4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
40 #include <lustre_dlm.h>
41 #include <linux/pagemap.h>
42 #include <linux/file.h>
43 #include <linux/sched.h>
44 #include <linux/user_namespace.h>
45 #include <linux/uidgid.h>
46 #include <linux/falloc.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <uapi/linux/llcrypt.h>
50 #include <lustre_swab.h>
52 #include "cl_object.h"
53 #include "llite_internal.h"
54 #include "vvp_internal.h"
57 struct inode *sp_inode;
62 __u64 pa_data_version;
68 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
70 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
73 static struct ll_file_data *ll_file_data_get(void)
75 struct ll_file_data *fd;
77 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
81 fd->fd_write_failed = false;
82 pcc_file_init(&fd->fd_pcc_file);
87 static void ll_file_data_put(struct ll_file_data *fd)
90 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
94 * Packs all the attributes into @op_data for the CLOSE rpc.
96 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
97 struct obd_client_handle *och)
101 ll_prep_md_op_data(op_data, inode, NULL, NULL,
102 0, 0, LUSTRE_OPC_ANY, NULL);
104 op_data->op_attr.ia_mode = inode->i_mode;
105 op_data->op_attr.ia_atime = inode->i_atime;
106 op_data->op_attr.ia_mtime = inode->i_mtime;
107 op_data->op_attr.ia_ctime = inode->i_ctime;
108 op_data->op_attr.ia_size = i_size_read(inode);
109 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
110 ATTR_MTIME | ATTR_MTIME_SET |
112 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
113 op_data->op_attr_blocks = inode->i_blocks;
114 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
115 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
116 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
117 op_data->op_open_handle = och->och_open_handle;
119 if (och->och_flags & FMODE_WRITE &&
120 test_and_clear_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags))
121 /* For HSM: if inode data has been modified, pack it so that
122 * MDT can set data dirty flag in the archive. */
123 op_data->op_bias |= MDS_DATA_MODIFIED;
129 * Perform a close, possibly with a bias.
130 * The meaning of "data" depends on the value of "bias".
132 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
133 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
136 static int ll_close_inode_openhandle(struct inode *inode,
137 struct obd_client_handle *och,
138 enum mds_op_bias bias, void *data)
140 struct obd_export *md_exp = ll_i2mdexp(inode);
141 const struct ll_inode_info *lli = ll_i2info(inode);
142 struct md_op_data *op_data;
143 struct ptlrpc_request *req = NULL;
147 if (class_exp2obd(md_exp) == NULL) {
148 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
149 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
153 OBD_ALLOC_PTR(op_data);
154 /* We leak openhandle and request here on error, but not much to be
155 * done in OOM case since app won't retry close on error either. */
157 GOTO(out, rc = -ENOMEM);
159 ll_prepare_close(inode, op_data, och);
161 case MDS_CLOSE_LAYOUT_MERGE:
162 /* merge blocks from the victim inode */
163 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
164 op_data->op_attr.ia_valid |= ATTR_SIZE;
165 op_data->op_xvalid |= OP_XVALID_BLOCKS;
167 case MDS_CLOSE_LAYOUT_SPLIT:
168 case MDS_CLOSE_LAYOUT_SWAP: {
169 struct split_param *sp = data;
171 LASSERT(data != NULL);
172 op_data->op_bias |= bias;
173 op_data->op_data_version = 0;
174 op_data->op_lease_handle = och->och_lease_handle;
175 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
176 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
177 op_data->op_mirror_id = sp->sp_mirror_id;
179 op_data->op_fid2 = *ll_inode2fid(data);
184 case MDS_CLOSE_RESYNC_DONE: {
185 struct ll_ioc_lease *ioc = data;
187 LASSERT(data != NULL);
188 op_data->op_attr_blocks +=
189 ioc->lil_count * op_data->op_attr_blocks;
190 op_data->op_attr.ia_valid |= ATTR_SIZE;
191 op_data->op_xvalid |= OP_XVALID_BLOCKS;
192 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
194 op_data->op_lease_handle = och->och_lease_handle;
195 op_data->op_data = &ioc->lil_ids[0];
196 op_data->op_data_size =
197 ioc->lil_count * sizeof(ioc->lil_ids[0]);
201 case MDS_PCC_ATTACH: {
202 struct pcc_param *param = data;
204 LASSERT(data != NULL);
205 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
206 op_data->op_archive_id = param->pa_archive_id;
207 op_data->op_data_version = param->pa_data_version;
208 op_data->op_lease_handle = och->och_lease_handle;
212 case MDS_HSM_RELEASE:
213 LASSERT(data != NULL);
214 op_data->op_bias |= MDS_HSM_RELEASE;
215 op_data->op_data_version = *(__u64 *)data;
216 op_data->op_lease_handle = och->och_lease_handle;
217 op_data->op_attr.ia_valid |= ATTR_SIZE;
218 op_data->op_xvalid |= OP_XVALID_BLOCKS;
222 LASSERT(data == NULL);
226 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
227 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
228 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
229 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
231 rc = md_close(md_exp, op_data, och->och_mod, &req);
232 if (rc != 0 && rc != -EINTR)
233 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
234 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
236 if (rc == 0 && op_data->op_bias & bias) {
237 struct mdt_body *body;
239 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
240 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
243 if (bias & MDS_PCC_ATTACH) {
244 struct pcc_param *param = data;
246 param->pa_layout_gen = body->mbo_layout_gen;
250 ll_finish_md_op_data(op_data);
254 md_clear_open_replay_data(md_exp, och);
255 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
258 ptlrpc_req_finished(req); /* This is close request */
262 int ll_md_real_close(struct inode *inode, fmode_t fmode)
264 struct ll_inode_info *lli = ll_i2info(inode);
265 struct obd_client_handle **och_p;
266 struct obd_client_handle *och;
271 if (fmode & FMODE_WRITE) {
272 och_p = &lli->lli_mds_write_och;
273 och_usecount = &lli->lli_open_fd_write_count;
274 } else if (fmode & FMODE_EXEC) {
275 och_p = &lli->lli_mds_exec_och;
276 och_usecount = &lli->lli_open_fd_exec_count;
278 LASSERT(fmode & FMODE_READ);
279 och_p = &lli->lli_mds_read_och;
280 och_usecount = &lli->lli_open_fd_read_count;
283 mutex_lock(&lli->lli_och_mutex);
284 if (*och_usecount > 0) {
285 /* There are still users of this handle, so skip
287 mutex_unlock(&lli->lli_och_mutex);
293 mutex_unlock(&lli->lli_och_mutex);
296 /* There might be a race and this handle may already
298 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
304 static int ll_md_close(struct inode *inode, struct file *file)
306 union ldlm_policy_data policy = {
307 .l_inodebits = { MDS_INODELOCK_OPEN },
309 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
310 struct ll_file_data *fd = file->private_data;
311 struct ll_inode_info *lli = ll_i2info(inode);
312 struct lustre_handle lockh;
313 enum ldlm_mode lockmode;
317 /* clear group lock, if present */
318 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
319 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
321 if (fd->fd_lease_och != NULL) {
324 /* Usually the lease is not released when the
325 * application crashed, we need to release here. */
326 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
327 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
328 "Clean up lease "DFID" %d/%d\n",
329 PFID(&lli->lli_fid), rc, lease_broken);
331 fd->fd_lease_och = NULL;
334 if (fd->fd_och != NULL) {
335 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
340 /* Let's see if we have good enough OPEN lock on the file and if
341 we can skip talking to MDS */
342 mutex_lock(&lli->lli_och_mutex);
343 if (fd->fd_omode & FMODE_WRITE) {
345 LASSERT(lli->lli_open_fd_write_count);
346 lli->lli_open_fd_write_count--;
347 } else if (fd->fd_omode & FMODE_EXEC) {
349 LASSERT(lli->lli_open_fd_exec_count);
350 lli->lli_open_fd_exec_count--;
353 LASSERT(lli->lli_open_fd_read_count);
354 lli->lli_open_fd_read_count--;
356 mutex_unlock(&lli->lli_och_mutex);
358 /* LU-4398: do not cache write open lock if the file has exec bit */
359 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
360 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
361 LDLM_IBITS, &policy, lockmode, &lockh))
362 rc = ll_md_real_close(inode, fd->fd_omode);
365 file->private_data = NULL;
366 ll_file_data_put(fd);
371 /* While this returns an error code, fput() the caller does not, so we need
372 * to make every effort to clean up all of our state here. Also, applications
373 * rarely check close errors and even if an error is returned they will not
374 * re-try the close call.
