843 PFID(ll_inode2fid(inode)), inode, file->f_flags);
845 it = file->private_data; /* XXX: compat macro */
846 file->private_data = NULL; /* prevent ll_local_open assertion */
848 if (S_ISREG(inode->i_mode)) {
849 rc = ll_file_open_encrypt(inode, file);
851 if (it && it->it_disposition)
852 ll_release_openhandle(file_dentry(file), it);
853 GOTO(out_nofiledata, rc);
857 fd = ll_file_data_get();
859 GOTO(out_nofiledata, rc = -ENOMEM);
862 if (S_ISDIR(inode->i_mode))
863 ll_authorize_statahead(inode, fd);
865 ll_track_file_opens(inode);
866 if (is_root_inode(inode)) {
867 file->private_data = fd;
871 if (!it || !it->it_disposition) {
872 /* Convert f_flags into access mode. We cannot use file->f_mode,
873 * because everything but O_ACCMODE mask was stripped from
875 if ((oit.it_flags + 1) & O_ACCMODE)
877 if (file->f_flags & O_TRUNC)
878 oit.it_flags |= FMODE_WRITE;
880 /* kernel only call f_op->open in dentry_open. filp_open calls
881 * dentry_open after call to open_namei that checks permissions.
882 * Only nfsd_open call dentry_open directly without checking
883 * permissions and because of that this code below is safe.
885 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
886 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
888 /* We do not want O_EXCL here, presumably we opened the file
889 * already? XXX - NFS implications? */
890 oit.it_flags &= ~O_EXCL;
892 /* bug20584, if "it_flags" contains O_CREAT, the file will be
893 * created if necessary, then "IT_CREAT" should be set to keep
894 * consistent with it */
895 if (oit.it_flags & O_CREAT)
896 oit.it_op |= IT_CREAT;
902 /* Let's see if we have file open on MDS already. */
903 if (it->it_flags & FMODE_WRITE) {
904 och_p = &lli->lli_mds_write_och;
905 och_usecount = &lli->lli_open_fd_write_count;
906 } else if (it->it_flags & FMODE_EXEC) {
907 och_p = &lli->lli_mds_exec_och;
908 och_usecount = &lli->lli_open_fd_exec_count;
910 och_p = &lli->lli_mds_read_och;
911 och_usecount = &lli->lli_open_fd_read_count;
914 mutex_lock(&lli->lli_och_mutex);
915 if (*och_p) { /* Open handle is present */
916 if (it_disposition(it, DISP_OPEN_OPEN)) {
917 /* Well, there's extra open request that we do not need,
918 * let's close it somehow. This will decref request. */
919 rc = it_open_error(DISP_OPEN_OPEN, it);
921 mutex_unlock(&lli->lli_och_mutex);
922 GOTO(out_openerr, rc);
925 ll_release_openhandle(file_dentry(file), it);
929 rc = ll_local_open(file, it, fd, NULL);
932 mutex_unlock(&lli->lli_och_mutex);
933 GOTO(out_openerr, rc);
936 LASSERT(*och_usecount == 0);
937 if (!it->it_disposition) {
938 struct dentry *dentry = file_dentry(file);
939 struct ll_sb_info *sbi = ll_i2sbi(inode);
940 int open_threshold = sbi->ll_oc_thrsh_count;
942 /* We cannot just request lock handle now, new ELC code
943 * means that one of other OPEN locks for this file
944 * could be cancelled, and since blocking ast handler
945 * would attempt to grab och_mutex as well, that would
946 * result in a deadlock
948 mutex_unlock(&lli->lli_och_mutex);
950 * Normally called under two situations:
951 * 1. fhandle / NFS export.
952 * 2. A race/condition on MDS resulting in no open
953 * handle to be returned from LOOKUP|OPEN request,
954 * for example if the target entry was a symlink.
956 * For NFSv3 we need to always cache the open lock
957 * for pre 5.5 Linux kernels.
959 * After reaching number of opens of this inode
960 * we always ask for an open lock on it to handle
961 * bad userspace actors that open and close files
962 * in a loop for absolutely no good reason
964 /* fhandle / NFS path. */
965 if (lli->lli_open_thrsh_count != UINT_MAX)
966 open_threshold = lli->lli_open_thrsh_count;
968 if (filename_is_volatile(dentry->d_name.name,
971 /* There really is nothing here, but this
972 * make this more readable I think.
973 * We do not want openlock for volatile
974 * files under any circumstances
976 } else if (open_threshold > 0) {
977 /* Take MDS_OPEN_LOCK with many opens */
978 if (lli->lli_open_fd_count >= open_threshold)
979 it->it_flags |= MDS_OPEN_LOCK;
981 /* If this is open after we just closed */
982 else if (ktime_before(ktime_get(),
983 ktime_add_ms(lli->lli_close_fd_time,
984 sbi->ll_oc_thrsh_ms)))
985 it->it_flags |= MDS_OPEN_LOCK;
989 * Always specify MDS_OPEN_BY_FID because we don't want
990 * to get file with different fid.
992 it->it_flags |= MDS_OPEN_BY_FID;
993 rc = ll_intent_file_open(dentry, NULL, 0, it);
995 GOTO(out_openerr, rc);
999 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
1001 GOTO(out_och_free, rc = -ENOMEM);
1005 /* md_intent_lock() didn't get a request ref if there was an
1006 * open error, so don't do cleanup on the request here
1008 /* XXX (green): Should not we bail out on any error here, not
1009 * just open error? */
1010 rc = it_open_error(DISP_OPEN_OPEN, it);
1012 GOTO(out_och_free, rc);
1014 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
1015 "inode %p: disposition %x, status %d\n", inode,
1016 it_disposition(it, ~0), it->it_status);
1018 rc = ll_local_open(file, it, fd, *och_p);
1020 GOTO(out_och_free, rc);
1023 rc = pcc_file_open(inode, file);
1025 GOTO(out_och_free, rc);
1027 mutex_unlock(&lli->lli_och_mutex);
1031 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1032 different kind of OPEN lock for this same inode gets cancelled
1033 by ldlm_cancel_lru */
1034 if (!S_ISREG(inode->i_mode))
1035 GOTO(out_och_free, rc);
1036 cl_lov_delay_create_clear(&file->f_flags);
1037 GOTO(out_och_free, rc);
1041 if (och_p && *och_p) {
1042 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1043 *och_p = NULL; /* OBD_FREE writes some magic there */
1046 mutex_unlock(&lli->lli_och_mutex);
1049 if (lli->lli_opendir_key == fd)
1050 ll_deauthorize_statahead(inode, fd);
1053 ll_file_data_put(fd);
1055 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1056 ktime_us_delta(ktime_get(), kstart));
1060 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1061 ptlrpc_req_finished(it->it_request);
1062 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1068 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1069 struct ldlm_lock_desc *desc, void *data, int flag)
1072 struct lustre_handle lockh;
1076 case LDLM_CB_BLOCKING:
1077 ldlm_lock2handle(lock, &lockh);
1078 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1080 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1084 case LDLM_CB_CANCELING:
1092 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1093 * and save it as fd->fd_och so as to force client to reopen the file even
1094 * if it has an open lock in cache already.
1096 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1097 struct lustre_handle *old_open_handle)
1099 struct ll_inode_info *lli = ll_i2info(inode);
1100 struct ll_file_data *fd = file->private_data;
1101 struct obd_client_handle **och_p;
1102 __u64 *och_usecount;
1106 /* Get the openhandle of the file */
1107 mutex_lock(&lli->lli_och_mutex);
1108 if (fd->fd_lease_och != NULL)
1109 GOTO(out_unlock, rc = -EBUSY);
1111 if (fd->fd_och == NULL) {
1112 if (file->f_mode & FMODE_WRITE) {
1113 LASSERT(lli->lli_mds_write_och != NULL);
1114 och_p = &lli->lli_mds_write_och;
1115 och_usecount = &lli->lli_open_fd_write_count;
1117 LASSERT(lli->lli_mds_read_och != NULL);
1118 och_p = &lli->lli_mds_read_och;
1119 och_usecount = &lli->lli_open_fd_read_count;
1122 if (*och_usecount > 1)
1123 GOTO(out_unlock, rc = -EBUSY);
1125 fd->fd_och = *och_p;
1130 *old_open_handle = fd->fd_och->och_open_handle;
1134 mutex_unlock(&lli->lli_och_mutex);
1139 * Release ownership on lli_mds_*_och when putting back a file lease.
1141 static int ll_lease_och_release(struct inode *inode, struct file *file)
1143 struct ll_inode_info *lli = ll_i2info(inode);
1144 struct ll_file_data *fd = file->private_data;
1145 struct obd_client_handle **och_p;
1146 struct obd_client_handle *old_och = NULL;
1147 __u64 *och_usecount;
1151 mutex_lock(&lli->lli_och_mutex);
1152 if (file->f_mode & FMODE_WRITE) {
1153 och_p = &lli->lli_mds_write_och;
1154 och_usecount = &lli->lli_open_fd_write_count;
1156 och_p = &lli->lli_mds_read_och;
1157 och_usecount = &lli->lli_open_fd_read_count;
1160 /* The file may have been open by another process (broken lease) so
1161 * *och_p is not NULL. In this case we should simply increase usecount
1164 if (*och_p != NULL) {
1165 old_och = fd->fd_och;
1168 *och_p = fd->fd_och;
1172 mutex_unlock(&lli->lli_och_mutex);
1174 if (old_och != NULL)
1175 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1181 * Acquire a lease and open the file.
1183 static struct obd_client_handle *
1184 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1187 struct lookup_intent it = { .it_op = IT_OPEN };
1188 struct ll_sb_info *sbi = ll_i2sbi(inode);
1189 struct md_op_data *op_data;
1190 struct ptlrpc_request *req = NULL;
1191 struct lustre_handle old_open_handle = { 0 };
1192 struct obd_client_handle *och = NULL;
1197 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1198 RETURN(ERR_PTR(-EINVAL));
1201 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1202 RETURN(ERR_PTR(-EPERM));
1204 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1206 RETURN(ERR_PTR(rc));
1211 RETURN(ERR_PTR(-ENOMEM));
1213 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1214 LUSTRE_OPC_ANY, NULL);
1215 if (IS_ERR(op_data))
1216 GOTO(out, rc = PTR_ERR(op_data));
1218 /* To tell the MDT this openhandle is from the same owner */
1219 op_data->op_open_handle = old_open_handle;
1221 it.it_flags = fmode | open_flags;
1222 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1223 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1224 &ll_md_blocking_lease_ast,
1225 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1226 * it can be cancelled which may mislead applications that the lease is
1228 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1229 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1230 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1231 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1232 ll_finish_md_op_data(op_data);
1233 ptlrpc_req_finished(req);
1235 GOTO(out_release_it, rc);
1237 if (it_disposition(&it, DISP_LOOKUP_NEG))
1238 GOTO(out_release_it, rc = -ENOENT);
1240 rc = it_open_error(DISP_OPEN_OPEN, &it);
1242 GOTO(out_release_it, rc);
1244 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1245 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1247 GOTO(out_release_it, rc);
1249 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1250 GOTO(out_close, rc = -EOPNOTSUPP);
1252 /* already get lease, handle lease lock */
1253 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1254 if (!it.it_lock_mode ||
1255 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1256 /* open lock must return for lease */
1257 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1258 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1260 GOTO(out_close, rc = -EPROTO);
1263 ll_intent_release(&it);
1267 /* Cancel open lock */
1268 if (it.it_lock_mode != 0) {
1269 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1271 it.it_lock_mode = 0;
1272 och->och_lease_handle.cookie = 0ULL;
1274 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1276 CERROR("%s: error closing file "DFID": %d\n",
1277 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1278 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1280 ll_intent_release(&it);
1284 RETURN(ERR_PTR(rc));
1288 * Check whether a layout swap can be done between two inodes.
1290 * \param[in] inode1 First inode to check
1291 * \param[in] inode2 Second inode to check
1293 * \retval 0 on success, layout swap can be performed between both inodes
1294 * \retval negative error code if requirements are not met
1296 static int ll_check_swap_layouts_validity(struct inode *inode1,
1297 struct inode *inode2)
1299 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1302 if (inode_permission(&nop_mnt_idmap, inode1, MAY_WRITE) ||
1303 inode_permission(&nop_mnt_idmap, inode2, MAY_WRITE))
1306 if (inode1->i_sb != inode2->i_sb)
1312 static int ll_swap_layouts_close(struct obd_client_handle *och,
1313 struct inode *inode, struct inode *inode2)
1315 const struct lu_fid *fid1 = ll_inode2fid(inode);
1316 const struct lu_fid *fid2;
1320 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1321 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1323 rc = ll_check_swap_layouts_validity(inode, inode2);
1325 GOTO(out_free_och, rc);
1327 /* We now know that inode2 is a lustre inode */
1328 fid2 = ll_inode2fid(inode2);
1330 rc = lu_fid_cmp(fid1, fid2);
1332 GOTO(out_free_och, rc = -EINVAL);
1334 /* Close the file and {swap,merge} layouts between inode & inode2.
1335 * NB: local lease handle is released in mdc_close_intent_pack()
1336 * because we still need it to pack l_remote_handle to MDT. */
1337 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1340 och = NULL; /* freed in ll_close_inode_openhandle() */
1350 * Release lease and close the file.
1351 * It will check if the lease has ever broken.
1353 static int ll_lease_close_intent(struct obd_client_handle *och,
1354 struct inode *inode,
1355 bool *lease_broken, enum mds_op_bias bias,
1358 struct ldlm_lock *lock;
1359 bool cancelled = true;
1363 lock = ldlm_handle2lock(&och->och_lease_handle);
1365 lock_res_and_lock(lock);
1366 cancelled = ldlm_is_cancel(lock);
1367 unlock_res_and_lock(lock);
1368 LDLM_LOCK_PUT(lock);
1371 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1372 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1374 if (lease_broken != NULL)
1375 *lease_broken = cancelled;
1377 if (!cancelled && !bias)
1378 ldlm_cli_cancel(&och->och_lease_handle, 0);
1380 if (cancelled) { /* no need to excute intent */
1385 rc = ll_close_inode_openhandle(inode, och, bias, data);
1389 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1392 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1396 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1398 static int ll_lease_file_resync(struct obd_client_handle *och,
1399 struct inode *inode, void __user *uarg)
1401 struct ll_sb_info *sbi = ll_i2sbi(inode);
1402 struct md_op_data *op_data;
1403 struct ll_ioc_lease_id ioc;
1404 __u64 data_version_unused;
1408 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1409 LUSTRE_OPC_ANY, NULL);
1410 if (IS_ERR(op_data))
1411 RETURN(PTR_ERR(op_data));
1413 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
1416 /* before starting file resync, it's necessary to clean up page cache
1417 * in client memory, otherwise once the layout version is increased,
1418 * writing back cached data will be denied the OSTs. */
1419 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1423 op_data->op_lease_handle = och->och_lease_handle;
1424 op_data->op_mirror_id = ioc.lil_mirror_id;
1425 rc = md_file_resync(sbi->ll_md_exp, op_data);
1431 ll_finish_md_op_data(op_data);
1435 static int ll_merge_attr_nolock(const struct lu_env *env, struct inode *inode)
1437 struct ll_inode_info *lli = ll_i2info(inode);
1438 struct cl_object *obj = lli->lli_clob;
1439 struct cl_attr *attr = vvp_env_thread_attr(env);
1447 /* Merge timestamps the most recently obtained from MDS with
1448 * timestamps obtained from OSTs.
1450 * Do not overwrite atime of inode because it may be refreshed
1451 * by file_accessed() function. If the read was served by cache
1452 * data, there is no RPC to be sent so that atime may not be
1453 * transferred to OSTs at all. MDT only updates atime at close time
1454 * if it's at least 'mdd.*.atime_diff' older.
1455 * All in all, the atime in Lustre does not strictly comply with
1456 * POSIX. Solving this problem needs to send an RPC to MDT for each
1457 * read, this will hurt performance.
1459 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1460 inode->i_atime.tv_sec < lli->lli_atime)
1461 inode->i_atime.tv_sec = lli->lli_atime;
1463 inode->i_mtime.tv_sec = lli->lli_mtime;
1464 inode->i_ctime.tv_sec = lli->lli_ctime;
1466 mtime = inode->i_mtime.tv_sec;
1467 atime = inode->i_atime.tv_sec;
1468 ctime = inode->i_ctime.tv_sec;
1470 cl_object_attr_lock(obj);
1471 if (CFS_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1474 rc = cl_object_attr_get(env, obj, attr);
1475 cl_object_attr_unlock(obj);
1478 GOTO(out, rc = (rc == -ENODATA ? 0 : rc));
1480 if (atime < attr->cat_atime)
1481 atime = attr->cat_atime;
1483 if (ctime < attr->cat_ctime)
1484 ctime = attr->cat_ctime;
1486 if (mtime < attr->cat_mtime)
1487 mtime = attr->cat_mtime;
1489 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu i_blocks %llu\n",
1490 PFID(&lli->lli_fid), attr->cat_size, attr->cat_blocks);
1492 if (llcrypt_require_key(inode) == -ENOKEY) {
1493 /* Without the key, round up encrypted file size to next
1494 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1495 * lli_lazysize for proper file size setting at close time.
1497 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1498 lli->lli_lazysize = attr->cat_size;
1499 attr->cat_size = round_up(attr->cat_size,
1500 LUSTRE_ENCRYPTION_UNIT_SIZE);
1502 i_size_write(inode, attr->cat_size);
1503 inode->i_blocks = attr->cat_blocks;
1505 inode->i_mtime.tv_sec = mtime;
1506 inode->i_atime.tv_sec = atime;
1507 inode->i_ctime.tv_sec = ctime;
1514 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1518 ll_inode_size_lock(inode);
1519 rc = ll_merge_attr_nolock(env, inode);
1520 ll_inode_size_unlock(inode);
1525 /* Use to update size and blocks on inode for LSOM if there is no contention */
1526 int ll_merge_attr_try(const struct lu_env *env, struct inode *inode)
1530 if (ll_inode_size_trylock(inode)) {
1531 rc = ll_merge_attr_nolock(env, inode);
1532 ll_inode_size_unlock(inode);
1539 * Set designated mirror for I/O.
1541 * So far only read, write, and truncated can support to issue I/O to
1542 * designated mirror.
1544 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1546 struct ll_file_data *fd = file->private_data;
1548 /* clear layout version for generic(non-resync) I/O in case it carries
1549 * stale layout version due to I/O restart */
1550 io->ci_layout_version = 0;
1552 /* FLR: disable non-delay for designated mirror I/O because obviously
1553 * only one mirror is available */
1554 if (fd->fd_designated_mirror > 0) {
1556 io->ci_designated_mirror = fd->fd_designated_mirror;
1557 io->ci_layout_version = fd->fd_layout_version;
1560 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1561 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1565 * This is relatime_need_update() from Linux 5.17, which is not exported.
1567 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1568 struct timespec64 now)
1571 if (!(mnt->mnt_flags & MNT_RELATIME))
1574 * Is mtime younger than atime? If yes, update atime:
1576 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1579 * Is ctime younger than atime? If yes, update atime:
1581 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1585 * Is the previous atime value older than a day? If yes,
1588 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1591 * Good, we can skip the atime update:
1597 * Very similar to kernel function: !__atime_needs_update()
1599 static bool file_is_noatime(const struct file *file)
1601 struct vfsmount *mnt = file->f_path.mnt;
1602 struct inode *inode = file_inode((struct file *)file);
1603 struct timespec64 now;
1605 if (file->f_flags & O_NOATIME)
1608 if (inode->i_flags & S_NOATIME)
1611 if (IS_NOATIME(inode))
1614 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1617 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1620 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1623 now = current_time(inode);
1625 if (!relatime_need_update(mnt, inode, now))
1631 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1632 struct vvp_io_args *args)
1634 struct inode *inode = file_inode(file);
1635 struct ll_file_data *fd = file->private_data;
1637 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1638 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1640 if (iot == CIT_WRITE) {
1641 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1642 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1643 file->f_flags & O_DIRECT ||
1645 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1646 io->u.ci_wr.wr_sync |= !!(args &&
1647 (args->u.normal.via_iocb->ki_flags &
1653 io->ci_iocb_nowait = !!(args &&
1654 (args->u.normal.via_iocb->ki_flags &
1658 io->ci_obj = ll_i2info(inode)->lli_clob;
1659 io->ci_lockreq = CILR_MAYBE;
1660 if (ll_file_nolock(file)) {
1661 io->ci_lockreq = CILR_NEVER;
1662 io->ci_no_srvlock = 1;
1663 } else if (file->f_flags & O_APPEND) {
1664 io->ci_lockreq = CILR_MANDATORY;
1666 io->ci_noatime = file_is_noatime(file);
1667 io->ci_async_readahead = false;
1669 /* FLR: only use non-delay I/O for read as there is only one
1670 * avaliable mirror for write. */
1671 io->ci_ndelay = !(iot == CIT_WRITE);
1673 ll_io_set_mirror(io, file);
1676 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1679 struct ll_inode_info *lli = ll_i2info(inode);
1680 struct ll_sb_info *sbi = ll_i2sbi(inode);
1681 enum obd_heat_type sample_type;
1682 enum obd_heat_type iobyte_type;
1683 __u64 now = ktime_get_real_seconds();
1685 if (!ll_sbi_has_file_heat(sbi) ||
1686 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1689 if (iot == CIT_READ) {
1690 sample_type = OBD_HEAT_READSAMPLE;
1691 iobyte_type = OBD_HEAT_READBYTE;
1692 } else if (iot == CIT_WRITE) {
1693 sample_type = OBD_HEAT_WRITESAMPLE;
1694 iobyte_type = OBD_HEAT_WRITEBYTE;
1699 spin_lock(&lli->lli_heat_lock);
1700 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1701 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1702 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1703 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1704 spin_unlock(&lli->lli_heat_lock);
1708 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1709 struct file *file, enum cl_io_type iot,
1710 loff_t *ppos, size_t bytes)
1712 struct vvp_io *vio = vvp_env_io(env);
1713 struct inode *inode = file_inode(file);
1714 struct ll_inode_info *lli = ll_i2info(inode);
1715 struct ll_sb_info *sbi = ll_i2sbi(inode);
1716 struct ll_file_data *fd = file->private_data;
1717 struct range_lock range;
1718 bool range_locked = false;
1724 unsigned int retried = 0, dio_lock = 0;
1725 bool is_aio = false;
1726 bool is_parallel_dio = false;
1727 struct cl_dio_aio *ci_dio_aio = NULL;
1728 size_t per_bytes, max_io_bytes;
1733 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, bytes: %zu\n",
1734 file_dentry(file)->d_name.name,
1735 iot == CIT_READ ? "read" : "write", *ppos, bytes);
1737 max_io_bytes = min_t(size_t, PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT,
1738 sbi->ll_cache->ccc_lru_max >> 2) << PAGE_SHIFT;
1740 io = vvp_env_thread_io(env);
1741 if (file->f_flags & O_DIRECT) {
1742 if (file->f_flags & O_APPEND)
1744 if (!is_sync_kiocb(args->u.normal.via_iocb))
1747 /* the kernel does not support AIO on pipes, and parallel DIO
1748 * uses part of the AIO path, so we must not do parallel dio
1751 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1754 if (!ll_sbi_has_parallel_dio(sbi))
1755 is_parallel_dio = false;
1757 ci_dio_aio = cl_dio_aio_alloc(args->u.normal.via_iocb,
1758 ll_i2info(inode)->lli_clob, is_aio);
1760 GOTO(out, rc = -ENOMEM);
1765 * IO block size need be aware of cached page limit, otherwise
1766 * if we have small max_cached_mb but large block IO issued, io
1767 * could not be finished and blocked whole client.
1769 if (file->f_flags & O_DIRECT || bytes < max_io_bytes) {
1773 per_bytes = max_io_bytes;
1776 io = vvp_env_thread_io(env);
1777 ll_io_init(io, file, iot, args);
1778 io->ci_dio_aio = ci_dio_aio;
1779 io->ci_dio_lock = dio_lock;
1780 io->ci_ndelay_tried = retried;
1781 io->ci_parallel_dio = is_parallel_dio;
1783 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1784 if (file->f_flags & O_APPEND)
1785 range_lock_init(&range, 0, LUSTRE_EOF);
1787 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1789 vio->vui_fd = file->private_data;
1790 vio->vui_iter = args->u.normal.via_iter;
1791 vio->vui_iocb = args->u.normal.via_iocb;
1792 /* Direct IO reads must also take range lock,
1793 * or multiple reads will try to work on the same pages
1794 * See LU-6227 for details.
1796 if (((iot == CIT_WRITE) ||
1797 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1798 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1799 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1801 rc = range_lock(&lli->lli_write_tree, &range);
1805 range_locked = true;
1808 ll_cl_add(inode, env, io, LCC_RW);
1809 rc = cl_io_loop(env, io);
1810 ll_cl_remove(inode, env);
1812 /* cl_io_rw_init() handled IO */
1816 if (io->ci_dio_aio && !is_aio) {
1817 struct cl_sync_io *anchor = &io->ci_dio_aio->cda_sync;
1819 /* for dio, EIOCBQUEUED is an implementation detail,
1820 * and we don't return it to userspace
1822 if (rc == -EIOCBQUEUED)
1825 /* N/B: parallel DIO may be disabled during i/o submission;
1826 * if that occurs, I/O shifts to sync, so it's all resolved
1827 * before we get here, and this wait call completes
1830 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1836 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1838 range_unlock(&lli->lli_write_tree, &range);
1839 range_locked = false;
1842 if (io->ci_bytes > 0) {
1844 result += io->ci_bytes;
1845 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1849 bytes -= io->ci_bytes;
1851 /* prepare IO restart */
1853 args->u.normal.via_iter = vio->vui_iter;
1857 * Reexpand iov count because it was zero
1860 iov_iter_reexpand(vio->vui_iter, bytes);
1861 if (per_bytes == io->ci_bytes)
1862 io->ci_need_restart = 1;
1866 cl_io_fini(env, io);
1869 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1870 file->f_path.dentry->d_name.name,
1871 iot, rc, result, io->ci_need_restart);
1873 if ((!rc || rc == -ENODATA || rc == -ENOLCK || rc == -EIOCBQUEUED) &&
1874 bytes > 0 && io->ci_need_restart && retries-- > 0) {
1876 "%s: restart %s from ppos=%lld bytes=%zu retries=%u ret=%zd: rc = %d\n",
1877 file_dentry(file)->d_name.name,
1878 iot == CIT_READ ? "read" : "write",
1879 *ppos, bytes, retries, result, rc);
1880 /* preserve the tried count for FLR */
1881 retried = io->ci_ndelay_tried;
1882 dio_lock = io->ci_dio_lock;
1886 if (io->ci_dio_aio) {
1887 /* set the number of bytes successfully moved in the aio */
1889 io->ci_dio_aio->cda_bytes = result;
1891 * VFS will call aio_complete() if no -EIOCBQUEUED
1892 * is returned for AIO, so we can not call aio_complete()
1893 * in our end_io(). (cda_no_aio_complete is always set for
1896 * NB: Setting cda_no_aio_complete like this is safe because
1897 * the atomic_dec_and_lock in cl_sync_io_note has implicit
1898 * memory barriers, so this will be seen by whichever thread
1899 * completes the DIO/AIO, even if it's not this one.
1901 if (is_aio && rc != -EIOCBQUEUED)
1902 io->ci_dio_aio->cda_no_aio_complete = 1;
1903 /* if an aio enqueued successfully (-EIOCBQUEUED), then Lustre
1904 * will call aio_complete rather than the vfs, so we return 0
1905 * to tell the VFS we're handling it
1907 else if (is_aio) /* rc == -EIOCBQUEUED */
1910 * Drop the reference held by the llite layer on this top level
1913 * For DIO, this frees it here, since IO is complete, and for
1914 * AIO, we will call aio_complete() (and then free this top
1915 * level context) once all the outstanding chunks of this AIO
1918 cl_sync_io_note(env, &io->ci_dio_aio->cda_sync,
1919 rc == -EIOCBQUEUED ? 0 : rc);
1921 LASSERT(io->ci_dio_aio->cda_creator_free);
1922 cl_dio_aio_free(env, io->ci_dio_aio);
1923 io->ci_dio_aio = NULL;
1927 if (iot == CIT_READ) {
1929 ll_stats_ops_tally(ll_i2sbi(inode),
1930 LPROC_LL_READ_BYTES, result);
1931 } else if (iot == CIT_WRITE) {
1933 ll_stats_ops_tally(ll_i2sbi(inode),
1934 LPROC_LL_WRITE_BYTES, result);
1935 fd->fd_write_failed = false;
1936 } else if (result == 0 && rc == 0) {
1939 fd->fd_write_failed = true;
1941 fd->fd_write_failed = false;
1942 } else if (rc != -ERESTARTSYS) {
1943 fd->fd_write_failed = true;
1947 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1949 ll_heat_add(inode, iot, result);
1951 RETURN(result > 0 ? result : rc);
1955 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1956 * especially for small I/O.
1958 * To serve a read request, CLIO has to create and initialize a cl_io and
1959 * then request DLM lock. This has turned out to have siginificant overhead
1960 * and affects the performance of small I/O dramatically.
1962 * It's not necessary to create a cl_io for each I/O. Under the help of read
1963 * ahead, most of the pages being read are already in memory cache and we can
1964 * read those pages directly because if the pages exist, the corresponding DLM
1965 * lock must exist so that page content must be valid.
1967 * In fast read implementation, the llite speculatively finds and reads pages
1968 * in memory cache. There are three scenarios for fast read:
1969 * - If the page exists and is uptodate, kernel VM will provide the data and
1970 * CLIO won't be intervened;
1971 * - If the page was brought into memory by read ahead, it will be exported
1972 * and read ahead parameters will be updated;
1973 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1974 * it will go back and invoke normal read, i.e., a cl_io will be created
1975 * and DLM lock will be requested.
1977 * POSIX compliance: posix standard states that read is intended to be atomic.
1978 * Lustre read implementation is in line with Linux kernel read implementation
1979 * and neither of them complies with POSIX standard in this matter. Fast read
1980 * doesn't make the situation worse on single node but it may interleave write
1981 * results from multiple nodes due to short read handling in ll_file_aio_read().
1983 * \param env - lu_env
1984 * \param iocb - kiocb from kernel
1985 * \param iter - user space buffers where the data will be copied
1987 * \retval - number of bytes have been read, or error code if error occurred.
1990 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1992 struct ll_inode_info *lli = ll_i2info(file_inode(iocb->ki_filp));
1995 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1998 /* NB: we can't do direct IO for fast read because it will need a lock
1999 * to make IO engine happy. */
2000 if (iocb->ki_filp->f_flags & O_DIRECT)
2003 if (ll_layout_version_get(lli) == CL_LAYOUT_GEN_NONE)
2006 result = generic_file_read_iter(iocb, iter);
2008 /* If the first page is not in cache, generic_file_aio_read() will be
2009 * returned with -ENODATA. Fall back to full read path.
2010 * See corresponding code in ll_readpage().
2012 * if we raced with page deletion, we might get EIO. Rather than add
2013 * locking to the fast path for this rare case, fall back to the full
2014 * read path. (See vvp_io_read_start() for rest of handling.
2016 if (result == -ENODATA || result == -EIO)
2020 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
2021 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
2022 LPROC_LL_READ_BYTES, result);
2029 * Confine read iter lest read beyond the EOF
2031 * \param iocb [in] kernel iocb
2032 * \param to [in] reader iov_iter
2034 * \retval <0 failure
2036 * \retval >0 @iocb->ki_pos has passed the EOF
2038 static int file_read_confine_iter(struct lu_env *env, struct kiocb *iocb,
2039 struct iov_iter *to)
2041 struct cl_attr *attr = vvp_env_thread_attr(env);
2042 struct file *file = iocb->ki_filp;
2043 struct inode *inode = file_inode(file);
2044 struct ll_inode_info *lli = ll_i2info(inode);
2045 loff_t read_end = iocb->ki_pos + iov_iter_count(to);
2050 cl_object_attr_lock(lli->lli_clob);
2051 rc = cl_object_attr_get(env, lli->lli_clob, attr);
2052 cl_object_attr_unlock(lli->lli_clob);
2056 kms = attr->cat_kms;
2057 /* if read beyond end-of-file, adjust read count */
2058 if (kms > 0 && (iocb->ki_pos >= kms || read_end > kms)) {
2059 rc = ll_glimpse_size(inode);
2063 size = i_size_read(inode);
2064 if (iocb->ki_pos >= size || read_end > size) {
2066 "%s: read [%llu, %llu] over eof, kms %llu, file_size %llu.\n",
2067 file_dentry(file)->d_name.name,
2068 iocb->ki_pos, read_end, kms, size);
2070 if (iocb->ki_pos >= size)
2073 if (read_end > size)
2074 iov_iter_truncate(to, size - iocb->ki_pos);
2082 * Read from a file (through the page cache).
2084 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2087 struct vvp_io_args *args;
2088 struct file *file = iocb->ki_filp;
2092 ktime_t kstart = ktime_get();
2094 bool stale_data = false;
2098 CDEBUG(D_VFSTRACE|D_IOTRACE,
2099 "START file %s:"DFID", ppos: %lld, count: %zu\n",
2100 file_dentry(file)->d_name.name,
2101 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2102 iov_iter_count(to));
2104 if (!iov_iter_count(to))
2107 env = cl_env_get(&refcheck);
2109 RETURN(PTR_ERR(env));
2111 result = file_read_confine_iter(env, iocb, to);
2114 else if (result > 0)
2118 * Currently when PCC read failed, we do not fall back to the
2119 * normal read path, just return the error.