376 int ll_file_release(struct inode *inode, struct file *file)
378 struct ll_file_data *fd;
379 struct ll_sb_info *sbi = ll_i2sbi(inode);
380 struct ll_inode_info *lli = ll_i2info(inode);
381 ktime_t kstart = ktime_get();
386 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
387 PFID(ll_inode2fid(inode)), inode);
389 fd = file->private_data;
392 /* The last ref on @file, maybe not the the owner pid of statahead,
393 * because parent and child process can share the same file handle. */
394 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
395 ll_deauthorize_statahead(inode, fd);
397 if (is_root_inode(inode)) {
398 file->private_data = NULL;
399 ll_file_data_put(fd);
403 pcc_file_release(inode, file);
405 if (!S_ISDIR(inode->i_mode)) {
406 if (lli->lli_clob != NULL)
407 lov_read_and_clear_async_rc(lli->lli_clob);
408 lli->lli_async_rc = 0;
411 rc = ll_md_close(inode, file);
413 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
414 libcfs_debug_dumplog();
417 if (!rc && !is_root_inode(inode))
418 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
419 ktime_us_delta(ktime_get(), kstart));
423 static inline int ll_dom_readpage(void *data, struct page *page)
425 struct niobuf_local *lnb = data;
429 struct inode *inode = page2inode(page);
431 kaddr = kmap_atomic(page);
432 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
433 if (lnb->lnb_len < PAGE_SIZE)
434 memset(kaddr + lnb->lnb_len, 0,
435 PAGE_SIZE - lnb->lnb_len);
436 flush_dcache_page(page);
437 SetPageUptodate(page);
438 kunmap_atomic(kaddr);
440 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
441 if (!llcrypt_has_encryption_key(inode))
442 CDEBUG(D_SEC, "no enc key for "DFID"\n",
443 PFID(ll_inode2fid(inode)));
445 unsigned int offs = 0;
447 while (offs < PAGE_SIZE) {
448 /* decrypt only if page is not empty */
449 if (memcmp(page_address(page) + offs,
450 page_address(ZERO_PAGE(0)),
451 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
454 rc = llcrypt_decrypt_pagecache_blocks(page,
455 LUSTRE_ENCRYPTION_UNIT_SIZE,
460 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
469 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
473 struct ll_inode_info *lli = ll_i2info(inode);
474 struct cl_object *obj = lli->lli_clob;
475 struct address_space *mapping = inode->i_mapping;
477 struct niobuf_remote *rnb;
478 struct mdt_body *body;
480 unsigned long index, start;
481 struct niobuf_local lnb;
490 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
494 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
495 if (rnb == NULL || rnb->rnb_len == 0)
498 /* LU-11595: Server may return whole file and that is OK always or
499 * it may return just file tail and its offset must be aligned with
500 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
501 * smaller then offset may be not aligned and that data is just ignored.
503 if (rnb->rnb_offset & ~PAGE_MASK)
506 /* Server returns whole file or just file tail if it fills in reply
507 * buffer, in both cases total size should be equal to the file size.
509 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
510 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
511 !(inode && IS_ENCRYPTED(inode))) {
512 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
513 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
514 rnb->rnb_len, body->mbo_dom_size);
518 env = cl_env_get(&refcheck);
521 io = vvp_env_thread_io(env);
523 io->ci_ignore_layout = 1;
524 rc = cl_io_init(env, io, CIT_MISC, obj);
528 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
529 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
531 data = (char *)rnb + sizeof(*rnb);
533 lnb.lnb_file_offset = rnb->rnb_offset;
534 start = lnb.lnb_file_offset >> PAGE_SHIFT;
536 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
537 lnb.lnb_page_offset = 0;
539 struct cl_page *page;
541 lnb.lnb_data = data + (index << PAGE_SHIFT);
542 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
543 if (lnb.lnb_len > PAGE_SIZE)
544 lnb.lnb_len = PAGE_SIZE;
546 vmpage = read_cache_page(mapping, index + start,
547 ll_dom_readpage, &lnb);
548 if (IS_ERR(vmpage)) {
549 CWARN("%s: cannot fill page %lu for "DFID
550 " with data: rc = %li\n",
551 ll_i2sbi(inode)->ll_fsname, index + start,
552 PFID(lu_object_fid(&obj->co_lu)),
557 if (vmpage->mapping == NULL) {
560 /* page was truncated */
563 /* attach VM page to CL page cache */
564 page = cl_page_find(env, obj, vmpage->index, vmpage,
567 ClearPageUptodate(vmpage);
572 cl_page_export(env, page, 1);
573 cl_page_put(env, page);
577 } while (rnb->rnb_len > (index << PAGE_SHIFT));
581 cl_env_put(env, &refcheck);
586 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
587 struct lookup_intent *itp)
589 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
590 struct dentry *parent = de->d_parent;
593 struct md_op_data *op_data;
594 struct ptlrpc_request *req = NULL;
598 LASSERT(parent != NULL);
599 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
601 /* if server supports open-by-fid, or file name is invalid, don't pack
602 * name in open request */
603 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
604 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
606 len = de->d_name.len;
607 name = kmalloc(len + 1, GFP_NOFS);
612 spin_lock(&de->d_lock);
613 if (len != de->d_name.len) {
614 spin_unlock(&de->d_lock);
618 memcpy(name, de->d_name.name, len);
620 spin_unlock(&de->d_lock);
622 if (!lu_name_is_valid_2(name, len)) {
628 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
629 name, len, 0, LUSTRE_OPC_ANY, NULL);
630 if (IS_ERR(op_data)) {
632 RETURN(PTR_ERR(op_data));
634 op_data->op_data = lmm;
635 op_data->op_data_size = lmmsize;
637 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
638 &ll_md_blocking_ast, 0);
640 ll_finish_md_op_data(op_data);
642 /* reason for keep own exit path - don`t flood log
643 * with messages with -ESTALE errors.
645 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
646 it_open_error(DISP_OPEN_OPEN, itp))
648 ll_release_openhandle(de, itp);
652 if (it_disposition(itp, DISP_LOOKUP_NEG))
653 GOTO(out, rc = -ENOENT);
655 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
656 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
657 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
661 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
663 if (!rc && itp->it_lock_mode) {
666 /* If we got a lock back and it has a LOOKUP bit set,
667 * make sure the dentry is marked as valid so we can find it.
668 * We don't need to care about actual hashing since other bits
669 * of kernel will deal with that later.
671 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
672 if (bits & MDS_INODELOCK_LOOKUP)
673 d_lustre_revalidate(de);
674 /* if DoM bit returned along with LAYOUT bit then there
675 * can be read-on-open data returned.
677 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
678 ll_dom_finish_open(de->d_inode, req);
682 ptlrpc_req_finished(req);
683 ll_intent_drop_lock(itp);
685 /* We did open by fid, but by the time we got to the server,
686 * the object disappeared. If this is a create, we cannot really
687 * tell the userspace that the file it was trying to create
688 * does not exist. Instead let's return -ESTALE, and the VFS will
689 * retry the create with LOOKUP_REVAL that we are going to catch
690 * in ll_revalidate_dentry() and use lookup then.
692 if (rc == -ENOENT && itp->it_op & IT_CREAT)
698 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
699 struct obd_client_handle *och)
701 struct mdt_body *body;
703 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
704 och->och_open_handle = body->mbo_open_handle;
705 och->och_fid = body->mbo_fid1;
706 och->och_lease_handle.cookie = it->it_lock_handle;
707 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
708 och->och_flags = it->it_flags;
710 return md_set_open_replay_data(md_exp, och, it);
713 static int ll_local_open(struct file *file, struct lookup_intent *it,
714 struct ll_file_data *fd, struct obd_client_handle *och)
716 struct inode *inode = file_inode(file);
719 LASSERT(!file->private_data);
726 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
731 file->private_data = fd;
732 ll_readahead_init(inode, &fd->fd_ras);
733 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
734 /* turn off the kernel's read-ahead */
735 file->f_ra.ra_pages = 0;
737 /* ll_cl_context initialize */
738 rwlock_init(&fd->fd_lock);
739 INIT_LIST_HEAD(&fd->fd_lccs);
744 /* Open a file, and (for the very first open) create objects on the OSTs at
745 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
746 * creation or open until ll_lov_setstripe() ioctl is called.
748 * If we already have the stripe MD locally then we don't request it in
749 * md_open(), by passing a lmm_size = 0.