2120 * The resaon is that: for RW-PCC, the file data may be modified
2121 * in the PCC and inconsistent with the data on OSTs (or file
2122 * data has been removed from the Lustre file system), at this
2123 * time, fallback to the normal read path may read the wrong
2125 * TODO: for RO-PCC (readonly PCC), fall back to normal read
2126 * path: read data from data copy on OSTs.
2128 result = pcc_file_read_iter(iocb, to, &cached);
2132 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
2134 result = ll_do_fast_read(iocb, to);
2135 if (result < 0 || iov_iter_count(to) == 0)
2138 args = ll_env_args(env);
2139 args->u.normal.via_iter = to;
2140 args->u.normal.via_iocb = iocb;
2142 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
2143 &iocb->ki_pos, iov_iter_count(to));
2146 else if (result == 0)
2150 cl_env_put(env, &refcheck);
2152 if (stale_data && result > 0) {
2154 * we've reached EOF before the read, the data read are cached
2157 iov_iter_truncate(to, 0);
2162 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2163 file->private_data, iocb->ki_pos, result,
2165 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2166 ktime_us_delta(ktime_get(), kstart));
2170 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2171 file_dentry(file)->d_name.name,
2172 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2173 iov_iter_count(to));
2179 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2180 * If a page is already in the page cache and dirty (and some other things -
2181 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2182 * write to it without doing a full I/O, because Lustre already knows about it
2183 * and will write it out. This saves a lot of processing time.
2185 * All writes here are within one page, so exclusion is handled by the page
2186 * lock on the vm page. We do not do tiny writes for writes which touch
2187 * multiple pages because it's very unlikely multiple sequential pages are
2188 * are already dirty.
2190 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2191 * and are unlikely to be to already dirty pages.
2193 * Attribute updates are important here, we do them in ll_tiny_write_end.
2195 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2197 ssize_t count = iov_iter_count(iter);
2198 struct file *file = iocb->ki_filp;
2199 struct inode *inode = file_inode(file);
2200 bool lock_inode = !IS_NOSEC(inode);
2205 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2206 * of function for why.
2208 if (count >= PAGE_SIZE ||
2209 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2212 if (unlikely(lock_inode))
2213 ll_inode_lock(inode);
2214 result = __generic_file_write_iter(iocb, iter);
2216 if (unlikely(lock_inode))
2217 ll_inode_unlock(inode);
2219 /* If the page is not already dirty, ll_tiny_write_begin returns
2220 * -ENODATA. We continue on to normal write.
2222 if (result == -ENODATA)
2226 ll_heat_add(inode, CIT_WRITE, result);
2227 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2230 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2236 * Write to a file (through the page cache).
2238 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2240 struct vvp_io_args *args;
2242 ssize_t rc_tiny = 0, rc_normal;
2243 struct file *file = iocb->ki_filp;
2246 ktime_t kstart = ktime_get();
2251 CDEBUG(D_VFSTRACE|D_IOTRACE,
2252 "START file %s:"DFID", ppos: %lld, count: %zu\n",
2253 file_dentry(file)->d_name.name,
2254 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2255 iov_iter_count(from));
2257 if (!iov_iter_count(from))
2258 GOTO(out, rc_normal = 0);
2261 * When PCC write failed, we usually do not fall back to the normal
2262 * write path, just return the error. But there is a special case when
2263 * returned error code is -ENOSPC due to running out of space on PCC HSM
2264 * bakcend. At this time, it will fall back to normal I/O path and
2265 * retry the I/O. As the file is in HSM released state, it will restore
2266 * the file data to OSTs first and redo the write again. And the
2267 * restore process will revoke the layout lock and detach the file
2268 * from PCC cache automatically.
2270 result = pcc_file_write_iter(iocb, from, &cached);
2271 if (cached && result != -ENOSPC && result != -EDQUOT)
2272 GOTO(out, rc_normal = result);
2274 /* NB: we can't do direct IO for tiny writes because they use the page
2275 * cache, we can't do sync writes because tiny writes can't flush
2276 * pages, and we can't do append writes because we can't guarantee the
2277 * required DLM locks are held to protect file size.
2279 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2280 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2281 rc_tiny = ll_do_tiny_write(iocb, from);
2283 /* In case of error, go on and try normal write - Only stop if tiny
2284 * write completed I/O.
2286 if (iov_iter_count(from) == 0)
2287 GOTO(out, rc_normal = rc_tiny);
2289 env = cl_env_get(&refcheck);
2291 RETURN(PTR_ERR(env));
2293 args = ll_env_args(env);
2294 args->u.normal.via_iter = from;
2295 args->u.normal.via_iocb = iocb;
2297 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2298 &iocb->ki_pos, iov_iter_count(from));
2300 /* On success, combine bytes written. */
2301 if (rc_tiny >= 0 && rc_normal > 0)
2302 rc_normal += rc_tiny;
2303 /* On error, only return error from normal write if tiny write did not
2304 * write any bytes. Otherwise return bytes written by tiny write.
2306 else if (rc_tiny > 0)
2307 rc_normal = rc_tiny;
2309 cl_env_put(env, &refcheck);
2311 if (rc_normal > 0) {
2312 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2313 file->private_data, iocb->ki_pos,
2315 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2316 ktime_us_delta(ktime_get(), kstart));
2320 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2321 file_dentry(file)->d_name.name,
2322 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2323 iov_iter_count(from));
2328 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2330 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2332 static int ll_file_get_iov_count(const struct iovec *iov,
2333 unsigned long *nr_segs, size_t *count,
2339 for (seg = 0; seg < *nr_segs; seg++) {
2340 const struct iovec *iv = &iov[seg];
2343 * If any segment has a negative length, or the cumulative
2344 * length ever wraps negative then return -EINVAL.
2347 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2349 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2354 cnt -= iv->iov_len; /* This segment is no good */
2361 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2362 unsigned long nr_segs, loff_t pos)
2369 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2376 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2377 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2378 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2379 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2380 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2382 result = ll_file_read_iter(iocb, &to);
2387 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2390 struct iovec iov = { .iov_base = buf, .iov_len = count };
2399 init_sync_kiocb(&kiocb, file);
2400 kiocb.ki_pos = *ppos;
2401 #ifdef HAVE_KIOCB_KI_LEFT
2402 kiocb.ki_left = count;
2403 #elif defined(HAVE_KI_NBYTES)
2404 kiocb.i_nbytes = count;
2407 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2408 *ppos = kiocb.ki_pos;
2414 * Write to a file (through the page cache).
2417 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2418 unsigned long nr_segs, loff_t pos)
2420 struct iov_iter from;
2425 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2432 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2433 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2434 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2435 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2436 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2438 result = ll_file_write_iter(iocb, &from);
2443 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2444 size_t count, loff_t *ppos)
2446 struct iovec iov = { .iov_base = (void __user *)buf,
2456 init_sync_kiocb(&kiocb, file);
2457 kiocb.ki_pos = *ppos;
2458 #ifdef HAVE_KIOCB_KI_LEFT
2459 kiocb.ki_left = count;
2460 #elif defined(HAVE_KI_NBYTES)
2461 kiocb.ki_nbytes = count;
2464 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2465 *ppos = kiocb.ki_pos;
2469 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2471 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2472 __u64 flags, struct lov_user_md *lum, int lum_size)
2474 struct lookup_intent oit = {
2476 .it_flags = flags | MDS_OPEN_BY_FID,
2481 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2482 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2483 /* this code will only exist for big-endian systems */
2484 lustre_swab_lov_user_md(lum, 0);
2487 ll_inode_size_lock(inode);
2488 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2490 GOTO(out_unlock, rc);
2492 ll_release_openhandle(dentry, &oit);
2495 ll_inode_size_unlock(inode);
2496 ll_intent_release(&oit);
2501 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2502 struct lov_mds_md **lmmp, int *lmm_size,
2503 struct ptlrpc_request **request)
2505 struct ll_sb_info *sbi = ll_i2sbi(inode);
2506 struct mdt_body *body;
2507 struct lov_mds_md *lmm = NULL;
2508 struct ptlrpc_request *req = NULL;
2509 struct md_op_data *op_data;
2514 rc = ll_get_default_mdsize(sbi, &lmmsize);
2518 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2519 strlen(filename), lmmsize,
2520 LUSTRE_OPC_ANY, NULL);
2521 if (IS_ERR(op_data))
2522 RETURN(PTR_ERR(op_data));
2524 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2525 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2526 ll_finish_md_op_data(op_data);
2528 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2533 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2534 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2536 lmmsize = body->mbo_eadatasize;
2538 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2540 GOTO(out, rc = -ENODATA);
2542 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2543 LASSERT(lmm != NULL);
2545 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2546 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2547 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2548 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2549 GOTO(out, rc = -EPROTO);
2552 * This is coming from the MDS, so is probably in
2553 * little endian. We convert it to host endian before
2554 * passing it to userspace.
2556 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2557 int stripe_count = 0;
2559 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2560 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2561 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2562 if (le32_to_cpu(lmm->lmm_pattern) &
2563 LOV_PATTERN_F_RELEASED)
2565 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2567 /* if function called for directory - we should
2568 * avoid swab not existent lsm objects
2570 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2571 S_ISREG(body->mbo_mode))
2572 lustre_swab_lov_user_md_objects(
2573 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2575 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2576 S_ISREG(body->mbo_mode))
2577 lustre_swab_lov_user_md_objects(
2578 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2580 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2581 lustre_swab_lov_comp_md_v1(
2582 (struct lov_comp_md_v1 *)lmm);
2586 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2587 struct lov_comp_md_v1 *comp_v1 = NULL;
2588 struct lov_comp_md_entry_v1 *ent;
2589 struct lov_user_md_v1 *v1 = NULL;
2593 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2594 /* Dump the striping information */
2595 for (; i < comp_v1->lcm_entry_count; i++) {
2596 ent = &comp_v1->lcm_entries[i];
2597 off = ent->lcme_offset;
2598 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2600 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2601 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2603 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LOV_COMP_MAGIC) &&
2604 (cfs_fail_val == i + 1)))
2605 v1->lmm_magic = LOV_MAGIC_BAD;
2607 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LOV_COMP_PATTERN) &&
2608 (cfs_fail_val == i + 1)))
2609 v1->lmm_pattern = LOV_PATTERN_BAD;
2613 GOTO(out, rc = -EINVAL);
2615 lmm->lmm_stripe_count = v1->lmm_stripe_count;
2616 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2618 * Return valid stripe_count and stripe_size instead of 0 for
2619 * DoM files to avoid divide-by-zero for older userspace that
2620 * calls this ioctl, e.g. lustre ADIO driver.
2622 if (lmm->lmm_stripe_count == 0)
2623 lmm->lmm_stripe_count = 1;
2624 if (lmm->lmm_stripe_size == 0) {
2625 /* Since the first component of the file data is placed
2626 * on the MDT for faster access, the stripe_size of the
2627 * second one is always that applications which are
2630 if (lmm->lmm_pattern & LOV_PATTERN_MDT)
2631 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2633 i = comp_v1->lcm_entry_count > 1 ?
2634 comp_v1->lcm_entry_count - 1 : 0;
2635 ent = &comp_v1->lcm_entries[i];
2636 off = ent->lcme_offset;
2637 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2638 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2643 *lmm_size = lmmsize;
2648 static int ll_lov_setea(struct inode *inode, struct file *file,
2651 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2652 struct lov_user_md *lump;
2653 int lum_size = sizeof(*lump) + sizeof(struct lov_user_ost_data);
2657 if (!capable(CAP_SYS_ADMIN))
2660 OBD_ALLOC_LARGE(lump, lum_size);
2664 if (copy_from_user(lump, arg, lum_size))
2665 GOTO(out_lump, rc = -EFAULT);
2667 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2669 cl_lov_delay_create_clear(&file->f_flags);
2672 OBD_FREE_LARGE(lump, lum_size);
2676 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2683 /* exit before doing any work if pointer is bad */
2684 if (unlikely(!ll_access_ok(lum, sizeof(struct lov_user_md))))
2687 env = cl_env_get(&refcheck);
2689 RETURN(PTR_ERR(env));
2691 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2692 cl_env_put(env, &refcheck);
2696 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2699 struct lov_user_md __user *lum = arg;
2700 struct lov_user_md *klum;
2702 __u64 flags = FMODE_WRITE;
2705 rc = ll_copy_user_md(lum, &klum);
2710 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2715 rc = put_user(0, &lum->lmm_stripe_count);
2719 rc = ll_layout_refresh(inode, &gen);
2723 rc = ll_file_getstripe(inode, arg, lum_size);
2724 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2725 ll_i2info(inode)->lli_clob) {
2726 struct iattr attr = { 0 };
2728 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2729 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2732 cl_lov_delay_create_clear(&file->f_flags);
2735 OBD_FREE_LARGE(klum, lum_size);
2741 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2743 struct ll_inode_info *lli = ll_i2info(inode);
2744 struct cl_object *obj = lli->lli_clob;
2745 struct ll_file_data *fd = file->private_data;
2746 struct ll_grouplock grouplock;
2751 CWARN("group id for group lock must not be 0\n");
2755 if (ll_file_nolock(file))
2756 RETURN(-EOPNOTSUPP);
2758 if (file->f_flags & O_NONBLOCK) {
2759 if (!mutex_trylock(&lli->lli_group_mutex))
2762 mutex_lock(&lli->lli_group_mutex);
2765 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2766 CWARN("group lock already existed with gid %lu\n",
2767 fd->fd_grouplock.lg_gid);
2768 GOTO(out, rc = -EINVAL);
2770 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2771 if (file->f_flags & O_NONBLOCK)
2772 GOTO(out, rc = -EAGAIN);
2773 mutex_unlock(&lli->lli_group_mutex);
2774 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2775 GOTO(retry, rc = 0);
2777 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2780 * XXX: group lock needs to protect all OST objects while PFL
2781 * can add new OST objects during the IO, so we'd instantiate
2782 * all OST objects before getting its group lock.
2787 struct cl_layout cl = {
2788 .cl_is_composite = false,
2790 struct lu_extent ext = {
2792 .e_end = OBD_OBJECT_EOF,
2795 env = cl_env_get(&refcheck);
2797 GOTO(out, rc = PTR_ERR(env));
2799 rc = cl_object_layout_get(env, obj, &cl);
2800 if (rc >= 0 && cl.cl_is_composite)
2801 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2804 cl_env_put(env, &refcheck);
2809 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2810 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2815 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2816 fd->fd_grouplock = grouplock;
2817 if (lli->lli_group_users == 0)
2818 lli->lli_group_gid = grouplock.lg_gid;
2819 lli->lli_group_users++;
2821 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2823 mutex_unlock(&lli->lli_group_mutex);
2828 static int ll_put_grouplock(struct inode *inode, struct file *file,
2831 struct ll_inode_info *lli = ll_i2info(inode);
2832 struct ll_file_data *fd = file->private_data;
2833 struct ll_grouplock grouplock;
2837 mutex_lock(&lli->lli_group_mutex);
2838 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2839 CWARN("no group lock held\n");
2840 GOTO(out, rc = -EINVAL);
2843 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2845 if (fd->fd_grouplock.lg_gid != arg) {
2846 CWARN("group lock %lu doesn't match current id %lu\n",
2847 arg, fd->fd_grouplock.lg_gid);
2848 GOTO(out, rc = -EINVAL);
2851 grouplock = fd->fd_grouplock;
2852 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2853 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2855 cl_put_grouplock(&grouplock);
2857 lli->lli_group_users--;
2858 if (lli->lli_group_users == 0) {
2859 lli->lli_group_gid = 0;
2860 wake_up_var(&lli->lli_group_users);
2862 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2865 mutex_unlock(&lli->lli_group_mutex);
2871 * Close inode open handle
2873 * \param dentry [in] dentry which contains the inode
2874 * \param it [in,out] intent which contains open info and result
2877 * \retval <0 failure
2879 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2881 struct inode *inode = dentry->d_inode;
2882 struct obd_client_handle *och;
2888 /* Root ? Do nothing. */
2889 if (is_root_inode(inode))
2892 /* No open handle to close? Move away */
2893 if (!it_disposition(it, DISP_OPEN_OPEN))
2896 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2898 OBD_ALLOC(och, sizeof(*och));
2900 GOTO(out, rc = -ENOMEM);
2902 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2906 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2908 /* this one is in place of ll_file_open */
2909 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2910 ptlrpc_req_finished(it->it_request);
2911 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2917 * Get size for inode for which FIEMAP mapping is requested.
2918 * Make the FIEMAP get_info call and returns the result.