751 * It is up to the application to ensure no other processes open this file
752 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
753 * used. We might be able to avoid races of that sort by getting lli_open_sem
754 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
755 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
757 int ll_file_open(struct inode *inode, struct file *file)
759 struct ll_inode_info *lli = ll_i2info(inode);
760 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
761 .it_flags = file->f_flags };
762 struct obd_client_handle **och_p = NULL;
763 __u64 *och_usecount = NULL;
764 struct ll_file_data *fd;
765 ktime_t kstart = ktime_get();
769 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
770 PFID(ll_inode2fid(inode)), inode, file->f_flags);
772 it = file->private_data; /* XXX: compat macro */
773 file->private_data = NULL; /* prevent ll_local_open assertion */
775 if (S_ISREG(inode->i_mode)) {
776 rc = llcrypt_file_open(inode, file);
778 GOTO(out_nofiledata, rc);
781 fd = ll_file_data_get();
783 GOTO(out_nofiledata, rc = -ENOMEM);
786 if (S_ISDIR(inode->i_mode))
787 ll_authorize_statahead(inode, fd);
789 if (is_root_inode(inode)) {
790 file->private_data = fd;
794 if (!it || !it->it_disposition) {
795 /* Convert f_flags into access mode. We cannot use file->f_mode,
796 * because everything but O_ACCMODE mask was stripped from
798 if ((oit.it_flags + 1) & O_ACCMODE)
800 if (file->f_flags & O_TRUNC)
801 oit.it_flags |= FMODE_WRITE;
803 /* kernel only call f_op->open in dentry_open. filp_open calls
804 * dentry_open after call to open_namei that checks permissions.
805 * Only nfsd_open call dentry_open directly without checking
806 * permissions and because of that this code below is safe.
808 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
809 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
811 /* We do not want O_EXCL here, presumably we opened the file
812 * already? XXX - NFS implications? */
813 oit.it_flags &= ~O_EXCL;
815 /* bug20584, if "it_flags" contains O_CREAT, the file will be
816 * created if necessary, then "IT_CREAT" should be set to keep
817 * consistent with it */
818 if (oit.it_flags & O_CREAT)
819 oit.it_op |= IT_CREAT;
825 /* Let's see if we have file open on MDS already. */
826 if (it->it_flags & FMODE_WRITE) {
827 och_p = &lli->lli_mds_write_och;
828 och_usecount = &lli->lli_open_fd_write_count;
829 } else if (it->it_flags & FMODE_EXEC) {
830 och_p = &lli->lli_mds_exec_och;
831 och_usecount = &lli->lli_open_fd_exec_count;
833 och_p = &lli->lli_mds_read_och;
834 och_usecount = &lli->lli_open_fd_read_count;
837 mutex_lock(&lli->lli_och_mutex);
838 if (*och_p) { /* Open handle is present */
839 if (it_disposition(it, DISP_OPEN_OPEN)) {
840 /* Well, there's extra open request that we do not need,
841 * let's close it somehow. This will decref request. */
842 rc = it_open_error(DISP_OPEN_OPEN, it);
844 mutex_unlock(&lli->lli_och_mutex);
845 GOTO(out_openerr, rc);
848 ll_release_openhandle(file_dentry(file), it);
852 rc = ll_local_open(file, it, fd, NULL);
855 mutex_unlock(&lli->lli_och_mutex);
856 GOTO(out_openerr, rc);
859 LASSERT(*och_usecount == 0);
860 if (!it->it_disposition) {
861 struct dentry *dentry = file_dentry(file);
862 struct ll_dentry_data *ldd;
864 /* We cannot just request lock handle now, new ELC code
865 * means that one of other OPEN locks for this file
866 * could be cancelled, and since blocking ast handler
867 * would attempt to grab och_mutex as well, that would
868 * result in a deadlock
870 mutex_unlock(&lli->lli_och_mutex);
872 * Normally called under two situations:
874 * 2. A race/condition on MDS resulting in no open
875 * handle to be returned from LOOKUP|OPEN request,
876 * for example if the target entry was a symlink.
878 * Only fetch MDS_OPEN_LOCK if this is in NFS path,
879 * marked by a bit set in ll_iget_for_nfs. Clear the
880 * bit so that it's not confusing later callers.
882 * NB; when ldd is NULL, it must have come via normal
883 * lookup path only, since ll_iget_for_nfs always calls
886 ldd = ll_d2d(dentry);
887 if (ldd && ldd->lld_nfs_dentry) {
888 ldd->lld_nfs_dentry = 0;
889 if (!filename_is_volatile(dentry->d_name.name,
892 it->it_flags |= MDS_OPEN_LOCK;
896 * Always specify MDS_OPEN_BY_FID because we don't want
897 * to get file with different fid.
899 it->it_flags |= MDS_OPEN_BY_FID;
900 rc = ll_intent_file_open(dentry, NULL, 0, it);
902 GOTO(out_openerr, rc);
906 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
908 GOTO(out_och_free, rc = -ENOMEM);
912 /* md_intent_lock() didn't get a request ref if there was an
913 * open error, so don't do cleanup on the request here
915 /* XXX (green): Should not we bail out on any error here, not
916 * just open error? */
917 rc = it_open_error(DISP_OPEN_OPEN, it);
919 GOTO(out_och_free, rc);
921 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
922 "inode %p: disposition %x, status %d\n", inode,
923 it_disposition(it, ~0), it->it_status);
925 rc = ll_local_open(file, it, fd, *och_p);
927 GOTO(out_och_free, rc);
930 rc = pcc_file_open(inode, file);
932 GOTO(out_och_free, rc);
934 mutex_unlock(&lli->lli_och_mutex);
938 /* Must do this outside lli_och_mutex lock to prevent deadlock where
939 different kind of OPEN lock for this same inode gets cancelled
940 by ldlm_cancel_lru */
941 if (!S_ISREG(inode->i_mode))
942 GOTO(out_och_free, rc);
943 cl_lov_delay_create_clear(&file->f_flags);
944 GOTO(out_och_free, rc);
948 if (och_p && *och_p) {
949 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
950 *och_p = NULL; /* OBD_FREE writes some magic there */
953 mutex_unlock(&lli->lli_och_mutex);
956 if (lli->lli_opendir_key == fd)
957 ll_deauthorize_statahead(inode, fd);
960 ll_file_data_put(fd);
962 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
963 ktime_us_delta(ktime_get(), kstart));
967 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
968 ptlrpc_req_finished(it->it_request);
969 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
975 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
976 struct ldlm_lock_desc *desc, void *data, int flag)
979 struct lustre_handle lockh;
983 case LDLM_CB_BLOCKING:
984 ldlm_lock2handle(lock, &lockh);
985 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
987 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
991 case LDLM_CB_CANCELING:
999 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1000 * and save it as fd->fd_och so as to force client to reopen the file even
1001 * if it has an open lock in cache already.
1003 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1004 struct lustre_handle *old_open_handle)
1006 struct ll_inode_info *lli = ll_i2info(inode);
1007 struct ll_file_data *fd = file->private_data;
1008 struct obd_client_handle **och_p;
1009 __u64 *och_usecount;
1013 /* Get the openhandle of the file */
1014 mutex_lock(&lli->lli_och_mutex);
1015 if (fd->fd_lease_och != NULL)
1016 GOTO(out_unlock, rc = -EBUSY);
1018 if (fd->fd_och == NULL) {
1019 if (file->f_mode & FMODE_WRITE) {
1020 LASSERT(lli->lli_mds_write_och != NULL);
1021 och_p = &lli->lli_mds_write_och;
1022 och_usecount = &lli->lli_open_fd_write_count;
1024 LASSERT(lli->lli_mds_read_och != NULL);
1025 och_p = &lli->lli_mds_read_och;
1026 och_usecount = &lli->lli_open_fd_read_count;
1029 if (*och_usecount > 1)
1030 GOTO(out_unlock, rc = -EBUSY);
1032 fd->fd_och = *och_p;
1037 *old_open_handle = fd->fd_och->och_open_handle;
1041 mutex_unlock(&lli->lli_och_mutex);
1046 * Release ownership on lli_mds_*_och when putting back a file lease.
1048 static int ll_lease_och_release(struct inode *inode, struct file *file)
1050 struct ll_inode_info *lli = ll_i2info(inode);
1051 struct ll_file_data *fd = file->private_data;
1052 struct obd_client_handle **och_p;
1053 struct obd_client_handle *old_och = NULL;
1054 __u64 *och_usecount;
1058 mutex_lock(&lli->lli_och_mutex);
1059 if (file->f_mode & FMODE_WRITE) {
1060 och_p = &lli->lli_mds_write_och;
1061 och_usecount = &lli->lli_open_fd_write_count;
1063 och_p = &lli->lli_mds_read_och;
1064 och_usecount = &lli->lli_open_fd_read_count;
1067 /* The file may have been open by another process (broken lease) so
1068 * *och_p is not NULL. In this case we should simply increase usecount
1071 if (*och_p != NULL) {
1072 old_och = fd->fd_och;
1075 *och_p = fd->fd_och;
1079 mutex_unlock(&lli->lli_och_mutex);
1081 if (old_och != NULL)
1082 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1088 * Acquire a lease and open the file.
1090 static struct obd_client_handle *
1091 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1094 struct lookup_intent it = { .it_op = IT_OPEN };
1095 struct ll_sb_info *sbi = ll_i2sbi(inode);
1096 struct md_op_data *op_data;
1097 struct ptlrpc_request *req = NULL;
1098 struct lustre_handle old_open_handle = { 0 };
1099 struct obd_client_handle *och = NULL;
1104 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1105 RETURN(ERR_PTR(-EINVAL));
1108 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1109 RETURN(ERR_PTR(-EPERM));
1111 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1113 RETURN(ERR_PTR(rc));
1118 RETURN(ERR_PTR(-ENOMEM));
1120 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1121 LUSTRE_OPC_ANY, NULL);
1122 if (IS_ERR(op_data))
1123 GOTO(out, rc = PTR_ERR(op_data));
1125 /* To tell the MDT this openhandle is from the same owner */
1126 op_data->op_open_handle = old_open_handle;
1128 it.it_flags = fmode | open_flags;
1129 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1130 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1131 &ll_md_blocking_lease_ast,
1132 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1133 * it can be cancelled which may mislead applications that the lease is
1135 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1136 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1137 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1138 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1139 ll_finish_md_op_data(op_data);
1140 ptlrpc_req_finished(req);
1142 GOTO(out_release_it, rc);
1144 if (it_disposition(&it, DISP_LOOKUP_NEG))
1145 GOTO(out_release_it, rc = -ENOENT);
1147 rc = it_open_error(DISP_OPEN_OPEN, &it);
1149 GOTO(out_release_it, rc);
1151 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1152 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1154 GOTO(out_release_it, rc);
1156 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1157 GOTO(out_close, rc = -EOPNOTSUPP);
1159 /* already get lease, handle lease lock */
1160 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1161 if (!it.it_lock_mode ||
1162 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1163 /* open lock must return for lease */
1164 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1165 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1167 GOTO(out_close, rc = -EPROTO);
1170 ll_intent_release(&it);
1174 /* Cancel open lock */
1175 if (it.it_lock_mode != 0) {
1176 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1178 it.it_lock_mode = 0;
1179 och->och_lease_handle.cookie = 0ULL;
1181 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1183 CERROR("%s: error closing file "DFID": %d\n",
1184 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1185 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1187 ll_intent_release(&it);
1191 RETURN(ERR_PTR(rc));
1195 * Check whether a layout swap can be done between two inodes.
1197 * \param[in] inode1 First inode to check
1198 * \param[in] inode2 Second inode to check
1200 * \retval 0 on success, layout swap can be performed between both inodes
1201 * \retval negative error code if requirements are not met
1203 static int ll_check_swap_layouts_validity(struct inode *inode1,
1204 struct inode *inode2)
1206 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1209 if (inode_permission(inode1, MAY_WRITE) ||
1210 inode_permission(inode2, MAY_WRITE))
1213 if (inode1->i_sb != inode2->i_sb)
1219 static int ll_swap_layouts_close(struct obd_client_handle *och,
1220 struct inode *inode, struct inode *inode2)
1222 const struct lu_fid *fid1 = ll_inode2fid(inode);
1223 const struct lu_fid *fid2;
1227 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1228 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1230 rc = ll_check_swap_layouts_validity(inode, inode2);
1232 GOTO(out_free_och, rc);
1234 /* We now know that inode2 is a lustre inode */
1235 fid2 = ll_inode2fid(inode2);
1237 rc = lu_fid_cmp(fid1, fid2);
1239 GOTO(out_free_och, rc = -EINVAL);
1241 /* Close the file and {swap,merge} layouts between inode & inode2.
1242 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1243 * because we still need it to pack l_remote_handle to MDT. */
1244 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1247 och = NULL; /* freed in ll_close_inode_openhandle() */
1257 * Release lease and close the file.
1258 * It will check if the lease has ever broken.
1260 static int ll_lease_close_intent(struct obd_client_handle *och,
1261 struct inode *inode,
1262 bool *lease_broken, enum mds_op_bias bias,
1265 struct ldlm_lock *lock;
1266 bool cancelled = true;
1270 lock = ldlm_handle2lock(&och->och_lease_handle);
1272 lock_res_and_lock(lock);
1273 cancelled = ldlm_is_cancel(lock);
1274 unlock_res_and_lock(lock);
1275 LDLM_LOCK_PUT(lock);
1278 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1279 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1281 if (lease_broken != NULL)
1282 *lease_broken = cancelled;
1284 if (!cancelled && !bias)
1285 ldlm_cli_cancel(&och->och_lease_handle, 0);
1287 if (cancelled) { /* no need to excute intent */
1292 rc = ll_close_inode_openhandle(inode, och, bias, data);
1296 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1299 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1303 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1305 static int ll_lease_file_resync(struct obd_client_handle *och,
1306 struct inode *inode, unsigned long arg)
1308 struct ll_sb_info *sbi = ll_i2sbi(inode);
1309 struct md_op_data *op_data;
1310 struct ll_ioc_lease_id ioc;
1311 __u64 data_version_unused;
1315 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1316 LUSTRE_OPC_ANY, NULL);
1317 if (IS_ERR(op_data))
1318 RETURN(PTR_ERR(op_data));
1320 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1324 /* before starting file resync, it's necessary to clean up page cache
1325 * in client memory, otherwise once the layout version is increased,
1326 * writing back cached data will be denied the OSTs. */
1327 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1331 op_data->op_lease_handle = och->och_lease_handle;
1332 op_data->op_mirror_id = ioc.lil_mirror_id;
1333 rc = md_file_resync(sbi->ll_md_exp, op_data);
1339 ll_finish_md_op_data(op_data);
1343 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1345 struct ll_inode_info *lli = ll_i2info(inode);
1346 struct cl_object *obj = lli->lli_clob;
1347 struct cl_attr *attr = vvp_env_thread_attr(env);
1355 ll_inode_size_lock(inode);
1357 /* Merge timestamps the most recently obtained from MDS with
1358 * timestamps obtained from OSTs.
1360 * Do not overwrite atime of inode because it may be refreshed
1361 * by file_accessed() function. If the read was served by cache
1362 * data, there is no RPC to be sent so that atime may not be
1363 * transferred to OSTs at all. MDT only updates atime at close time
1364 * if it's at least 'mdd.*.atime_diff' older.
1365 * All in all, the atime in Lustre does not strictly comply with
1366 * POSIX. Solving this problem needs to send an RPC to MDT for each
1367 * read, this will hurt performance.
1369 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1370 inode->i_atime.tv_sec < lli->lli_atime)
1371 inode->i_atime.tv_sec = lli->lli_atime;
1373 inode->i_mtime.tv_sec = lli->lli_mtime;
1374 inode->i_ctime.tv_sec = lli->lli_ctime;
1376 mtime = inode->i_mtime.tv_sec;
1377 atime = inode->i_atime.tv_sec;
1378 ctime = inode->i_ctime.tv_sec;
1380 cl_object_attr_lock(obj);
1381 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1384 rc = cl_object_attr_get(env, obj, attr);
1385 cl_object_attr_unlock(obj);
1388 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1390 if (atime < attr->cat_atime)
1391 atime = attr->cat_atime;
1393 if (ctime < attr->cat_ctime)
1394 ctime = attr->cat_ctime;
1396 if (mtime < attr->cat_mtime)
1397 mtime = attr->cat_mtime;
1399 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1400 PFID(&lli->lli_fid), attr->cat_size);
1402 i_size_write(inode, attr->cat_size);
1403 inode->i_blocks = attr->cat_blocks;
1405 inode->i_mtime.tv_sec = mtime;
1406 inode->i_atime.tv_sec = atime;
1407 inode->i_ctime.tv_sec = ctime;
1410 ll_inode_size_unlock(inode);
1416 * Set designated mirror for I/O.