2919 * \param fiemap kernel buffer to hold extens
2920 * \param num_bytes kernel buffer size
2922 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2928 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2931 /* Checks for fiemap flags */
2932 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2933 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2937 /* Check for FIEMAP_FLAG_SYNC */
2938 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2939 rc = filemap_fdatawrite(inode->i_mapping);
2944 env = cl_env_get(&refcheck);
2946 RETURN(PTR_ERR(env));
2948 if (i_size_read(inode) == 0) {
2949 rc = ll_glimpse_size(inode);
2954 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLPROJID;
2955 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2956 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2958 /* If filesize is 0, then there would be no objects for mapping */
2959 if (fmkey.lfik_oa.o_size == 0) {
2960 fiemap->fm_mapped_extents = 0;
2964 fmkey.lfik_fiemap = *fiemap;
2966 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2967 &fmkey, fiemap, &num_bytes);
2969 cl_env_put(env, &refcheck);
2973 static int fid2path_for_enc_file(struct inode *parent, char *gfpath,
2976 struct dentry *de = NULL, *de_parent = d_find_any_alias(parent);
2977 struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
2978 struct llcrypt_str de_name;
2979 char *p, *ptr = gfpath;
2980 size_t len = 0, len_orig = 0;
2981 int enckey = -1, nameenc = -1;
2985 while ((p = strsep(&gfpath, "/")) != NULL) {
2993 len_orig = strlen(p);
2995 rc = sscanf(p, "["SFID"]", RFID(&fid));
2997 p = strchr(p, ']') + 1;
3003 if (!IS_ENCRYPTED(parent)) {
3004 if (gfpathlen < len + 1) {
3009 memmove(ptr, p, len);
3013 gfpathlen -= len + 1;
3017 /* From here, we know parent is encrypted */
3020 rc = llcrypt_prepare_readdir(parent);
3021 if (rc && rc != -ENOKEY) {
3028 if (llcrypt_has_encryption_key(parent))
3034 llcrypt_policy_has_filename_enc(parent);
3037 /* Even if names are not encrypted, we still need to call
3038 * ll_fname_disk_to_usr in order to decode names as they are
3039 * coming from the wire.
3041 rc = llcrypt_fname_alloc_buffer(parent, NAME_MAX + 1, &lltr);
3049 rc = ll_fname_disk_to_usr(parent, 0, 0, &de_name,
3052 llcrypt_fname_free_buffer(&lltr);
3056 lltr.name[lltr.len] = '\0';
3058 if (lltr.len <= len_orig && gfpathlen >= lltr.len + 1) {
3059 memcpy(ptr, lltr.name, lltr.len);
3064 gfpathlen -= lltr.len + 1;
3068 llcrypt_fname_free_buffer(&lltr);
3070 if (rc == -EOVERFLOW) {
3077 /* We reached the end of the string, which means
3078 * we are dealing with the last component in the path.
3079 * So save a useless lookup and exit.
3085 if (enckey == 0 || nameenc == 0)
3088 ll_inode_lock(parent);
3089 de = lookup_one_len(p, de_parent, len);
3090 ll_inode_unlock(parent);
3091 if (IS_ERR_OR_NULL(de) || !de->d_inode) {
3097 parent = de->d_inode;
3104 if (!IS_ERR_OR_NULL(de))
3109 int __ll_fid2path(struct inode *inode, struct getinfo_fid2path *gfout,
3110 size_t outsize, __u32 pathlen_orig)
3112 struct obd_export *exp = ll_i2mdexp(inode);
3115 /* Append root FID after gfout to let MDT know the root FID so that
3116 * it can lookup the correct path, this is mainly for fileset.
3117 * old server without fileset mount support will ignore this.
3119 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
3121 /* Call mdc_iocontrol */
3122 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
3124 if (!rc && gfout->gf_pathlen && gfout->gf_u.gf_path[0] == '/') {
3125 /* by convention, server side (mdt_path_current()) puts
3126 * a leading '/' to tell client that we are dealing with
3129 rc = fid2path_for_enc_file(inode, gfout->gf_u.gf_path,
3131 if (!rc && strlen(gfout->gf_u.gf_path) > pathlen_orig)
3138 int ll_fid2path(struct inode *inode, void __user *arg)
3140 const struct getinfo_fid2path __user *gfin = arg;
3141 __u32 pathlen, pathlen_orig;
3142 struct getinfo_fid2path *gfout;
3148 if (!capable(CAP_DAC_READ_SEARCH) &&
3149 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3152 /* Only need to get the buflen */
3153 if (get_user(pathlen, &gfin->gf_pathlen))
3156 if (pathlen > PATH_MAX)
3158 pathlen_orig = pathlen;
3161 outsize = sizeof(*gfout) + pathlen;
3162 OBD_ALLOC(gfout, outsize);
3166 if (copy_from_user(gfout, arg, sizeof(*gfout)))
3167 GOTO(gf_free, rc = -EFAULT);
3169 gfout->gf_pathlen = pathlen;
3170 rc = __ll_fid2path(inode, gfout, outsize, pathlen_orig);
3174 if (copy_to_user(arg, gfout, sizeof(*gfout) + pathlen_orig))
3178 OBD_FREE(gfout, outsize);
3179 if (rc == -ENAMETOOLONG) {
3180 pathlen += PATH_MAX;
3187 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
3189 struct cl_object *obj = ll_i2info(inode)->lli_clob;
3197 ioc->idv_version = 0;
3198 ioc->idv_layout_version = UINT_MAX;
3200 /* If no file object initialized, we consider its version is 0. */
3204 env = cl_env_get(&refcheck);
3206 RETURN(PTR_ERR(env));
3208 io = vvp_env_thread_io(env);
3210 io->u.ci_data_version.dv_data_version = 0;
3211 io->u.ci_data_version.dv_layout_version = UINT_MAX;
3212 io->u.ci_data_version.dv_flags = ioc->idv_flags;
3215 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
3216 result = cl_io_loop(env, io);
3218 result = io->ci_result;
3220 ioc->idv_version = io->u.ci_data_version.dv_data_version;
3221 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
3222 cl_io_fini(env, io);
3224 if (unlikely(io->ci_need_restart))
3227 cl_env_put(env, &refcheck);
3233 * Read the data_version for inode.
3235 * This value is computed using stripe object version on OST.
3236 * Version is computed using server side locking.
3238 * @param flags if do sync on the OST side;
3240 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
3241 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
3243 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
3245 struct ioc_data_version ioc = { .idv_flags = flags };
3248 rc = ll_ioc_data_version(inode, &ioc);
3250 *data_version = ioc.idv_version;
3256 * Trigger a HSM release request for the provided inode.
3258 int ll_hsm_release(struct inode *inode)
3261 struct obd_client_handle *och = NULL;
3262 __u64 data_version = 0;
3268 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
3269 ll_i2sbi(inode)->ll_fsname,
3270 PFID(&ll_i2info(inode)->lli_fid));
3272 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
3274 GOTO(out, rc = PTR_ERR(och));
3276 /* Grab latest data_version and [am]time values */
3277 rc = ll_data_version(inode, &data_version,
3278 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
3282 env = cl_env_get(&refcheck);
3284 GOTO(out, rc = PTR_ERR(env));
3286 rc = ll_merge_attr(env, inode);
3287 cl_env_put(env, &refcheck);
3289 /* If error happen, we have the wrong size for a file.
3295 /* Release the file.
3296 * NB: lease lock handle is released in mdc_hsm_release_pack() because
3297 * we still need it to pack l_remote_handle to MDT. */
3298 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
3304 if (och != NULL && !IS_ERR(och)) /* close the file */
3305 ll_lease_close(och, inode, NULL);
3310 struct ll_swap_stack {
3313 struct inode *inode1;
3314 struct inode *inode2;
3319 static int ll_swap_layouts(struct file *file1, struct file *file2,
3320 struct lustre_swap_layouts *lsl)
3322 struct mdc_swap_layouts msl;
3323 struct md_op_data *op_data;
3326 struct ll_swap_stack *llss = NULL;
3329 OBD_ALLOC_PTR(llss);
3333 llss->inode1 = file_inode(file1);
3334 llss->inode2 = file_inode(file2);
3336 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
3340 /* we use 2 bool because it is easier to swap than 2 bits */
3341 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
3342 llss->check_dv1 = true;
3344 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
3345 llss->check_dv2 = true;
3347 /* we cannot use lsl->sl_dvX directly because we may swap them */
3348 llss->dv1 = lsl->sl_dv1;
3349 llss->dv2 = lsl->sl_dv2;
3351 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
3352 if (rc == 0) /* same file, done! */
3355 if (rc < 0) { /* sequentialize it */
3356 swap(llss->inode1, llss->inode2);
3358 swap(llss->dv1, llss->dv2);
3359 swap(llss->check_dv1, llss->check_dv2);
3363 if (gid != 0) { /* application asks to flush dirty cache */
3364 rc = ll_get_grouplock(llss->inode1, file1, gid);
3368 rc = ll_get_grouplock(llss->inode2, file2, gid);
3370 ll_put_grouplock(llss->inode1, file1, gid);
3375 /* ultimate check, before swaping the layouts we check if
3376 * dataversion has changed (if requested) */
3377 if (llss->check_dv1) {
3378 rc = ll_data_version(llss->inode1, &dv, 0);
3381 if (dv != llss->dv1)
3382 GOTO(putgl, rc = -EAGAIN);
3385 if (llss->check_dv2) {
3386 rc = ll_data_version(llss->inode2, &dv, 0);
3389 if (dv != llss->dv2)
3390 GOTO(putgl, rc = -EAGAIN);
3393 /* struct md_op_data is used to send the swap args to the mdt
3394 * only flags is missing, so we use struct mdc_swap_layouts
3395 * through the md_op_data->op_data */
3396 /* flags from user space have to be converted before they are send to
3397 * server, no flag is sent today, they are only used on the client */
3400 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3401 0, LUSTRE_OPC_ANY, &msl);
3402 if (IS_ERR(op_data))
3403 GOTO(free, rc = PTR_ERR(op_data));
3405 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3406 sizeof(*op_data), op_data, NULL);
3407 ll_finish_md_op_data(op_data);
3414 ll_put_grouplock(llss->inode2, file2, gid);
3415 ll_put_grouplock(llss->inode1, file1, gid);
3425 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3427 struct obd_export *exp = ll_i2mdexp(inode);
3428 struct md_op_data *op_data;
3432 /* Detect out-of range masks */
3433 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3436 /* Non-root users are forbidden to set or clear flags which are
3437 * NOT defined in HSM_USER_MASK. */
3438 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3439 !capable(CAP_SYS_ADMIN))
3442 if (!exp_connect_archive_id_array(exp)) {
3443 /* Detect out-of range archive id */
3444 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3445 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3449 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3450 LUSTRE_OPC_ANY, hss);
3451 if (IS_ERR(op_data))
3452 RETURN(PTR_ERR(op_data));
3454 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3457 ll_finish_md_op_data(op_data);
3462 static int ll_hsm_import(struct inode *inode, struct file *file,
3463 struct hsm_user_import *hui)
3465 struct hsm_state_set *hss = NULL;
3466 struct iattr *attr = NULL;
3470 if (!S_ISREG(inode->i_mode))
3476 GOTO(out, rc = -ENOMEM);
3478 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3479 hss->hss_archive_id = hui->hui_archive_id;
3480 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3481 rc = ll_hsm_state_set(inode, hss);
3485 OBD_ALLOC_PTR(attr);
3487 GOTO(out, rc = -ENOMEM);
3489 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3490 attr->ia_mode |= S_IFREG;
3491 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3492 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3493 attr->ia_size = hui->hui_size;
3494 attr->ia_mtime.tv_sec = hui->hui_mtime;
3495 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3496 attr->ia_atime.tv_sec = hui->hui_atime;
3497 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3499 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3500 ATTR_UID | ATTR_GID |
3501 ATTR_MTIME | ATTR_MTIME_SET |
3502 ATTR_ATIME | ATTR_ATIME_SET;
3505 /* inode lock owner set in ll_setattr_raw()*/
3506 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3509 inode_unlock(inode);
3521 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3523 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3524 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3527 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3529 struct inode *inode = file_inode(file);
3531 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3532 ATTR_MTIME | ATTR_MTIME_SET |
3535 .tv_sec = lfu->lfu_atime_sec,
3536 .tv_nsec = lfu->lfu_atime_nsec,
3539 .tv_sec = lfu->lfu_mtime_sec,
3540 .tv_nsec = lfu->lfu_mtime_nsec,
3543 .tv_sec = lfu->lfu_ctime_sec,
3544 .tv_nsec = lfu->lfu_ctime_nsec,
3550 if (!capable(CAP_SYS_ADMIN))
3553 if (!S_ISREG(inode->i_mode))
3557 /* inode lock owner set in ll_setattr_raw()*/
3558 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3560 inode_unlock(inode);
3565 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3568 case MODE_READ_USER:
3570 case MODE_WRITE_USER:
3577 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3579 /* Used to allow the upper layers of the client to request an LDLM lock
3580 * without doing an actual read or write.
3582 * Used for ladvise lockahead to manually request specific locks.
3584 * \param[in] file file this ladvise lock request is on
3585 * \param[in] ladvise ladvise struct describing this lock request
3587 * \retval 0 success, no detailed result available (sync requests
3588 * and requests sent to the server [not handled locally]
3589 * cannot return detailed results)
3590 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3591 * see definitions for details.
3592 * \retval negative negative errno on error
3594 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3596 struct lu_env *env = NULL;
3597 struct cl_io *io = NULL;
3598 struct cl_lock *lock = NULL;
3599 struct cl_lock_descr *descr = NULL;
3600 struct dentry *dentry = file->f_path.dentry;
3601 struct inode *inode = dentry->d_inode;
3602 enum cl_lock_mode cl_mode;
3603 off_t start = ladvise->lla_start;
3604 off_t end = ladvise->lla_end;
3611 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3612 dentry, dentry->d_inode,
3613 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3616 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3618 GOTO(out, result = cl_mode);
3620 /* Get IO environment */
3621 result = cl_io_get(inode, &env, &io, &refcheck);
3625 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3628 * nothing to do for this io. This currently happens when
3629 * stripe sub-object's are not yet created.
3631 result = io->ci_result;
3632 } else if (result == 0) {
3633 lock = vvp_env_lock(env);
3634 descr = &lock->cll_descr;
3636 descr->cld_obj = io->ci_obj;
3637 /* Convert byte offsets to pages */
3638 descr->cld_start = start >> PAGE_SHIFT;
3639 descr->cld_end = end >> PAGE_SHIFT;
3640 descr->cld_mode = cl_mode;
3641 /* CEF_MUST is used because we do not want to convert a
3642 * lockahead request to a lockless lock */
3643 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3645 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3646 descr->cld_enq_flags |= CEF_SPECULATIVE;
3648 result = cl_lock_request(env, io, lock);
3650 /* On success, we need to release the lock */
3652 cl_lock_release(env, lock);
3654 cl_io_fini(env, io);
3655 cl_env_put(env, &refcheck);
3657 /* -ECANCELED indicates a matching lock with a different extent
3658 * was already present, and -EEXIST indicates a matching lock
3659 * on exactly the same extent was already present.
3660 * We convert them to positive values for userspace to make
3661 * recognizing true errors easier.
3662 * Note we can only return these detailed results on async requests,
3663 * as sync requests look the same as i/o requests for locking. */
3664 if (result == -ECANCELED)
3665 result = LLA_RESULT_DIFFERENT;
3666 else if (result == -EEXIST)
3667 result = LLA_RESULT_SAME;
3672 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3674 static int ll_ladvise_sanity(struct inode *inode,
3675 struct llapi_lu_ladvise *ladvise)
3677 struct ll_sb_info *sbi = ll_i2sbi(inode);
3678 enum lu_ladvise_type advice = ladvise->lla_advice;
3679 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3680 * be in the first 32 bits of enum ladvise_flags */
3681 __u32 flags = ladvise->lla_peradvice_flags;
3682 /* 3 lines at 80 characters per line, should be plenty */
3685 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3688 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3689 sbi->ll_fsname, advice,
3690 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3694 /* Per-advice checks */
3696 case LU_LADVISE_LOCKNOEXPAND:
3697 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3699 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3700 "rc = %d\n", sbi->ll_fsname, flags,
3701 ladvise_names[advice], rc);
3705 case LU_LADVISE_LOCKAHEAD:
3706 /* Currently only READ and WRITE modes can be requested */
3707 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3708 ladvise->lla_lockahead_mode == 0) {
3710 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3711 "rc = %d\n", sbi->ll_fsname,
3712 ladvise->lla_lockahead_mode,
3713 ladvise_names[advice], rc);
3717 case LU_LADVISE_WILLREAD:
3718 case LU_LADVISE_DONTNEED:
3720 /* Note fall through above - These checks apply to all advices
3721 * except LOCKNOEXPAND */
3722 if (flags & ~LF_DEFAULT_MASK) {
3724 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3725 "rc = %d\n", sbi->ll_fsname, flags,
3726 ladvise_names[advice], rc);
3729 if (ladvise->lla_start >= ladvise->lla_end) {
3731 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3732 "for %s: rc = %d\n", sbi->ll_fsname,
3733 ladvise->lla_start, ladvise->lla_end,
3734 ladvise_names[advice], rc);
3746 * Give file access advices
3748 * The ladvise interface is similar to Linux fadvise() system call, except it
3749 * forwards the advices directly from Lustre client to server. The server side
3750 * codes will apply appropriate read-ahead and caching techniques for the
3751 * corresponding files.