1418 * So far only read, write, and truncated can support to issue I/O to
1419 * designated mirror.
1421 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1423 struct ll_file_data *fd = file->private_data;
1425 /* clear layout version for generic(non-resync) I/O in case it carries
1426 * stale layout version due to I/O restart */
1427 io->ci_layout_version = 0;
1429 /* FLR: disable non-delay for designated mirror I/O because obviously
1430 * only one mirror is available */
1431 if (fd->fd_designated_mirror > 0) {
1433 io->ci_designated_mirror = fd->fd_designated_mirror;
1434 io->ci_layout_version = fd->fd_layout_version;
1437 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1438 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1441 static bool file_is_noatime(const struct file *file)
1443 const struct vfsmount *mnt = file->f_path.mnt;
1444 const struct inode *inode = file_inode((struct file *)file);
1446 /* Adapted from file_accessed() and touch_atime().*/
1447 if (file->f_flags & O_NOATIME)
1450 if (inode->i_flags & S_NOATIME)
1453 if (IS_NOATIME(inode))
1456 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1459 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1462 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1468 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1469 struct vvp_io_args *args)
1471 struct inode *inode = file_inode(file);
1472 struct ll_file_data *fd = file->private_data;
1474 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1475 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1477 if (iot == CIT_WRITE) {
1478 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1479 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1480 file->f_flags & O_DIRECT ||
1482 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1483 io->u.ci_wr.wr_sync |= !!(args &&
1484 (args->u.normal.via_iocb->ki_flags &
1489 io->ci_obj = ll_i2info(inode)->lli_clob;
1490 io->ci_lockreq = CILR_MAYBE;
1491 if (ll_file_nolock(file)) {
1492 io->ci_lockreq = CILR_NEVER;
1493 io->ci_no_srvlock = 1;
1494 } else if (file->f_flags & O_APPEND) {
1495 io->ci_lockreq = CILR_MANDATORY;
1497 io->ci_noatime = file_is_noatime(file);
1498 io->ci_async_readahead = false;
1500 /* FLR: only use non-delay I/O for read as there is only one
1501 * avaliable mirror for write. */
1502 io->ci_ndelay = !(iot == CIT_WRITE);
1504 ll_io_set_mirror(io, file);
1507 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1510 struct ll_inode_info *lli = ll_i2info(inode);
1511 struct ll_sb_info *sbi = ll_i2sbi(inode);
1512 enum obd_heat_type sample_type;
1513 enum obd_heat_type iobyte_type;
1514 __u64 now = ktime_get_real_seconds();
1516 if (!ll_sbi_has_file_heat(sbi) ||
1517 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1520 if (iot == CIT_READ) {
1521 sample_type = OBD_HEAT_READSAMPLE;
1522 iobyte_type = OBD_HEAT_READBYTE;
1523 } else if (iot == CIT_WRITE) {
1524 sample_type = OBD_HEAT_WRITESAMPLE;
1525 iobyte_type = OBD_HEAT_WRITEBYTE;
1530 spin_lock(&lli->lli_heat_lock);
1531 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1532 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1533 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1534 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1535 spin_unlock(&lli->lli_heat_lock);
1539 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1540 struct file *file, enum cl_io_type iot,
1541 loff_t *ppos, size_t count)
1543 struct vvp_io *vio = vvp_env_io(env);
1544 struct inode *inode = file_inode(file);
1545 struct ll_inode_info *lli = ll_i2info(inode);
1546 struct ll_sb_info *sbi = ll_i2sbi(inode);
1547 struct ll_file_data *fd = file->private_data;
1548 struct range_lock range;
1552 unsigned int retried = 0, dio_lock = 0;
1553 bool is_aio = false;
1554 struct cl_dio_aio *ci_aio = NULL;
1556 bool partial_io = false;
1557 size_t max_io_pages, max_cached_pages;
1561 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1562 file_dentry(file)->d_name.name,
1563 iot == CIT_READ ? "read" : "write", *ppos, count);
1565 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1566 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1567 if (max_io_pages > (max_cached_pages >> 2))
1568 max_io_pages = max_cached_pages >> 2;
1570 io = vvp_env_thread_io(env);
1571 if (file->f_flags & O_DIRECT) {
1572 if (!is_sync_kiocb(args->u.normal.via_iocb))
1574 ci_aio = cl_aio_alloc(args->u.normal.via_iocb);
1576 GOTO(out, rc = -ENOMEM);
1581 * IO block size need be aware of cached page limit, otherwise
1582 * if we have small max_cached_mb but large block IO issued, io
1583 * could not be finished and blocked whole client.
1585 if (file->f_flags & O_DIRECT)
1588 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1589 partial_io = per_bytes < count;
1590 io = vvp_env_thread_io(env);
1591 ll_io_init(io, file, iot, args);
1592 io->ci_aio = ci_aio;
1593 io->ci_dio_lock = dio_lock;
1594 io->ci_ndelay_tried = retried;
1596 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1597 bool range_locked = false;
1599 if (file->f_flags & O_APPEND)
1600 range_lock_init(&range, 0, LUSTRE_EOF);
1602 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1604 vio->vui_fd = file->private_data;
1605 vio->vui_iter = args->u.normal.via_iter;
1606 vio->vui_iocb = args->u.normal.via_iocb;
1607 /* Direct IO reads must also take range lock,
1608 * or multiple reads will try to work on the same pages
1609 * See LU-6227 for details.
1611 if (((iot == CIT_WRITE) ||
1612 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1613 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1614 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1616 rc = range_lock(&lli->lli_write_tree, &range);
1620 range_locked = true;
1623 ll_cl_add(file, env, io, LCC_RW);
1624 rc = cl_io_loop(env, io);
1625 ll_cl_remove(file, env);
1628 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1630 range_unlock(&lli->lli_write_tree, &range);
1633 /* cl_io_rw_init() handled IO */
1638 * In order to move forward AIO, ci_nob was increased,
1639 * but that doesn't mean io have been finished, it just
1640 * means io have been submited, we will always return
1641 * EIOCBQUEUED to the caller, So we could only return
1642 * number of bytes in non-AIO case.
1644 if (io->ci_nob > 0) {
1646 result += io->ci_nob;
1647 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1649 count -= io->ci_nob;
1651 /* prepare IO restart */
1653 args->u.normal.via_iter = vio->vui_iter;
1657 * Reexpand iov count because it was zero
1660 iov_iter_reexpand(vio->vui_iter, count);
1661 if (per_bytes == io->ci_nob)
1662 io->ci_need_restart = 1;
1666 cl_io_fini(env, io);
1669 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1670 file->f_path.dentry->d_name.name,
1671 iot, rc, result, io->ci_need_restart);
1673 if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
1674 count > 0 && io->ci_need_restart) {
1676 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1677 file_dentry(file)->d_name.name,
1678 iot == CIT_READ ? "read" : "write",
1679 *ppos, count, result, rc);
1680 /* preserve the tried count for FLR */
1681 retried = io->ci_ndelay_tried;
1682 dio_lock = io->ci_dio_lock;
1688 * VFS will call aio_complete() if no -EIOCBQUEUED
1689 * is returned for AIO, so we can not call aio_complete()
1692 if (rc != -EIOCBQUEUED)
1693 io->ci_aio->cda_no_aio_complete = 1;
1695 * Drop one extra reference so that end_io() could be
1696 * called for this IO context, we could call it after
1697 * we make sure all AIO requests have been proceed.
1699 cl_sync_io_note(env, &io->ci_aio->cda_sync,
1700 rc == -EIOCBQUEUED ? 0 : rc);
1702 cl_aio_free(io->ci_aio);
1707 if (iot == CIT_READ) {
1709 ll_stats_ops_tally(ll_i2sbi(inode),
1710 LPROC_LL_READ_BYTES, result);
1711 } else if (iot == CIT_WRITE) {
1713 ll_stats_ops_tally(ll_i2sbi(inode),
1714 LPROC_LL_WRITE_BYTES, result);
1715 fd->fd_write_failed = false;
1716 } else if (result == 0 && rc == 0) {
1719 fd->fd_write_failed = true;
1721 fd->fd_write_failed = false;
1722 } else if (rc != -ERESTARTSYS) {
1723 fd->fd_write_failed = true;
1727 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1729 ll_heat_add(inode, iot, result);
1731 RETURN(result > 0 ? result : rc);
1735 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1736 * especially for small I/O.
1738 * To serve a read request, CLIO has to create and initialize a cl_io and
1739 * then request DLM lock. This has turned out to have siginificant overhead
1740 * and affects the performance of small I/O dramatically.