3753 * A typical workload for ladvise is e.g. a bunch of different clients are
3754 * doing small random reads of a file, so prefetching pages into OSS cache
3755 * with big linear reads before the random IO is a net benefit. Fetching
3756 * all that data into each client cache with fadvise() may not be, due to
3757 * much more data being sent to the client.
3759 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3760 struct llapi_lu_ladvise *ladvise)
3764 struct cl_ladvise_io *lio;
3769 env = cl_env_get(&refcheck);
3771 RETURN(PTR_ERR(env));
3773 io = vvp_env_thread_io(env);
3774 io->ci_obj = ll_i2info(inode)->lli_clob;
3776 /* initialize parameters for ladvise */
3777 lio = &io->u.ci_ladvise;
3778 lio->li_start = ladvise->lla_start;
3779 lio->li_end = ladvise->lla_end;
3780 lio->li_fid = ll_inode2fid(inode);
3781 lio->li_advice = ladvise->lla_advice;
3782 lio->li_flags = flags;
3784 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3785 rc = cl_io_loop(env, io);
3789 cl_io_fini(env, io);
3790 cl_env_put(env, &refcheck);
3794 static int ll_lock_noexpand(struct file *file, int flags)
3796 struct ll_file_data *fd = file->private_data;
3798 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3803 #ifndef HAVE_FILEATTR_GET
3804 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3807 struct fsxattr fsxattr;
3809 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3812 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3813 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3814 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3815 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3816 if (copy_to_user(uarg, &fsxattr, sizeof(fsxattr)))
3823 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3827 * Project Quota ID state is only allowed to change from within the init
3828 * namespace. Enforce that restriction only if we are trying to change
3829 * the quota ID state. Everything else is allowed in user namespaces.
3831 if (current_user_ns() == &init_user_ns) {
3833 * Caller is allowed to change the project ID. if it is being
3834 * changed, make sure that the new value is valid.
3836 if (ll_i2info(inode)->lli_projid != projid &&
3837 !projid_valid(make_kprojid(&init_user_ns, projid)))
3843 if (ll_i2info(inode)->lli_projid != projid)
3846 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3847 if (!(xflags & FS_XFLAG_PROJINHERIT))
3850 if (xflags & FS_XFLAG_PROJINHERIT)
3857 int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3859 struct ptlrpc_request *req = NULL;
3860 struct md_op_data *op_data;
3861 struct cl_object *obj;
3862 unsigned int inode_flags;
3865 CDEBUG(D_QUOTA, DFID" xflags=%x projid=%u\n",
3866 PFID(ll_inode2fid(inode)), xflags, projid);
3867 rc = ll_ioctl_check_project(inode, xflags, projid);
3871 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3872 LUSTRE_OPC_ANY, NULL);
3873 if (IS_ERR(op_data))
3874 RETURN(PTR_ERR(op_data));
3876 inode_flags = ll_xflags_to_inode_flags(xflags);
3877 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3878 if (xflags & FS_XFLAG_PROJINHERIT)
3879 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3881 /* pass projid to md_op_data */
3882 op_data->op_projid = projid;
3884 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3885 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3886 ptlrpc_req_finished(req);
3888 GOTO(out_fsxattr, rc);
3889 ll_update_inode_flags(inode, op_data->op_attr_flags);
3891 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3892 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3893 GOTO(out_fsxattr, rc);
3895 obj = ll_i2info(inode)->lli_clob;
3897 struct iattr attr = { 0 };
3899 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3903 ll_finish_md_op_data(op_data);
3907 #ifndef HAVE_FILEATTR_GET
3908 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3911 struct fsxattr fsxattr;
3915 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3918 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3919 fsxattr.fsx_projid));
3923 int ll_ioctl_project(struct file *file, unsigned int cmd, void __user *uarg)
3925 struct lu_project lu_project;
3926 struct dentry *dentry = file_dentry(file);
3927 struct inode *inode = file_inode(file);
3928 struct dentry *child_dentry = NULL;
3929 int rc = 0, name_len;
3931 if (copy_from_user(&lu_project, uarg, sizeof(lu_project)))
3934 /* apply child dentry if name is valid */
3935 name_len = strnlen(lu_project.project_name, NAME_MAX);
3936 if (name_len > 0 && name_len <= NAME_MAX) {
3937 ll_inode_lock(inode);
3938 child_dentry = lookup_one_len(lu_project.project_name,
3940 ll_inode_unlock(inode);
3941 if (IS_ERR(child_dentry)) {
3942 rc = PTR_ERR(child_dentry);
3945 inode = child_dentry->d_inode;
3950 } else if (name_len > NAME_MAX) {
3955 switch (lu_project.project_type) {
3956 case LU_PROJECT_SET:
3957 rc = ll_set_project(inode, lu_project.project_xflags,
3958 lu_project.project_id);
3960 case LU_PROJECT_GET:
3961 lu_project.project_xflags =
3962 ll_inode_flags_to_xflags(inode->i_flags);
3963 if (test_bit(LLIF_PROJECT_INHERIT,
3964 &ll_i2info(inode)->lli_flags))
3965 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3966 lu_project.project_id = ll_i2info(inode)->lli_projid;
3967 if (copy_to_user(uarg, &lu_project, sizeof(lu_project))) {
3977 if (!IS_ERR_OR_NULL(child_dentry))
3982 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3985 struct inode *inode = file_inode(file);
3986 struct ll_file_data *fd = file->private_data;
3987 struct ll_inode_info *lli = ll_i2info(inode);
3988 struct obd_client_handle *och = NULL;
3989 struct split_param sp;
3990 struct pcc_param param;
3991 bool lease_broken = false;
3993 enum mds_op_bias bias = 0;
3995 struct file *layout_file = NULL;
3997 size_t data_size = 0;
3998 bool attached = false;
4003 mutex_lock(&lli->lli_och_mutex);
4004 if (fd->fd_lease_och != NULL) {
4005 och = fd->fd_lease_och;
4006 fd->fd_lease_och = NULL;
4008 mutex_unlock(&lli->lli_och_mutex);
4013 fmode = och->och_flags;
4015 switch (ioc->lil_flags) {
4016 case LL_LEASE_RESYNC_DONE:
4017 if (ioc->lil_count > IOC_IDS_MAX)
4018 GOTO(out_lease_close, rc = -EINVAL);
4020 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
4021 OBD_ALLOC(data, data_size);
4023 GOTO(out_lease_close, rc = -ENOMEM);
4025 if (copy_from_user(data, uarg, data_size))
4026 GOTO(out_lease_close, rc = -EFAULT);
4028 bias = MDS_CLOSE_RESYNC_DONE;
4030 case LL_LEASE_LAYOUT_MERGE:
4031 if (ioc->lil_count != 1)
4032 GOTO(out_lease_close, rc = -EINVAL);
4034 uarg += sizeof(*ioc);
4035 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4036 GOTO(out_lease_close, rc = -EFAULT);
4038 layout_file = fget(fdv);
4040 GOTO(out_lease_close, rc = -EBADF);
4042 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
4043 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
4044 GOTO(out_lease_close, rc = -EPERM);
4046 data = file_inode(layout_file);
4047 bias = MDS_CLOSE_LAYOUT_MERGE;
4049 case LL_LEASE_LAYOUT_SPLIT: {
4052 if (ioc->lil_count != 2)
4053 GOTO(out_lease_close, rc = -EINVAL);
4055 uarg += sizeof(*ioc);
4056 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4057 GOTO(out_lease_close, rc = -EFAULT);
4059 uarg += sizeof(fdv);
4060 if (copy_from_user(&mirror_id, uarg, sizeof(mirror_id)))
4061 GOTO(out_lease_close, rc = -EFAULT);
4062 if (mirror_id >= MIRROR_ID_NEG)
4063 GOTO(out_lease_close, rc = -EINVAL);
4065 layout_file = fget(fdv);
4067 GOTO(out_lease_close, rc = -EBADF);
4069 /* if layout_file == file, it means to destroy the mirror */
4070 sp.sp_inode = file_inode(layout_file);
4071 sp.sp_mirror_id = (__u16)mirror_id;
4073 bias = MDS_CLOSE_LAYOUT_SPLIT;
4076 case LL_LEASE_PCC_ATTACH:
4077 if (ioc->lil_count != 1)
4080 if (IS_ENCRYPTED(inode))
4081 RETURN(-EOPNOTSUPP);
4083 uarg += sizeof(*ioc);
4084 if (copy_from_user(¶m.pa_archive_id, uarg, sizeof(__u32)))
4085 GOTO(out_lease_close, rc2 = -EFAULT);
4087 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
4089 GOTO(out_lease_close, rc2);
4092 /* Grab latest data version */
4093 rc2 = ll_data_version(inode, ¶m.pa_data_version,
4096 GOTO(out_lease_close, rc2);
4099 bias = MDS_PCC_ATTACH;
4102 /* without close intent */
4107 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
4111 rc = ll_lease_och_release(inode, file);
4120 if (ioc->lil_flags == LL_LEASE_RESYNC_DONE && data)
4121 OBD_FREE(data, data_size);
4126 if (ioc->lil_flags == LL_LEASE_PCC_ATTACH) {
4129 rc = pcc_readwrite_attach_fini(file, inode,
4130 param.pa_layout_gen,
4135 ll_layout_refresh(inode, &fd->fd_layout_version);
4138 rc = ll_lease_type_from_fmode(fmode);
4142 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
4145 struct inode *inode = file_inode(file);
4146 struct ll_inode_info *lli = ll_i2info(inode);
4147 struct ll_file_data *fd = file->private_data;
4148 struct obd_client_handle *och = NULL;
4149 __u64 open_flags = 0;
4155 switch (ioc->lil_mode) {
4156 case LL_LEASE_WRLCK:
4157 if (!(file->f_mode & FMODE_WRITE))
4159 fmode = FMODE_WRITE;
4161 case LL_LEASE_RDLCK:
4162 if (!(file->f_mode & FMODE_READ))
4166 case LL_LEASE_UNLCK:
4167 RETURN(ll_file_unlock_lease(file, ioc, uarg));
4172 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
4174 /* apply for lease */
4175 if (ioc->lil_flags & LL_LEASE_RESYNC)
4176 open_flags = MDS_OPEN_RESYNC;
4177 och = ll_lease_open(inode, file, fmode, open_flags);
4179 RETURN(PTR_ERR(och));
4181 if (ioc->lil_flags & LL_LEASE_RESYNC) {
4182 rc = ll_lease_file_resync(och, inode, uarg);
4184 ll_lease_close(och, inode, NULL);
4187 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
4189 ll_lease_close(och, inode, NULL);
4195 mutex_lock(&lli->lli_och_mutex);
4196 if (fd->fd_lease_och == NULL) {
4197 fd->fd_lease_och = och;
4200 mutex_unlock(&lli->lli_och_mutex);
4202 /* impossible now that only excl is supported for now */
4203 ll_lease_close(och, inode, &lease_broken);
4209 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
4211 struct ll_inode_info *lli = ll_i2info(inode);
4212 struct ll_sb_info *sbi = ll_i2sbi(inode);
4213 __u64 now = ktime_get_real_seconds();
4216 spin_lock(&lli->lli_heat_lock);
4217 heat->lh_flags = lli->lli_heat_flags;
4218 for (i = 0; i < heat->lh_count; i++)
4219 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
4220 now, sbi->ll_heat_decay_weight,
4221 sbi->ll_heat_period_second);
4222 spin_unlock(&lli->lli_heat_lock);
4225 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
4227 struct ll_inode_info *lli = ll_i2info(inode);
4230 spin_lock(&lli->lli_heat_lock);
4231 if (flags & LU_HEAT_FLAG_CLEAR)
4232 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
4234 if (flags & LU_HEAT_FLAG_OFF)
4235 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
4237 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
4239 spin_unlock(&lli->lli_heat_lock);
4245 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4247 struct inode *inode = file_inode(file);
4248 struct ll_file_data *fd = file->private_data;
4249 void __user *uarg = (void __user *)arg;
4253 CDEBUG(D_VFSTRACE|D_IOCTL, "VFS Op:inode="DFID"(%pK) cmd=%x arg=%lx\n",
4254 PFID(ll_inode2fid(inode)), inode, cmd, arg);
4255 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
4257 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
4258 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
4261 /* can't do a generic karg == NULL check here, since it is too noisy and
4262 * we need to return -ENOTTY for unsupported ioctls instead of -EINVAL.
4265 case LL_IOC_GETFLAGS:
4266 /* Get the current value of the file flags */
4267 return put_user(fd->fd_flags, (int __user *)arg);
4268 case LL_IOC_SETFLAGS:
4269 case LL_IOC_CLRFLAGS:
4270 /* Set or clear specific file flags */
4271 /* XXX This probably needs checks to ensure the flags are
4272 * not abused, and to handle any flag side effects.
4274 if (get_user(flags, (int __user *)arg))
4277 if (cmd == LL_IOC_SETFLAGS) {
4278 if ((flags & LL_FILE_IGNORE_LOCK) &&
4279 !(file->f_flags & O_DIRECT)) {
4281 CERROR("%s: unable to disable locking on non-O_DIRECT file "DFID": rc = %d\n",
4282 current->comm, PFID(ll_inode2fid(inode)),
4287 fd->fd_flags |= flags;
4289 fd->fd_flags &= ~flags;
4292 case LL_IOC_LOV_SETSTRIPE:
4293 case LL_IOC_LOV_SETSTRIPE_NEW:
4294 RETURN(ll_lov_setstripe(inode, file, uarg));
4295 case LL_IOC_LOV_SETEA:
4296 RETURN(ll_lov_setea(inode, file, uarg));
4297 case LL_IOC_LOV_SWAP_LAYOUTS: {
4299 struct lustre_swap_layouts lsl;
4301 if (copy_from_user(&lsl, uarg, sizeof(lsl)))
4304 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
4307 file2 = fget(lsl.sl_fd);
4311 /* O_WRONLY or O_RDWR */
4312 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
4313 GOTO(out, rc = -EPERM);
4315 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
4316 struct obd_client_handle *och = NULL;
4317 struct ll_inode_info *lli;
4318 struct inode *inode2;
4320 lli = ll_i2info(inode);
4321 mutex_lock(&lli->lli_och_mutex);
4322 if (fd->fd_lease_och != NULL) {
4323 och = fd->fd_lease_och;
4324 fd->fd_lease_och = NULL;
4326 mutex_unlock(&lli->lli_och_mutex);
4328 GOTO(out, rc = -ENOLCK);
4329 inode2 = file_inode(file2);
4330 rc = ll_swap_layouts_close(och, inode, inode2);
4332 rc = ll_swap_layouts(file, file2, &lsl);
4338 case LL_IOC_LOV_GETSTRIPE:
4339 case LL_IOC_LOV_GETSTRIPE_NEW:
4340 RETURN(ll_file_getstripe(inode, uarg, 0));
4341 case LL_IOC_GROUP_LOCK:
4342 RETURN(ll_get_grouplock(inode, file, arg));
4343 case LL_IOC_GROUP_UNLOCK:
4344 RETURN(ll_put_grouplock(inode, file, arg));
4345 case LL_IOC_DATA_VERSION: {
4346 struct ioc_data_version idv;
4349 if (copy_from_user(&idv, uarg, sizeof(idv)))
4352 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4353 rc = ll_ioc_data_version(inode, &idv);
4355 if (rc == 0 && copy_to_user(uarg, &idv, sizeof(idv)))
4360 case LL_IOC_HSM_STATE_GET: {
4361 struct md_op_data *op_data;
4362 struct hsm_user_state *hus;
4365 if (!ll_access_ok(uarg, sizeof(*hus)))
4372 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4373 LUSTRE_OPC_ANY, hus);
4374 if (IS_ERR(op_data)) {
4375 rc = PTR_ERR(op_data);
4377 rc = obd_iocontrol(cmd, ll_i2mdexp(inode),
4378 sizeof(*op_data), op_data, NULL);
4380 if (copy_to_user(uarg, hus, sizeof(*hus)))
4383 ll_finish_md_op_data(op_data);
4388 case LL_IOC_HSM_STATE_SET: {
4389 struct hsm_state_set *hss;
4396 if (copy_from_user(hss, uarg, sizeof(*hss)))
4399 rc = ll_hsm_state_set(inode, hss);
4404 case LL_IOC_HSM_ACTION: {
4405 struct md_op_data *op_data;
4406 struct hsm_current_action *hca;
4410 if (!ll_access_ok(uarg, sizeof(*hca)))
4417 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4418 LUSTRE_OPC_ANY, hca);
4419 if (IS_ERR(op_data)) {
4421 RETURN(PTR_ERR(op_data));
4424 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4427 GOTO(skip_copy, rc);
4429 /* The hsm_current_action retreived from the server could
4430 * contain corrupt information. If it is incorrect data collect
4431 * debug information. We still send the data even if incorrect
4432 * to user land to handle.