1742 * It's not necessary to create a cl_io for each I/O. Under the help of read
1743 * ahead, most of the pages being read are already in memory cache and we can
1744 * read those pages directly because if the pages exist, the corresponding DLM
1745 * lock must exist so that page content must be valid.
1747 * In fast read implementation, the llite speculatively finds and reads pages
1748 * in memory cache. There are three scenarios for fast read:
1749 * - If the page exists and is uptodate, kernel VM will provide the data and
1750 * CLIO won't be intervened;
1751 * - If the page was brought into memory by read ahead, it will be exported
1752 * and read ahead parameters will be updated;
1753 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1754 * it will go back and invoke normal read, i.e., a cl_io will be created
1755 * and DLM lock will be requested.
1757 * POSIX compliance: posix standard states that read is intended to be atomic.
1758 * Lustre read implementation is in line with Linux kernel read implementation
1759 * and neither of them complies with POSIX standard in this matter. Fast read
1760 * doesn't make the situation worse on single node but it may interleave write
1761 * results from multiple nodes due to short read handling in ll_file_aio_read().
1763 * \param env - lu_env
1764 * \param iocb - kiocb from kernel
1765 * \param iter - user space buffers where the data will be copied
1767 * \retval - number of bytes have been read, or error code if error occurred.
1770 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1774 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1777 /* NB: we can't do direct IO for fast read because it will need a lock
1778 * to make IO engine happy. */
1779 if (iocb->ki_filp->f_flags & O_DIRECT)
1782 result = generic_file_read_iter(iocb, iter);
1784 /* If the first page is not in cache, generic_file_aio_read() will be
1785 * returned with -ENODATA.
1786 * See corresponding code in ll_readpage(). */
1787 if (result == -ENODATA)
1791 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1792 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1793 LPROC_LL_READ_BYTES, result);
1800 * Read from a file (through the page cache).
1802 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1805 struct vvp_io_args *args;
1806 struct file *file = iocb->ki_filp;
1810 ktime_t kstart = ktime_get();
1813 if (!iov_iter_count(to))
1817 * Currently when PCC read failed, we do not fall back to the
1818 * normal read path, just return the error.
1819 * The resaon is that: for RW-PCC, the file data may be modified
1820 * in the PCC and inconsistent with the data on OSTs (or file
1821 * data has been removed from the Lustre file system), at this
1822 * time, fallback to the normal read path may read the wrong
1824 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1825 * path: read data from data copy on OSTs.
1827 result = pcc_file_read_iter(iocb, to, &cached);
1831 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
1833 result = ll_do_fast_read(iocb, to);
1834 if (result < 0 || iov_iter_count(to) == 0)
1837 env = cl_env_get(&refcheck);
1839 return PTR_ERR(env);
1841 args = ll_env_args(env);
1842 args->u.normal.via_iter = to;
1843 args->u.normal.via_iocb = iocb;
1845 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
1846 &iocb->ki_pos, iov_iter_count(to));
1849 else if (result == 0)
1852 cl_env_put(env, &refcheck);
1855 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1856 file->private_data, iocb->ki_pos, result,
1858 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
1859 ktime_us_delta(ktime_get(), kstart));
1866 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
1867 * If a page is already in the page cache and dirty (and some other things -
1868 * See ll_tiny_write_begin for the instantiation of these rules), then we can
1869 * write to it without doing a full I/O, because Lustre already knows about it
1870 * and will write it out. This saves a lot of processing time.
1872 * All writes here are within one page, so exclusion is handled by the page
1873 * lock on the vm page. We do not do tiny writes for writes which touch
1874 * multiple pages because it's very unlikely multiple sequential pages are
1875 * are already dirty.
1877 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
1878 * and are unlikely to be to already dirty pages.
1880 * Attribute updates are important here, we do them in ll_tiny_write_end.
1882 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
1884 ssize_t count = iov_iter_count(iter);
1885 struct file *file = iocb->ki_filp;
1886 struct inode *inode = file_inode(file);
1887 bool lock_inode = !IS_NOSEC(inode);
1892 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
1893 * of function for why.
1895 if (count >= PAGE_SIZE ||
1896 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
1899 if (unlikely(lock_inode))
1901 result = __generic_file_write_iter(iocb, iter);
1903 if (unlikely(lock_inode))
1904 inode_unlock(inode);
1906 /* If the page is not already dirty, ll_tiny_write_begin returns
1907 * -ENODATA. We continue on to normal write.
1909 if (result == -ENODATA)
1913 ll_heat_add(inode, CIT_WRITE, result);
1914 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
1916 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
1919 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
1925 * Write to a file (through the page cache).
1927 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1929 struct vvp_io_args *args;
1931 ssize_t rc_tiny = 0, rc_normal;
1932 struct file *file = iocb->ki_filp;
1935 ktime_t kstart = ktime_get();
1940 if (!iov_iter_count(from))
1941 GOTO(out, rc_normal = 0);
1944 * When PCC write failed, we usually do not fall back to the normal
1945 * write path, just return the error. But there is a special case when
1946 * returned error code is -ENOSPC due to running out of space on PCC HSM
1947 * bakcend. At this time, it will fall back to normal I/O path and
1948 * retry the I/O. As the file is in HSM released state, it will restore
1949 * the file data to OSTs first and redo the write again. And the
1950 * restore process will revoke the layout lock and detach the file
1951 * from PCC cache automatically.
1953 result = pcc_file_write_iter(iocb, from, &cached);
1954 if (cached && result != -ENOSPC && result != -EDQUOT)
1955 GOTO(out, rc_normal = result);
1957 /* NB: we can't do direct IO for tiny writes because they use the page
1958 * cache, we can't do sync writes because tiny writes can't flush
1959 * pages, and we can't do append writes because we can't guarantee the
1960 * required DLM locks are held to protect file size.
1962 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
1963 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
1964 rc_tiny = ll_do_tiny_write(iocb, from);
1966 /* In case of error, go on and try normal write - Only stop if tiny
1967 * write completed I/O.
1969 if (iov_iter_count(from) == 0)
1970 GOTO(out, rc_normal = rc_tiny);
1972 env = cl_env_get(&refcheck);
1974 return PTR_ERR(env);
1976 args = ll_env_args(env);
1977 args->u.normal.via_iter = from;
1978 args->u.normal.via_iocb = iocb;
1980 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
1981 &iocb->ki_pos, iov_iter_count(from));
1983 /* On success, combine bytes written. */
1984 if (rc_tiny >= 0 && rc_normal > 0)
1985 rc_normal += rc_tiny;
1986 /* On error, only return error from normal write if tiny write did not
1987 * write any bytes. Otherwise return bytes written by tiny write.
1989 else if (rc_tiny > 0)
1990 rc_normal = rc_tiny;
1992 cl_env_put(env, &refcheck);
1994 if (rc_normal > 0) {
1995 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1996 file->private_data, iocb->ki_pos,
1998 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
1999 ktime_us_delta(ktime_get(), kstart));
2005 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2007 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2009 static int ll_file_get_iov_count(const struct iovec *iov,
2010 unsigned long *nr_segs, size_t *count,
2016 for (seg = 0; seg < *nr_segs; seg++) {
2017 const struct iovec *iv = &iov[seg];
2020 * If any segment has a negative length, or the cumulative
2021 * length ever wraps negative then return -EINVAL.
2024 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2026 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2031 cnt -= iv->iov_len; /* This segment is no good */
2038 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2039 unsigned long nr_segs, loff_t pos)
2046 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2053 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2054 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2055 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2056 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2057 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2059 result = ll_file_read_iter(iocb, &to);
2064 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2067 struct iovec iov = { .iov_base = buf, .iov_len = count };
2076 init_sync_kiocb(&kiocb, file);
2077 kiocb.ki_pos = *ppos;
2078 #ifdef HAVE_KIOCB_KI_LEFT
2079 kiocb.ki_left = count;
2080 #elif defined(HAVE_KI_NBYTES)
2081 kiocb.i_nbytes = count;
2084 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2085 *ppos = kiocb.ki_pos;
2091 * Write to a file (through the page cache).
2094 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2095 unsigned long nr_segs, loff_t pos)
2097 struct iov_iter from;
2102 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2109 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2110 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2111 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2112 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2113 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2115 result = ll_file_write_iter(iocb, &from);
2120 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2121 size_t count, loff_t *ppos)
2123 struct iovec iov = { .iov_base = (void __user *)buf,
2133 init_sync_kiocb(&kiocb, file);
2134 kiocb.ki_pos = *ppos;
2135 #ifdef HAVE_KIOCB_KI_LEFT
2136 kiocb.ki_left = count;
2137 #elif defined(HAVE_KI_NBYTES)
2138 kiocb.ki_nbytes = count;
2141 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2142 *ppos = kiocb.ki_pos;
2146 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2148 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2149 __u64 flags, struct lov_user_md *lum, int lum_size)
2151 struct lookup_intent oit = {
2153 .it_flags = flags | MDS_OPEN_BY_FID,
2158 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2159 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2160 /* this code will only exist for big-endian systems */
2161 lustre_swab_lov_user_md(lum, 0);
2164 ll_inode_size_lock(inode);
2165 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2167 GOTO(out_unlock, rc);
2169 ll_release_openhandle(dentry, &oit);
2172 ll_inode_size_unlock(inode);
2173 ll_intent_release(&oit);
2178 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2179 struct lov_mds_md **lmmp, int *lmm_size,
2180 struct ptlrpc_request **request)
2182 struct ll_sb_info *sbi = ll_i2sbi(inode);
2183 struct mdt_body *body;
2184 struct lov_mds_md *lmm = NULL;
2185 struct ptlrpc_request *req = NULL;
2186 struct md_op_data *op_data;
2191 rc = ll_get_default_mdsize(sbi, &lmmsize);
2195 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2196 strlen(filename), lmmsize,
2197 LUSTRE_OPC_ANY, NULL);
2198 if (IS_ERR(op_data))
2199 RETURN(PTR_ERR(op_data));
2201 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2202 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2203 ll_finish_md_op_data(op_data);
2205 CDEBUG(D_INFO, "md_getattr_name failed "
2206 "on %s: rc %d\n", filename, rc);
2210 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2211 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2213 lmmsize = body->mbo_eadatasize;
2215 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2217 GOTO(out, rc = -ENODATA);
2219 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2220 LASSERT(lmm != NULL);
2222 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2223 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2224 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2225 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2226 GOTO(out, rc = -EPROTO);
2229 * This is coming from the MDS, so is probably in
2230 * little endian. We convert it to host endian before
2231 * passing it to userspace.
2233 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2234 int stripe_count = 0;
2236 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2237 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2238 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2239 if (le32_to_cpu(lmm->lmm_pattern) &
2240 LOV_PATTERN_F_RELEASED)
2242 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2244 /* if function called for directory - we should
2245 * avoid swab not existent lsm objects
2247 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2248 S_ISREG(body->mbo_mode))
2249 lustre_swab_lov_user_md_objects(
2250 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2252 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2253 S_ISREG(body->mbo_mode))
2254 lustre_swab_lov_user_md_objects(
2255 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2257 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2258 lustre_swab_lov_comp_md_v1(
2259 (struct lov_comp_md_v1 *)lmm);
2263 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2264 struct lov_comp_md_v1 *comp_v1 = NULL;
2265 struct lov_comp_md_entry_v1 *ent;
2266 struct lov_user_md_v1 *v1;
2270 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2271 /* Dump the striping information */
2272 for (; i < comp_v1->lcm_entry_count; i++) {
2273 ent = &comp_v1->lcm_entries[i];
2274 off = ent->lcme_offset;
2275 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2277 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2278 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2282 * Return valid stripe_count and stripe_size instead of 0 for
2283 * DoM files to avoid divide-by-zero for older userspace that
2284 * calls this ioctl, e.g. lustre ADIO driver.
2286 if (lmm->lmm_stripe_count == 0)
2287 lmm->lmm_stripe_count = 1;
2288 if (lmm->lmm_stripe_size == 0) {
2289 /* Since the first component of the file data is placed
2290 * on the MDT for faster access, the stripe_size of the
2291 * second one is always that applications which are
2294 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2295 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2297 i = comp_v1->lcm_entry_count > 1 ?
2298 comp_v1->lcm_entry_count - 1 : 0;
2299 ent = &comp_v1->lcm_entries[i];
2300 off = ent->lcme_offset;
2301 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2302 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2307 *lmm_size = lmmsize;
2312 static int ll_lov_setea(struct inode *inode, struct file *file,
2315 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2316 struct lov_user_md *lump;
2317 int lum_size = sizeof(struct lov_user_md) +
2318 sizeof(struct lov_user_ost_data);
2322 if (!capable(CAP_SYS_ADMIN))
2325 OBD_ALLOC_LARGE(lump, lum_size);
2329 if (copy_from_user(lump, arg, lum_size))
2330 GOTO(out_lump, rc = -EFAULT);
2332 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2334 cl_lov_delay_create_clear(&file->f_flags);
2337 OBD_FREE_LARGE(lump, lum_size);
2341 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2348 env = cl_env_get(&refcheck);
2350 RETURN(PTR_ERR(env));
2352 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2353 cl_env_put(env, &refcheck);
2357 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2360 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2361 struct lov_user_md *klum;
2363 __u64 flags = FMODE_WRITE;
2366 rc = ll_copy_user_md(lum, &klum);
2371 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2376 rc = put_user(0, &lum->lmm_stripe_count);
2380 rc = ll_layout_refresh(inode, &gen);
2384 rc = ll_file_getstripe(inode, arg, lum_size);
2385 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2386 ll_i2info(inode)->lli_clob) {
2387 struct iattr attr = { 0 };
2389 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2390 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2393 cl_lov_delay_create_clear(&file->f_flags);
2396 OBD_FREE_LARGE(klum, lum_size);
2402 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2404 struct ll_inode_info *lli = ll_i2info(inode);
2405 struct cl_object *obj = lli->lli_clob;
2406 struct ll_file_data *fd = file->private_data;
2407 struct ll_grouplock grouplock;
2412 CWARN("group id for group lock must not be 0\n");
2416 if (ll_file_nolock(file))
2417 RETURN(-EOPNOTSUPP);
2419 if (file->f_flags & O_NONBLOCK) {
2420 if (!mutex_trylock(&lli->lli_group_mutex))
2423 mutex_lock(&lli->lli_group_mutex);
2425 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2426 CWARN("group lock already existed with gid %lu\n",
2427 fd->fd_grouplock.lg_gid);
2428 GOTO(out, rc = -EINVAL);
2430 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2431 if (file->f_flags & O_NONBLOCK)
2432 GOTO(out, rc = -EAGAIN);
2433 mutex_unlock(&lli->lli_group_mutex);
2434 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2435 GOTO(retry, rc = 0);
2437 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2440 * XXX: group lock needs to protect all OST objects while PFL
2441 * can add new OST objects during the IO, so we'd instantiate
2442 * all OST objects before getting its group lock.
2447 struct cl_layout cl = {
2448 .cl_is_composite = false,
2450 struct lu_extent ext = {
2452 .e_end = OBD_OBJECT_EOF,
2455 env = cl_env_get(&refcheck);
2457 GOTO(out, rc = PTR_ERR(env));
2459 rc = cl_object_layout_get(env, obj, &cl);
2460 if (rc >= 0 && cl.cl_is_composite)
2461 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2464 cl_env_put(env, &refcheck);
2469 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2470 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2475 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2476 fd->fd_grouplock = grouplock;
2477 if (lli->lli_group_users == 0)
2478 lli->lli_group_gid = grouplock.lg_gid;
2479 lli->lli_group_users++;
2481 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2483 mutex_unlock(&lli->lli_group_mutex);
2488 static int ll_put_grouplock(struct inode *inode, struct file *file,
2491 struct ll_inode_info *lli = ll_i2info(inode);
2492 struct ll_file_data *fd = file->private_data;
2493 struct ll_grouplock grouplock;
2497 mutex_lock(&lli->lli_group_mutex);
2498 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2499 CWARN("no group lock held\n");
2500 GOTO(out, rc = -EINVAL);
2503 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2505 if (fd->fd_grouplock.lg_gid != arg) {
2506 CWARN("group lock %lu doesn't match current id %lu\n",
2507 arg, fd->fd_grouplock.lg_gid);
2508 GOTO(out, rc = -EINVAL);
2511 grouplock = fd->fd_grouplock;
2512 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2513 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2515 cl_put_grouplock(&grouplock);
2517 lli->lli_group_users--;
2518 if (lli->lli_group_users == 0) {
2519 lli->lli_group_gid = 0;
2520 wake_up_var(&lli->lli_group_users);
2522 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2525 mutex_unlock(&lli->lli_group_mutex);
2531 * Close inode open handle
2533 * \param dentry [in] dentry which contains the inode
2534 * \param it [in,out] intent which contains open info and result
2537 * \retval <0 failure
2539 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2541 struct inode *inode = dentry->d_inode;
2542 struct obd_client_handle *och;
2548 /* Root ? Do nothing. */
2549 if (is_root_inode(inode))
2552 /* No open handle to close? Move away */
2553 if (!it_disposition(it, DISP_OPEN_OPEN))
2556 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2558 OBD_ALLOC(och, sizeof(*och));
2560 GOTO(out, rc = -ENOMEM);
2562 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2566 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2568 /* this one is in place of ll_file_open */
2569 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2570 ptlrpc_req_finished(it->it_request);
2571 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2577 * Get size for inode for which FIEMAP mapping is requested.