4434 action = hsm_user_action2name(hca->hca_action);
4435 if (strcmp(action, "UNKNOWN") == 0 ||
4436 hca->hca_state > HPS_DONE) {
4438 "HSM current state %s action %s, offset = %llu, length %llu\n",
4439 hsm_progress_state2name(hca->hca_state), action,
4440 hca->hca_location.offset, hca->hca_location.length);
4443 if (copy_to_user(uarg, hca, sizeof(*hca)))
4446 ll_finish_md_op_data(op_data);
4450 case LL_IOC_SET_LEASE_OLD: {
4451 struct ll_ioc_lease ioc = { .lil_mode = arg };
4453 RETURN(ll_file_set_lease(file, &ioc, 0));
4455 case LL_IOC_SET_LEASE: {
4456 struct ll_ioc_lease ioc;
4458 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
4461 RETURN(ll_file_set_lease(file, &ioc, uarg));
4463 case LL_IOC_GET_LEASE: {
4464 struct ll_inode_info *lli = ll_i2info(inode);
4465 struct ldlm_lock *lock = NULL;
4468 mutex_lock(&lli->lli_och_mutex);
4469 if (fd->fd_lease_och != NULL) {
4470 struct obd_client_handle *och = fd->fd_lease_och;
4472 lock = ldlm_handle2lock(&och->och_lease_handle);
4474 lock_res_and_lock(lock);
4475 if (!ldlm_is_cancel(lock))
4476 fmode = och->och_flags;
4478 unlock_res_and_lock(lock);
4479 LDLM_LOCK_PUT(lock);
4482 mutex_unlock(&lli->lli_och_mutex);
4484 RETURN(ll_lease_type_from_fmode(fmode));
4486 case LL_IOC_HSM_IMPORT: {
4487 struct hsm_user_import *hui;
4493 if (copy_from_user(hui, uarg, sizeof(*hui)))
4496 rc = ll_hsm_import(inode, file, hui);
4501 case LL_IOC_FUTIMES_3: {
4502 struct ll_futimes_3 lfu;
4504 if (copy_from_user(&lfu, uarg, sizeof(lfu)))
4507 RETURN(ll_file_futimes_3(file, &lfu));
4509 case LL_IOC_LADVISE: {
4510 struct llapi_ladvise_hdr *k_ladvise_hdr;
4511 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4514 int alloc_size = sizeof(*k_ladvise_hdr);
4517 u_ladvise_hdr = uarg;
4518 OBD_ALLOC_PTR(k_ladvise_hdr);
4519 if (k_ladvise_hdr == NULL)
4522 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4523 GOTO(out_ladvise, rc = -EFAULT);
4525 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4526 k_ladvise_hdr->lah_count < 1)
4527 GOTO(out_ladvise, rc = -EINVAL);
4529 num_advise = k_ladvise_hdr->lah_count;
4530 if (num_advise >= LAH_COUNT_MAX)
4531 GOTO(out_ladvise, rc = -EFBIG);
4533 OBD_FREE_PTR(k_ladvise_hdr);
4534 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4535 lah_advise[num_advise]);
4536 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4537 if (k_ladvise_hdr == NULL)
4541 * TODO: submit multiple advices to one server in a single RPC
4543 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4544 GOTO(out_ladvise, rc = -EFAULT);
4546 for (i = 0; i < num_advise; i++) {
4547 struct llapi_lu_ladvise *k_ladvise =
4548 &k_ladvise_hdr->lah_advise[i];
4549 struct llapi_lu_ladvise __user *u_ladvise =
4550 &u_ladvise_hdr->lah_advise[i];
4552 rc = ll_ladvise_sanity(inode, k_ladvise);
4554 GOTO(out_ladvise, rc);
4556 switch (k_ladvise->lla_advice) {
4557 case LU_LADVISE_LOCKNOEXPAND:
4558 rc = ll_lock_noexpand(file,
4559 k_ladvise->lla_peradvice_flags);
4560 GOTO(out_ladvise, rc);
4561 case LU_LADVISE_LOCKAHEAD:
4563 rc = ll_file_lock_ahead(file, k_ladvise);
4566 GOTO(out_ladvise, rc);
4569 &u_ladvise->lla_lockahead_result))
4570 GOTO(out_ladvise, rc = -EFAULT);
4573 rc = ll_ladvise(inode, file,
4574 k_ladvise_hdr->lah_flags,
4577 GOTO(out_ladvise, rc);
4584 OBD_FREE(k_ladvise_hdr, alloc_size);
4587 case LL_IOC_FLR_SET_MIRROR: {
4588 /* mirror I/O must be direct to avoid polluting page cache
4590 if (!(file->f_flags & O_DIRECT))
4593 fd->fd_designated_mirror = arg;
4596 case LL_IOC_HEAT_GET: {
4597 struct lu_heat uheat;
4598 struct lu_heat *heat;
4601 if (copy_from_user(&uheat, uarg, sizeof(uheat)))
4604 if (uheat.lh_count > OBD_HEAT_COUNT)
4605 uheat.lh_count = OBD_HEAT_COUNT;
4607 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4608 OBD_ALLOC(heat, size);
4612 heat->lh_count = uheat.lh_count;
4613 ll_heat_get(inode, heat);
4614 rc = copy_to_user(uarg, heat, size);
4615 OBD_FREE(heat, size);
4616 RETURN(rc ? -EFAULT : 0);
4618 case LL_IOC_HEAT_SET: {
4621 if (copy_from_user(&flags, uarg, sizeof(flags)))
4624 rc = ll_heat_set(inode, flags);
4627 case LL_IOC_PCC_DETACH: {
4628 struct lu_pcc_detach *detach;
4630 OBD_ALLOC_PTR(detach);
4634 if (copy_from_user(detach, uarg, sizeof(*detach)))
4635 GOTO(out_detach_free, rc = -EFAULT);
4637 if (!S_ISREG(inode->i_mode))
4638 GOTO(out_detach_free, rc = -EINVAL);
4640 if (!inode_owner_or_capable(&nop_mnt_idmap, inode))
4641 GOTO(out_detach_free, rc = -EPERM);
4643 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4645 OBD_FREE_PTR(detach);
4648 case LL_IOC_PCC_STATE: {
4649 struct lu_pcc_state __user *ustate = uarg;
4650 struct lu_pcc_state *state;
4652 OBD_ALLOC_PTR(state);
4656 if (copy_from_user(state, ustate, sizeof(*state)))
4657 GOTO(out_state, rc = -EFAULT);
4659 rc = pcc_ioctl_state(file, inode, state);
4661 GOTO(out_state, rc);
4663 if (copy_to_user(ustate, state, sizeof(*state)))
4664 GOTO(out_state, rc = -EFAULT);
4667 OBD_FREE_PTR(state);
4671 rc = ll_iocontrol(inode, file, cmd, uarg);
4674 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, uarg));
4678 static loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4680 struct inode *inode = file_inode(file);
4683 struct cl_lseek_io *lsio;
4690 env = cl_env_get(&refcheck);
4692 RETURN(PTR_ERR(env));
4694 io = vvp_env_thread_io(env);
4695 io->ci_obj = ll_i2info(inode)->lli_clob;
4696 ll_io_set_mirror(io, file);
4698 lsio = &io->u.ci_lseek;
4699 lsio->ls_start = offset;
4700 lsio->ls_whence = whence;
4701 lsio->ls_result = -ENXIO;
4704 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4706 struct vvp_io *vio = vvp_env_io(env);
4708 vio->vui_fd = file->private_data;
4709 rc = cl_io_loop(env, io);
4713 retval = rc ? : lsio->ls_result;
4714 cl_io_fini(env, io);
4715 } while (unlikely(io->ci_need_restart));
4717 cl_env_put(env, &refcheck);
4719 /* Without the key, SEEK_HOLE return value has to be
4720 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4722 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4723 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4728 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4730 struct inode *inode = file_inode(file);
4731 loff_t retval = offset, eof = 0;
4732 ktime_t kstart = ktime_get();
4736 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4737 PFID(ll_inode2fid(inode)), inode, retval, retval,
4740 if (origin == SEEK_END) {
4741 retval = ll_glimpse_size(inode);
4744 eof = i_size_read(inode);
4747 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4751 /* flush local cache first if any */
4752 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4755 retval = ll_lseek(file, offset, origin);
4758 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4760 retval = generic_file_llseek_size(file, offset, origin,
4761 ll_file_maxbytes(inode), eof);
4764 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4765 ktime_us_delta(ktime_get(), kstart));
4769 static int ll_flush(struct file *file, fl_owner_t id)
4771 struct inode *inode = file_inode(file);
4772 struct ll_inode_info *lli = ll_i2info(inode);
4773 struct ll_file_data *fd = file->private_data;
4776 LASSERT(!S_ISDIR(inode->i_mode));
4778 /* catch async errors that were recorded back when async writeback
4779 * failed for pages in this mapping. */
4780 rc = lli->lli_async_rc;
4781 lli->lli_async_rc = 0;
4782 if (lli->lli_clob != NULL) {
4783 err = lov_read_and_clear_async_rc(lli->lli_clob);
4788 /* The application has been told write failure already.
4789 * Do not report failure again. */
4790 if (fd->fd_write_failed)
4792 return rc ? -EIO : 0;
4796 * Called to make sure a portion of file has been written out.
4797 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4799 * Return how many pages have been written.
4801 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4802 enum cl_fsync_mode mode, int ignore_layout)
4806 struct cl_fsync_io *fio;
4811 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4812 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL &&
4813 mode != CL_FSYNC_RECLAIM)
4816 env = cl_env_get(&refcheck);
4818 RETURN(PTR_ERR(env));
4820 io = vvp_env_thread_io(env);
4821 io->ci_obj = ll_i2info(inode)->lli_clob;
4822 io->ci_ignore_layout = ignore_layout;
4824 /* initialize parameters for sync */
4825 fio = &io->u.ci_fsync;
4826 fio->fi_start = start;
4828 fio->fi_fid = ll_inode2fid(inode);
4829 fio->fi_mode = mode;
4830 fio->fi_nr_written = 0;
4832 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4833 result = cl_io_loop(env, io);
4835 result = io->ci_result;
4837 result = fio->fi_nr_written;
4838 cl_io_fini(env, io);
4839 cl_env_put(env, &refcheck);
4845 * When dentry is provided (the 'else' case), file_dentry() may be
4846 * null and dentry must be used directly rather than pulled from
4847 * file_dentry() as is done otherwise.
4850 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4852 struct dentry *dentry = file_dentry(file);
4853 struct inode *inode = dentry->d_inode;
4854 struct ll_inode_info *lli = ll_i2info(inode);
4855 struct ptlrpc_request *req;
4856 ktime_t kstart = ktime_get();
4862 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4863 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4865 /* fsync's caller has already called _fdata{sync,write}, we want
4866 * that IO to finish before calling the osc and mdc sync methods */
4867 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4869 /* catch async errors that were recorded back when async writeback
4870 * failed for pages in this mapping. */
4871 if (!S_ISDIR(inode->i_mode)) {
4872 err = lli->lli_async_rc;
4873 lli->lli_async_rc = 0;
4876 if (lli->lli_clob != NULL) {
4877 err = lov_read_and_clear_async_rc(lli->lli_clob);
4883 if (S_ISREG(inode->i_mode) && !lli->lli_synced_to_mds) {
4885 * only the first sync on MDS makes sense,
4886 * everything else is stored on OSTs
4888 err = md_fsync(ll_i2sbi(inode)->ll_md_exp,
4889 ll_inode2fid(inode), &req);
4893 lli->lli_synced_to_mds = true;
4894 ptlrpc_req_finished(req);
4898 if (S_ISREG(inode->i_mode)) {
4899 struct ll_file_data *fd = file->private_data;
4902 /* Sync metadata on MDT first, and then sync the cached data
4905 err = pcc_fsync(file, start, end, datasync, &cached);
4907 err = cl_sync_file_range(inode, start, end,
4909 if (rc == 0 && err < 0)
4912 fd->fd_write_failed = true;
4914 fd->fd_write_failed = false;
4918 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4919 ktime_us_delta(ktime_get(), kstart));
4924 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4926 struct inode *inode = file_inode(file);
4927 struct ll_sb_info *sbi = ll_i2sbi(inode);
4928 struct ldlm_enqueue_info einfo = {
4929 .ei_type = LDLM_FLOCK,
4930 .ei_cb_cp = ldlm_flock_completion_ast,
4931 .ei_cbdata = file_lock,
4933 struct md_op_data *op_data;
4934 struct lustre_handle lockh = { 0 };
4935 union ldlm_policy_data flock = { { 0 } };
4936 int fl_type = file_lock->fl_type;
4937 ktime_t kstart = ktime_get();
4943 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4944 PFID(ll_inode2fid(inode)), file_lock);
4946 if (file_lock->fl_flags & FL_FLOCK) {
4947 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4948 /* flocks are whole-file locks */
4949 flock.l_flock.end = OFFSET_MAX;
4950 /* For flocks owner is determined by the local file desctiptor*/
4951 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4952 } else if (file_lock->fl_flags & FL_POSIX) {
4953 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4954 flock.l_flock.start = file_lock->fl_start;
4955 flock.l_flock.end = file_lock->fl_end;
4959 flock.l_flock.pid = file_lock->fl_pid;
4961 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4962 /* Somewhat ugly workaround for svc lockd.
4963 * lockd installs custom fl_lmops->lm_compare_owner that checks
4964 * for the fl_owner to be the same (which it always is on local node
4965 * I guess between lockd processes) and then compares pid.
4966 * As such we assign pid to the owner field to make it all work,
4967 * conflict with normal locks is unlikely since pid space and
4968 * pointer space for current->files are not intersecting */
4969 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4970 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4975 einfo.ei_mode = LCK_PR;
4978 /* An unlock request may or may not have any relation to
4979 * existing locks so we may not be able to pass a lock handle
4980 * via a normal ldlm_lock_cancel() request. The request may even
4981 * unlock a byte range in the middle of an existing lock. In
4982 * order to process an unlock request we need all of the same
4983 * information that is given with a normal read or write record
4984 * lock request. To avoid creating another ldlm unlock (cancel)
4985 * message we'll treat a LCK_NL flock request as an unlock. */
4986 einfo.ei_mode = LCK_NL;
4989 einfo.ei_mode = LCK_PW;
4993 CERROR("%s: fcntl from '%s' unknown lock type=%d: rc = %d\n",
4994 sbi->ll_fsname, current->comm, fl_type, rc);
5009 flags = LDLM_FL_BLOCK_NOWAIT;
5015 flags = LDLM_FL_TEST_LOCK;
5019 CERROR("%s: fcntl from '%s' unknown lock command=%d: rc = %d\n",
5020 sbi->ll_fsname, current->comm, cmd, rc);
5024 /* Save the old mode so that if the mode in the lock changes we
5025 * can decrement the appropriate reader or writer refcount. */
5026 file_lock->fl_type = einfo.ei_mode;
5028 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
5029 LUSTRE_OPC_ANY, NULL);
5030 if (IS_ERR(op_data))
5031 RETURN(PTR_ERR(op_data));
5033 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
5034 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
5035 flock.l_flock.pid, flags, einfo.ei_mode,
5036 flock.l_flock.start, flock.l_flock.end);
5038 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
5041 /* Restore the file lock type if not TEST lock. */
5042 if (!(flags & LDLM_FL_TEST_LOCK))
5043 file_lock->fl_type = fl_type;
5045 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
5046 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
5047 !(flags & LDLM_FL_TEST_LOCK))
5048 rc2 = locks_lock_file_wait(file, file_lock);
5050 if ((file_lock->fl_flags & FL_FLOCK) &&
5051 (rc == 0 || file_lock->fl_type == F_UNLCK))
5052 rc2 = flock_lock_file_wait(file, file_lock);
5053 if ((file_lock->fl_flags & FL_POSIX) &&
5054 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
5055 !(flags & LDLM_FL_TEST_LOCK))
5056 rc2 = posix_lock_file_wait(file, file_lock);
5057 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
5059 if (rc2 && file_lock->fl_type != F_UNLCK) {
5060 einfo.ei_mode = LCK_NL;
5061 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
5066 ll_finish_md_op_data(op_data);
5069 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
5070 ktime_us_delta(ktime_get(), kstart));
5074 int ll_get_fid_by_name(struct inode *parent, const char *name,
5075 int namelen, struct lu_fid *fid,
5076 struct inode **inode)
5078 struct md_op_data *op_data = NULL;
5079 struct mdt_body *body;
5080 struct ptlrpc_request *req;
5084 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
5085 LUSTRE_OPC_ANY, NULL);
5086 if (IS_ERR(op_data))
5087 RETURN(PTR_ERR(op_data));
5089 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
5090 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
5091 ll_finish_md_op_data(op_data);
5095 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
5097 GOTO(out_req, rc = -EFAULT);
5099 *fid = body->mbo_fid1;
5102 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
5104 ptlrpc_req_finished(req);
5108 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
5109 const char *name, __u32 flags)
5111 struct dentry *dchild = NULL;
5112 struct inode *child_inode = NULL;
5113 struct md_op_data *op_data;
5114 struct ptlrpc_request *request = NULL;
5115 struct obd_client_handle *och = NULL;
5117 struct mdt_body *body;
5118 __u64 data_version = 0;
5119 size_t namelen = strlen(name);
5120 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
5124 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
5125 PFID(ll_inode2fid(parent)), name,
5126 lum->lum_stripe_offset, lum->lum_stripe_count);
5128 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
5129 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
5130 lustre_swab_lmv_user_md(lum);
5132 /* Get child FID first */
5133 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
5136 dchild = d_lookup(file_dentry(file), &qstr);
5138 if (dchild->d_inode)
5139 child_inode = igrab(dchild->d_inode);
5144 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
5153 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
5154 OBD_CONNECT2_DIR_MIGRATE)) {
5155 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
5156 ll_dir_striped(child_inode)) {
5157 CERROR("%s: MDT doesn't support stripe directory "
5158 "migration!\n", ll_i2sbi(parent)->ll_fsname);
5159 GOTO(out_iput, rc = -EOPNOTSUPP);
5164 * lfs migrate command needs to be blocked on the client
5165 * by checking the migrate FID against the FID of the
5168 if (is_root_inode(child_inode))
5169 GOTO(out_iput, rc = -EINVAL);
5171 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
5172 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
5173 if (IS_ERR(op_data))
5174 GOTO(out_iput, rc = PTR_ERR(op_data));
5176 ll_inode_lock(child_inode);
5177 op_data->op_fid3 = *ll_inode2fid(child_inode);
5178 if (!fid_is_sane(&op_data->op_fid3)) {
5179 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
5180 ll_i2sbi(parent)->ll_fsname, name,
5181 PFID(&op_data->op_fid3));
5182 GOTO(out_unlock, rc = -EINVAL);
5185 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
5186 op_data->op_data = lum;
5187 op_data->op_data_size = lumlen;
5189 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
5190 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
5191 lmv_dir_layout_changing(op_data->op_lso1))
5192 op_data->op_bias |= MDS_MIGRATE_NSONLY;
5195 if (S_ISREG(child_inode->i_mode)) {
5196 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
5200 GOTO(out_unlock, rc);
5203 rc = ll_data_version(child_inode, &data_version,
5206 GOTO(out_close, rc);
5208 op_data->op_open_handle = och->och_open_handle;
5209 op_data->op_data_version = data_version;
5210 op_data->op_lease_handle = och->och_lease_handle;
5211 op_data->op_bias |= MDS_CLOSE_MIGRATE;
5213 spin_lock(&och->och_mod->mod_open_req->rq_lock);
5214 och->och_mod->mod_open_req->rq_replay = 0;
5215 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
5218 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
5219 op_data->op_name, op_data->op_namelen,
5220 op_data->op_name, op_data->op_namelen, &request);
5222 LASSERT(request != NULL);
5223 ll_update_times(request, parent);
5226 if (rc == 0 || rc == -EAGAIN) {
5227 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
5228 LASSERT(body != NULL);
5230 /* If the server does release layout lock, then we cleanup
5231 * the client och here, otherwise release it in out_close: */
5232 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
5233 obd_mod_put(och->och_mod);
5234 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
5236 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
5242 if (request != NULL) {
5243 ptlrpc_req_finished(request);
5247 /* Try again if the lease has cancelled. */
5248 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
5253 ll_lease_close(och, child_inode, NULL);
5255 clear_nlink(child_inode);
5257 ll_inode_unlock(child_inode);
5258 ll_finish_md_op_data(op_data);
5265 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
5267 struct ll_file_data *fd = file->private_data;
5271 * In order to avoid flood of warning messages, only print one message
5272 * for one file. And the entire message rate on the client is limited
5273 * by CDEBUG_LIMIT too.
5275 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5276 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5277 CDEBUG_LIMIT(D_CONSOLE,
5278 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5284 * test if some locks matching bits and l_req_mode are acquired
5285 * - bits can be in different locks
5286 * - if found clear the common lock bits in *bits
5287 * - the bits not found, are kept in *bits
5289 * \param bits [IN] searched lock bits [IN]
5290 * \param l_req_mode [IN] searched lock mode
5291 * \retval boolean, true iff all bits are found
5293 int ll_have_md_lock(struct obd_export *exp, struct inode *inode, __u64 *bits,
5294 enum ldlm_mode l_req_mode)
5296 struct lustre_handle lockh;
5297 union ldlm_policy_data policy;
5298 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5299 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5308 fid = &ll_i2info(inode)->lli_fid;
5309 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5310 ldlm_lockname[mode]);
5312 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5313 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5314 policy.l_inodebits.bits = *bits & BIT(i);
5315 if (policy.l_inodebits.bits == 0)
5318 if (md_lock_match(exp, flags, fid, LDLM_IBITS, &policy, mode,
5320 struct ldlm_lock *lock;
5322 lock = ldlm_handle2lock(&lockh);
5325 ~(lock->l_policy_data.l_inodebits.bits);
5326 LDLM_LOCK_PUT(lock);
5328 *bits &= ~policy.l_inodebits.bits;
5335 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5336 struct lustre_handle *lockh, __u64 flags,
5337 enum ldlm_mode mode)
5339 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5344 fid = &ll_i2info(inode)->lli_fid;
5345 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5347 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5348 fid, LDLM_IBITS, &policy, mode, lockh);
5353 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5355 /* Already unlinked. Just update nlink and return success */
5356 if (rc == -ENOENT) {
5358 /* If it is striped directory, and there is bad stripe
5359 * Let's revalidate the dentry again, instead of returning
5361 if (ll_dir_striped(inode))
5364 /* This path cannot be hit for regular files unless in
5365 * case of obscure races, so no need to to validate
5367 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5369 } else if (rc != 0) {
5370 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5371 "%s: revalidate FID "DFID" error: rc = %d\n",
5372 ll_i2sbi(inode)->ll_fsname,
5373 PFID(ll_inode2fid(inode)), rc);
5379 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5381 struct inode *parent;
5382 struct inode *inode = dentry->d_inode;
5383 struct obd_export *exp = ll_i2mdexp(inode);
5384 struct lookup_intent oit = {
5387 struct ptlrpc_request *req = NULL;
5388 struct md_op_data *op_data;
5389 const char *name = NULL;
5394 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5395 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5397 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5398 parent = dentry->d_parent->d_inode;
5399 name = dentry->d_name.name;
5400 namelen = dentry->d_name.len;
5405 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5406 LUSTRE_OPC_ANY, NULL);
5407 if (IS_ERR(op_data))
5408 RETURN(PTR_ERR(op_data));
5410 /* Call getattr by fid */
5411 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5412 op_data->op_flags = MF_GETATTR_BY_FID;
5413 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5414 ll_finish_md_op_data(op_data);
5416 rc = ll_inode_revalidate_fini(inode, rc);
5420 rc = ll_revalidate_it_finish(req, &oit, dentry);
5422 ll_intent_release(&oit);
5426 /* Unlinked? Unhash dentry, so it is not picked up later by
5427 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5428 * here to preserve get_cwd functionality on 2.6.
5430 if (!dentry->d_inode->i_nlink)
5431 d_lustre_invalidate(dentry);
5433 ll_lookup_finish_locks(&oit, dentry);
5435 ptlrpc_req_finished(req);
5440 static int ll_merge_md_attr(struct inode *inode)
5442 struct ll_inode_info *lli = ll_i2info(inode);
5443 struct lmv_stripe_object *lsm_obj;
5444 struct cl_attr attr = { 0 };
5447 if (!ll_dir_striped(inode))
5450 down_read(&lli->lli_lsm_sem);
5451 LASSERT(lli->lli_lsm_obj != NULL);
5453 lsm_obj = lmv_stripe_object_get(lli->lli_lsm_obj);
5454 up_read(&lli->lli_lsm_sem);
5456 rc = md_merge_attr(ll_i2mdexp(inode), lsm_obj,
5457 &attr, ll_md_blocking_ast);
5458 lmv_stripe_object_put(&lsm_obj);
5462 spin_lock(&inode->i_lock);
5463 set_nlink(inode, attr.cat_nlink);
5464 spin_unlock(&inode->i_lock);
5466 inode->i_blocks = attr.cat_blocks;
5467 i_size_write(inode, attr.cat_size);
5469 ll_i2info(inode)->lli_atime = attr.cat_atime;
5470 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5471 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5476 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5477 unsigned int flags, bool foreign)
5479 struct inode *inode = de->d_inode;
5480 struct ll_sb_info *sbi = ll_i2sbi(inode);
5481 struct ll_inode_info *lli = ll_i2info(inode);
5482 struct inode *dir = de->d_parent->d_inode;
5483 bool need_glimpse = true;
5484 ktime_t kstart = ktime_get();
5487 /* The OST object(s) determine the file size, blocks and mtime. */
5488 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5489 request_mask & STATX_MTIME))
5490 need_glimpse = false;
5492 if (dentry_may_statahead(dir, de))
5493 ll_start_statahead(dir, de, need_glimpse &&
5494 !(flags & AT_STATX_DONT_SYNC));
5496 if (flags & AT_STATX_DONT_SYNC)
5497 GOTO(fill_attr, rc = 0);
5499 rc = ll_inode_revalidate(de, IT_GETATTR);
5503 /* foreign file/dir are always of zero length, so don't
5504 * need to validate size.
5506 if (S_ISREG(inode->i_mode) && !foreign) {
5510 GOTO(fill_attr, rc);
5512 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5513 if (cached && rc < 0)
5517 GOTO(fill_attr, rc);
5520 * If the returned attr is masked with OBD_MD_FLSIZE &
5521 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5522 * or blocks obtained from MDT is strictly correct, and the file
5523 * is usually not being modified by clients, and the [a|m|c]time
5524 * got from MDT is also strictly correct.
5525 * Under this circumstance, it does not need to send glimpse
5526 * RPCs to OSTs for file attributes such as the size and blocks.
5528 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5529 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5530 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5531 inode->i_mtime.tv_sec = lli->lli_mtime;
5532 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5533 inode->i_atime.tv_sec = lli->lli_atime;
5534 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5535 inode->i_ctime.tv_sec = lli->lli_ctime;
5536 GOTO(fill_attr, rc);
5539 /* In case of restore, the MDT has the right size and has
5540 * already send it back without granting the layout lock,
5541 * inode is up-to-date so glimpse is useless.
5542 * Also to glimpse we need the layout, in case of a running
5543 * restore the MDT holds the layout lock so the glimpse will
5544 * block up to the end of restore (getattr will block)
5546 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5547 rc = ll_glimpse_size(inode);
5552 /* If object isn't regular a file then don't validate size. */
5553 /* foreign dir is not striped dir */
5555 rc = ll_merge_md_attr(inode);
5560 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5561 inode->i_atime.tv_sec = lli->lli_atime;
5562 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5563 inode->i_mtime.tv_sec = lli->lli_mtime;
5564 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5565 inode->i_ctime.tv_sec = lli->lli_ctime;
5569 CFS_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5571 if (ll_need_32bit_api(sbi)) {
5572 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5573 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5574 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5576 stat->ino = inode->i_ino;
5577 stat->dev = inode->i_sb->s_dev;
5578 stat->rdev = inode->i_rdev;
5581 /* foreign symlink to be exposed as a real symlink */
5583 stat->mode = inode->i_mode;
5585 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5587 stat->uid = inode->i_uid;
5588 stat->gid = inode->i_gid;
5589 stat->atime = inode->i_atime;
5590 stat->mtime = inode->i_mtime;
5591 stat->ctime = inode->i_ctime;
5592 /* stat->blksize is used to tell about preferred IO size */
5593 if (sbi->ll_stat_blksize)
5594 stat->blksize = sbi->ll_stat_blksize;
5595 else if (S_ISREG(inode->i_mode))
5596 stat->blksize = min(PTLRPC_MAX_BRW_SIZE,
5597 1U << LL_MAX_BLKSIZE_BITS);
5598 else if (S_ISDIR(inode->i_mode))
5599 stat->blksize = min(MD_MAX_BRW_SIZE,
5600 1U << LL_MAX_BLKSIZE_BITS);
5602 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5604 stat->nlink = inode->i_nlink;
5605 stat->size = i_size_read(inode);
5606 stat->blocks = inode->i_blocks;
5608 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5609 if (flags & AT_STATX_DONT_SYNC) {
5610 if (stat->size == 0 &&
5611 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5612 stat->size = lli->lli_lazysize;
5613 if (stat->blocks == 0 &&
5614 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5615 stat->blocks = lli->lli_lazyblocks;
5618 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5619 stat->result_mask |= STATX_BTIME;
5620 stat->btime.tv_sec = lli->lli_btime;
5623 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5624 #ifdef HAVE_LUSTRE_CRYPTO
5625 stat->attributes_mask |= STATX_ATTR_ENCRYPTED;
5627 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5628 /* if Lustre specific LUSTRE_ENCRYPT_FL flag is set, also set
5629 * ext4 equivalent to please statx
5631 if (stat->attributes & LUSTRE_ENCRYPT_FL)
5632 stat->attributes |= STATX_ATTR_ENCRYPTED;
5633 stat->result_mask &= request_mask;
5636 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5637 ktime_us_delta(ktime_get(), kstart));
5642 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5643 int ll_getattr(struct mnt_idmap *map, const struct path *path,
5644 struct kstat *stat, u32 request_mask, unsigned int flags)
5646 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5650 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5652 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5653 AT_STATX_SYNC_AS_STAT, false);
5657 static int cl_falloc(struct file *file, struct inode *inode, int mode,
5658 loff_t offset, loff_t len)
5660 loff_t size = i_size_read(inode);
5668 env = cl_env_get(&refcheck);
5670 RETURN(PTR_ERR(env));
5672 io = vvp_env_thread_io(env);
5673 io->ci_obj = ll_i2info(inode)->lli_clob;
5674 ll_io_set_mirror(io, file);
5676 io->ci_verify_layout = 1;
5677 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5678 io->u.ci_setattr.sa_falloc_mode = mode;
5679 io->u.ci_setattr.sa_falloc_offset = offset;
5680 io->u.ci_setattr.sa_falloc_end = offset + len;
5681 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5683 CDEBUG(D_INODE, "UID %u GID %u PRJID %u\n",
5684 from_kuid(&init_user_ns, inode->i_uid),
5685 from_kgid(&init_user_ns, inode->i_gid),
5686 ll_i2info(inode)->lli_projid);
5688 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5689 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5690 io->u.ci_setattr.sa_falloc_projid = ll_i2info(inode)->lli_projid;
5692 if (io->u.ci_setattr.sa_falloc_end > size) {
5693 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5695 /* Check new size against VFS/VM file size limit and rlimit */
5696 rc = inode_newsize_ok(inode, newsize);
5699 if (newsize > ll_file_maxbytes(inode)) {
5700 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5701 (unsigned long long)newsize,
5702 ll_file_maxbytes(inode));
5709 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5711 rc = cl_io_loop(env, io);
5714 cl_io_fini(env, io);
5715 } while (unlikely(io->ci_need_restart));
5718 cl_env_put(env, &refcheck);
5722 static long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5724 struct inode *inode = file_inode(filp);
5727 if (offset < 0 || len <= 0)
5730 * Encrypted inodes can't handle collapse range or zero range or insert
5731 * range since we would need to re-encrypt blocks with a different IV or
5732 * XTS tweak (which are based on the logical block number).
5733 * Similar to what ext4 does.
5735 if (IS_ENCRYPTED(inode) &&
5736 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5737 FALLOC_FL_ZERO_RANGE)))
5738 RETURN(-EOPNOTSUPP);
5741 * mode == 0 (which is standard prealloc) and PUNCH is supported
5742 * Rest of mode options are not supported yet.
5744 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5745 RETURN(-EOPNOTSUPP);
5747 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5749 rc = cl_falloc(filp, inode, mode, offset, len);
5751 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5752 * used by Lustre in several places. Retuning it here would
5753 * confuse applications that explicity test for EOPNOTSUPP
5754 * (95) and fall back to ftruncate().
5756 if (rc == -ENOTSUPP)
5762 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5763 __u64 start, __u64 len)
5767 struct fiemap *fiemap;
5768 unsigned int extent_count = fieinfo->fi_extents_max;
5770 num_bytes = sizeof(*fiemap) + (extent_count *
5771 sizeof(struct fiemap_extent));
5772 OBD_ALLOC_LARGE(fiemap, num_bytes);
5777 fiemap->fm_flags = fieinfo->fi_flags;
5778 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5779 fiemap->fm_start = start;
5780 fiemap->fm_length = len;
5781 if (extent_count > 0 &&
5782 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5783 sizeof(struct fiemap_extent)) != 0)
5784 GOTO(out, rc = -EFAULT);
5786 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5788 if (IS_ENCRYPTED(inode)) {
5791 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5792 fiemap->fm_extents[i].fe_flags |=
5793 FIEMAP_EXTENT_DATA_ENCRYPTED |
5794 FIEMAP_EXTENT_ENCODED;
5797 fieinfo->fi_flags = fiemap->fm_flags;
5798 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5799 if (extent_count > 0 &&
5800 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5801 fiemap->fm_mapped_extents *
5802 sizeof(struct fiemap_extent)) != 0)
5803 GOTO(out, rc = -EFAULT);
5805 OBD_FREE_LARGE(fiemap, num_bytes);
5809 int ll_inode_permission(struct mnt_idmap *idmap, struct inode *inode, int mask)
5812 struct ll_sb_info *sbi;
5813 struct root_squash_info *squash;
5814 struct cred *cred = NULL;
5815 const struct cred *old_cred = NULL;
5816 bool squash_id = false;
5817 ktime_t kstart = ktime_get();
5821 if (mask & MAY_NOT_BLOCK)
5825 * as root inode are NOT getting validated in lookup operation,
5826 * need to revalidate PERM before permission check.
5828 if (is_root_inode(inode)) {
5829 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_GETATTR);
5834 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5835 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5837 /* squash fsuid/fsgid if needed */
5838 sbi = ll_i2sbi(inode);
5839 squash = &sbi->ll_squash;
5840 if (unlikely(squash->rsi_uid != 0 &&
5841 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5842 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5846 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5847 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5848 squash->rsi_uid, squash->rsi_gid);
5850 /* update current process's credentials
5851 * and FS capability */
5852 cred = prepare_creds();
5856 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5857 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5858 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5859 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5861 old_cred = override_creds(cred);
5864 rc = generic_permission(idmap, inode, mask);
5865 /* restore current process's credentials and FS capability */
5867 revert_creds(old_cred);
5872 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5873 ktime_us_delta(ktime_get(), kstart));
5878 /* -o localflock - only provides locally consistent flock locks */
5879 static const struct file_operations ll_file_operations = {
5880 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5881 # ifdef HAVE_SYNC_READ_WRITE
5882 .read = new_sync_read,
5883 .write = new_sync_write,
5885 .read_iter = ll_file_read_iter,
5886 .write_iter = ll_file_write_iter,
5887 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5888 .read = ll_file_read,
5889 .aio_read = ll_file_aio_read,
5890 .write = ll_file_write,
5891 .aio_write = ll_file_aio_write,
5892 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5893 .unlocked_ioctl = ll_file_ioctl,
5894 .open = ll_file_open,
5895 .release = ll_file_release,
5896 .mmap = ll_file_mmap,
5897 .llseek = ll_file_seek,
5898 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5899 .splice_read = generic_file_splice_read,
5901 .splice_read = pcc_file_splice_read,
5903 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5904 .splice_write = iter_file_splice_write,
5908 .fallocate = ll_fallocate,
5911 static const struct file_operations ll_file_operations_flock = {
5912 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5913 # ifdef HAVE_SYNC_READ_WRITE
5914 .read = new_sync_read,
5915 .write = new_sync_write,
5916 # endif /* HAVE_SYNC_READ_WRITE */
5917 .read_iter = ll_file_read_iter,
5918 .write_iter = ll_file_write_iter,
5919 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5920 .read = ll_file_read,
5921 .aio_read = ll_file_aio_read,
5922 .write = ll_file_write,
5923 .aio_write = ll_file_aio_write,
5924 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5925 .unlocked_ioctl = ll_file_ioctl,
5926 .open = ll_file_open,
5927 .release = ll_file_release,
5928 .mmap = ll_file_mmap,
5929 .llseek = ll_file_seek,
5930 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5931 .splice_read = generic_file_splice_read,
5933 .splice_read = pcc_file_splice_read,
5935 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5936 .splice_write = iter_file_splice_write,
5940 .flock = ll_file_flock,
5941 .lock = ll_file_flock,
5942 .fallocate = ll_fallocate,
5945 /* These are for -o noflock - to return ENOSYS on flock calls */
5946 static const struct file_operations ll_file_operations_noflock = {
5947 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5948 # ifdef HAVE_SYNC_READ_WRITE
5949 .read = new_sync_read,
5950 .write = new_sync_write,
5951 # endif /* HAVE_SYNC_READ_WRITE */
5952 .read_iter = ll_file_read_iter,
5953 .write_iter = ll_file_write_iter,
5954 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5955 .read = ll_file_read,
5956 .aio_read = ll_file_aio_read,
5957 .write = ll_file_write,
5958 .aio_write = ll_file_aio_write,
5959 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5960 .unlocked_ioctl = ll_file_ioctl,
5961 .open = ll_file_open,
5962 .release = ll_file_release,
5963 .mmap = ll_file_mmap,
5964 .llseek = ll_file_seek,
5965 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5966 .splice_read = generic_file_splice_read,
5968 .splice_read = pcc_file_splice_read,
5970 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5971 .splice_write = iter_file_splice_write,
5975 .flock = ll_file_noflock,
5976 .lock = ll_file_noflock,
5977 .fallocate = ll_fallocate,
5980 const struct inode_operations ll_file_inode_operations = {
5981 .setattr = ll_setattr,
5982 .getattr = ll_getattr,
5983 .permission = ll_inode_permission,
5984 #ifdef HAVE_IOP_XATTR
5985 .setxattr = ll_setxattr,
5986 .getxattr = ll_getxattr,
5987 .removexattr = ll_removexattr,
5989 .listxattr = ll_listxattr,
5990 .fiemap = ll_fiemap,
5991 .get_acl = ll_get_acl,
5992 #ifdef HAVE_IOP_SET_ACL
5993 .set_acl = ll_set_acl,
5995 #ifdef HAVE_FILEATTR_GET
5996 .fileattr_get = ll_fileattr_get,
5997 .fileattr_set = ll_fileattr_set,
6001 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
6003 const struct file_operations *fops = &ll_file_operations_noflock;
6005 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
6006 fops = &ll_file_operations_flock;
6007 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
6008 fops = &ll_file_operations;
6013 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
6015 struct ll_inode_info *lli = ll_i2info(inode);
6016 struct cl_object *obj = lli->lli_clob;
6025 env = cl_env_get(&refcheck);
6027 RETURN(PTR_ERR(env));
6029 rc = cl_conf_set(env, lli->lli_clob, conf);
6033 if (conf->coc_opc == OBJECT_CONF_SET) {
6034 struct ldlm_lock *lock = conf->coc_lock;
6035 struct cl_layout cl = {
6039 LASSERT(lock != NULL);
6040 LASSERT(ldlm_has_layout(lock));
6042 /* it can only be allowed to match after layout is
6043 * applied to inode otherwise false layout would be
6044 * seen. Applying layout shoud happen before dropping
6045 * the intent lock. */
6046 ldlm_lock_allow_match(lock);
6048 rc = cl_object_layout_get(env, obj, &cl);
6053 DFID": layout version change: %u -> %u\n",
6054 PFID(&lli->lli_fid), ll_layout_version_get(lli),
6056 ll_layout_version_set(lli, cl.cl_layout_gen);
6060 cl_env_put(env, &refcheck);
6062 RETURN(rc < 0 ? rc : 0);
6065 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
6066 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
6069 struct ll_sb_info *sbi = ll_i2sbi(inode);
6070 struct ptlrpc_request *req;
6077 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
6078 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
6079 lock->l_lvb_data, lock->l_lvb_len);
6081 if (lock->l_lvb_data != NULL)
6084 /* if layout lock was granted right away, the layout is returned
6085 * within DLM_LVB of dlm reply; otherwise if the lock was ever
6086 * blocked and then granted via completion ast, we have to fetch
6087 * layout here. Please note that we can't use the LVB buffer in
6088 * completion AST because it doesn't have a large enough buffer */
6089 rc = ll_get_default_mdsize(sbi, &lmmsize);
6093 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
6094 XATTR_NAME_LOV, lmmsize, &req);
6097 GOTO(out, rc = 0); /* empty layout */
6104 if (lmmsize == 0) /* empty layout */
6107 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
6109 GOTO(out, rc = -EFAULT);
6111 OBD_ALLOC_LARGE(lvbdata, lmmsize);
6112 if (lvbdata == NULL)
6113 GOTO(out, rc = -ENOMEM);
6115 memcpy(lvbdata, lmm, lmmsize);
6116 lock_res_and_lock(lock);
6117 if (unlikely(lock->l_lvb_data == NULL)) {
6118 lock->l_lvb_type = LVB_T_LAYOUT;
6119 lock->l_lvb_data = lvbdata;
6120 lock->l_lvb_len = lmmsize;
6123 unlock_res_and_lock(lock);
6126 OBD_FREE_LARGE(lvbdata, lmmsize);
6131 ptlrpc_req_finished(req);
6136 * Apply the layout to the inode. Layout lock is held and will be released
6139 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
6140 struct inode *inode, bool try)
6142 struct ll_inode_info *lli = ll_i2info(inode);
6143 struct ll_sb_info *sbi = ll_i2sbi(inode);
6144 struct ldlm_lock *lock;
6145 struct cl_object_conf conf;
6148 bool wait_layout = false;
6151 LASSERT(lustre_handle_is_used(lockh));
6153 lock = ldlm_handle2lock(lockh);
6154 LASSERT(lock != NULL);
6156 if (!ldlm_has_layout(lock))
6157 GOTO(out, rc = -EAGAIN);
6159 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
6160 PFID(&lli->lli_fid), inode);
6162 /* in case this is a caching lock and reinstate with new inode */
6163 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
6165 lock_res_and_lock(lock);
6166 lvb_ready = ldlm_is_lvb_ready(lock);
6167 unlock_res_and_lock(lock);
6169 /* checking lvb_ready is racy but this is okay. The worst case is
6170 * that multi processes may configure the file on the same time. */
6174 rc = ll_layout_fetch(inode, lock);
6178 /* for layout lock, lmm is stored in lock's lvb.
6179 * lvb_data is immutable if the lock is held so it's safe to access it
6182 * set layout to file. Unlikely this will fail as old layout was
6183 * surely eliminated */
6184 memset(&conf, 0, sizeof conf);
6185 conf.coc_opc = OBJECT_CONF_SET;
6186 conf.coc_inode = inode;
6187 conf.coc_lock = lock;
6189 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
6190 conf.u.coc_layout.lb_len = lock->l_lvb_len;
6191 rc = ll_layout_conf(inode, &conf);
6193 /* refresh layout failed, need to wait */
6194 wait_layout = rc == -EBUSY;
6197 LDLM_LOCK_PUT(lock);
6198 ldlm_lock_decref(lockh, mode);
6200 /* wait for IO to complete if it's still being used. */
6202 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
6203 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6205 memset(&conf, 0, sizeof conf);
6206 conf.coc_opc = OBJECT_CONF_WAIT;
6207 conf.coc_inode = inode;
6208 rc = ll_layout_conf(inode, &conf);
6212 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
6213 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
6216 if (rc == -ERESTARTSYS) {
6219 struct cl_object * obj = lli->lli_clob;
6221 env = cl_env_get(&refcheck);
6223 RETURN(PTR_ERR(env));
6225 CDEBUG(D_INODE, "prune without lock "DFID"\n",
6226 PFID(lu_object_fid(&obj->co_lu)));
6228 trunc_sem_down_write(&lli->lli_trunc_sem);
6229 cl_object_prune(env, obj);
6230 trunc_sem_up_write(&lli->lli_trunc_sem);
6231 cl_env_put(env, &refcheck);
6240 * Issue layout intent RPC to MDS.
6241 * \param inode [in] file inode
6242 * \param intent [in] layout intent
6244 * \retval 0 on success
6245 * \retval < 0 error code
6247 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
6249 struct ll_inode_info *lli = ll_i2info(inode);
6250 struct ll_sb_info *sbi = ll_i2sbi(inode);
6251 struct md_op_data *op_data;
6252 struct lookup_intent it;
6253 struct ptlrpc_request *req;
6257 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
6258 0, 0, LUSTRE_OPC_ANY, NULL);
6259 if (IS_ERR(op_data))
6260 RETURN(PTR_ERR(op_data));
6262 op_data->op_data = intent;
6263 op_data->op_data_size = sizeof(*intent);
6265 memset(&it, 0, sizeof(it));
6266 it.it_op = IT_LAYOUT;
6267 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
6268 intent->li_opc == LAYOUT_INTENT_TRUNC)
6269 it.it_flags = FMODE_WRITE;
6271 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
6272 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6274 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
6275 &ll_md_blocking_ast, 0);
6276 if (it.it_request != NULL)
6277 ptlrpc_req_finished(it.it_request);
6278 it.it_request = NULL;
6280 ll_finish_md_op_data(op_data);
6282 /* set lock data in case this is a new lock */
6284 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
6286 ll_intent_drop_lock(&it);
6292 * This function checks if there exists a LAYOUT lock on the client side,
6293 * or enqueues it if it doesn't have one in cache.
6295 * This function will not hold layout lock so it may be revoked any time after
6296 * this function returns. Any operations depend on layout should be redone
6299 * This function should be called before lov_io_init() to get an uptodate
6300 * layout version, the caller should save the version number and after IO
6301 * is finished, this function should be called again to verify that layout
6302 * is not changed during IO time.
6304 int ll_layout_refresh(struct inode *inode, __u32 *gen)
6306 struct ll_inode_info *lli = ll_i2info(inode);
6307 struct ll_sb_info *sbi = ll_i2sbi(inode);
6308 struct lustre_handle lockh;
6309 struct layout_intent intent = {
6310 .li_opc = LAYOUT_INTENT_ACCESS,
6312 enum ldlm_mode mode;
6317 *gen = ll_layout_version_get(lli);
6318 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6319 *gen != CL_LAYOUT_GEN_NONE)
6323 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6324 LASSERT(S_ISREG(inode->i_mode));
6326 /* take layout lock mutex to enqueue layout lock exclusively. */
6327 mutex_lock(&lli->lli_layout_mutex);
6328 lli->lli_layout_lock_owner = current;
6331 /* mostly layout lock is caching on the local side, so try to
6332 * match it before grabbing layout lock mutex. */
6333 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6334 LCK_CR | LCK_CW | LCK_PR |
6336 if (mode != 0) { /* hit cached lock */
6337 rc = ll_layout_lock_set(&lockh, mode, inode, try);
6344 rc = ll_layout_intent(inode, &intent);
6350 *gen = ll_layout_version_get(lli);
6351 lli->lli_layout_lock_owner = NULL;
6352 mutex_unlock(&lli->lli_layout_mutex);
6358 * Issue layout intent RPC indicating where in a file an IO is about to write.
6360 * \param[in] inode file inode.
6361 * \param[in] ext write range with start offset of fille in bytes where
6362 * an IO is about to write, and exclusive end offset in
6365 * \retval 0 on success
6366 * \retval < 0 error code
6368 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6369 struct lu_extent *ext)
6371 struct layout_intent intent = {
6373 .li_extent.e_start = ext->e_start,
6374 .li_extent.e_end = ext->e_end,
6379 rc = ll_layout_intent(inode, &intent);
6385 * This function send a restore request to the MDT
6387 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6389 struct hsm_user_request *hur;
6393 len = sizeof(struct hsm_user_request) +
6394 sizeof(struct hsm_user_item);
6395 OBD_ALLOC(hur, len);
6399 hur->hur_request.hr_action = HUA_RESTORE;
6400 hur->hur_request.hr_archive_id = 0;
6401 hur->hur_request.hr_flags = 0;
6402 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6403 sizeof(hur->hur_user_item[0].hui_fid));
6404 hur->hur_user_item[0].hui_extent.offset = offset;
6405 hur->hur_user_item[0].hui_extent.length = length;
6406 hur->hur_request.hr_itemcount = 1;
6407 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,