2578 * Make the FIEMAP get_info call and returns the result.
2579 * \param fiemap kernel buffer to hold extens
2580 * \param num_bytes kernel buffer size
2582 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2588 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2591 /* Checks for fiemap flags */
2592 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2593 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2597 /* Check for FIEMAP_FLAG_SYNC */
2598 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2599 rc = filemap_fdatawrite(inode->i_mapping);
2604 env = cl_env_get(&refcheck);
2606 RETURN(PTR_ERR(env));
2608 if (i_size_read(inode) == 0) {
2609 rc = ll_glimpse_size(inode);
2614 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2615 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2616 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2618 /* If filesize is 0, then there would be no objects for mapping */
2619 if (fmkey.lfik_oa.o_size == 0) {
2620 fiemap->fm_mapped_extents = 0;
2624 fmkey.lfik_fiemap = *fiemap;
2626 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2627 &fmkey, fiemap, &num_bytes);
2629 cl_env_put(env, &refcheck);
2633 int ll_fid2path(struct inode *inode, void __user *arg)
2635 struct obd_export *exp = ll_i2mdexp(inode);
2636 const struct getinfo_fid2path __user *gfin = arg;
2638 struct getinfo_fid2path *gfout;
2644 if (!capable(CAP_DAC_READ_SEARCH) &&
2645 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2648 /* Only need to get the buflen */
2649 if (get_user(pathlen, &gfin->gf_pathlen))
2652 if (pathlen > PATH_MAX)
2655 outsize = sizeof(*gfout) + pathlen;
2656 OBD_ALLOC(gfout, outsize);
2660 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2661 GOTO(gf_free, rc = -EFAULT);
2662 /* append root FID after gfout to let MDT know the root FID so that it
2663 * can lookup the correct path, this is mainly for fileset.
2664 * old server without fileset mount support will ignore this. */
2665 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2667 /* Call mdc_iocontrol */
2668 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2672 if (copy_to_user(arg, gfout, outsize))
2676 OBD_FREE(gfout, outsize);
2681 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2683 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2691 ioc->idv_version = 0;
2692 ioc->idv_layout_version = UINT_MAX;
2694 /* If no file object initialized, we consider its version is 0. */
2698 env = cl_env_get(&refcheck);
2700 RETURN(PTR_ERR(env));
2702 io = vvp_env_thread_io(env);
2704 io->u.ci_data_version.dv_data_version = 0;
2705 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2706 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2709 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2710 result = cl_io_loop(env, io);
2712 result = io->ci_result;
2714 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2715 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2717 cl_io_fini(env, io);
2719 if (unlikely(io->ci_need_restart))
2722 cl_env_put(env, &refcheck);
2728 * Read the data_version for inode.
2730 * This value is computed using stripe object version on OST.
2731 * Version is computed using server side locking.
2733 * @param flags if do sync on the OST side;
2735 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2736 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2738 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2740 struct ioc_data_version ioc = { .idv_flags = flags };
2743 rc = ll_ioc_data_version(inode, &ioc);
2745 *data_version = ioc.idv_version;
2751 * Trigger a HSM release request for the provided inode.
2753 int ll_hsm_release(struct inode *inode)
2756 struct obd_client_handle *och = NULL;
2757 __u64 data_version = 0;
2762 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2763 ll_i2sbi(inode)->ll_fsname,
2764 PFID(&ll_i2info(inode)->lli_fid));
2766 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2768 GOTO(out, rc = PTR_ERR(och));
2770 /* Grab latest data_version and [am]time values */
2771 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2775 env = cl_env_get(&refcheck);
2777 GOTO(out, rc = PTR_ERR(env));
2779 rc = ll_merge_attr(env, inode);
2780 cl_env_put(env, &refcheck);
2782 /* If error happen, we have the wrong size for a file.
2788 /* Release the file.
2789 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2790 * we still need it to pack l_remote_handle to MDT. */
2791 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2797 if (och != NULL && !IS_ERR(och)) /* close the file */
2798 ll_lease_close(och, inode, NULL);
2803 struct ll_swap_stack {
2806 struct inode *inode1;
2807 struct inode *inode2;
2812 static int ll_swap_layouts(struct file *file1, struct file *file2,
2813 struct lustre_swap_layouts *lsl)
2815 struct mdc_swap_layouts msl;
2816 struct md_op_data *op_data;
2819 struct ll_swap_stack *llss = NULL;
2822 OBD_ALLOC_PTR(llss);
2826 llss->inode1 = file_inode(file1);
2827 llss->inode2 = file_inode(file2);
2829 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2833 /* we use 2 bool because it is easier to swap than 2 bits */
2834 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2835 llss->check_dv1 = true;
2837 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2838 llss->check_dv2 = true;
2840 /* we cannot use lsl->sl_dvX directly because we may swap them */
2841 llss->dv1 = lsl->sl_dv1;
2842 llss->dv2 = lsl->sl_dv2;
2844 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2845 if (rc == 0) /* same file, done! */
2848 if (rc < 0) { /* sequentialize it */
2849 swap(llss->inode1, llss->inode2);
2851 swap(llss->dv1, llss->dv2);
2852 swap(llss->check_dv1, llss->check_dv2);
2856 if (gid != 0) { /* application asks to flush dirty cache */
2857 rc = ll_get_grouplock(llss->inode1, file1, gid);
2861 rc = ll_get_grouplock(llss->inode2, file2, gid);
2863 ll_put_grouplock(llss->inode1, file1, gid);
2868 /* ultimate check, before swaping the layouts we check if
2869 * dataversion has changed (if requested) */
2870 if (llss->check_dv1) {
2871 rc = ll_data_version(llss->inode1, &dv, 0);
2874 if (dv != llss->dv1)
2875 GOTO(putgl, rc = -EAGAIN);
2878 if (llss->check_dv2) {
2879 rc = ll_data_version(llss->inode2, &dv, 0);
2882 if (dv != llss->dv2)
2883 GOTO(putgl, rc = -EAGAIN);
2886 /* struct md_op_data is used to send the swap args to the mdt
2887 * only flags is missing, so we use struct mdc_swap_layouts
2888 * through the md_op_data->op_data */
2889 /* flags from user space have to be converted before they are send to
2890 * server, no flag is sent today, they are only used on the client */
2893 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2894 0, LUSTRE_OPC_ANY, &msl);
2895 if (IS_ERR(op_data))
2896 GOTO(free, rc = PTR_ERR(op_data));
2898 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2899 sizeof(*op_data), op_data, NULL);
2900 ll_finish_md_op_data(op_data);
2907 ll_put_grouplock(llss->inode2, file2, gid);
2908 ll_put_grouplock(llss->inode1, file1, gid);
2918 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2920 struct obd_export *exp = ll_i2mdexp(inode);
2921 struct md_op_data *op_data;
2925 /* Detect out-of range masks */
2926 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2929 /* Non-root users are forbidden to set or clear flags which are
2930 * NOT defined in HSM_USER_MASK. */
2931 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2932 !capable(CAP_SYS_ADMIN))
2935 if (!exp_connect_archive_id_array(exp)) {
2936 /* Detect out-of range archive id */
2937 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2938 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
2942 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2943 LUSTRE_OPC_ANY, hss);
2944 if (IS_ERR(op_data))
2945 RETURN(PTR_ERR(op_data));
2947 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
2950 ll_finish_md_op_data(op_data);
2955 static int ll_hsm_import(struct inode *inode, struct file *file,
2956 struct hsm_user_import *hui)
2958 struct hsm_state_set *hss = NULL;
2959 struct iattr *attr = NULL;
2963 if (!S_ISREG(inode->i_mode))
2969 GOTO(out, rc = -ENOMEM);
2971 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2972 hss->hss_archive_id = hui->hui_archive_id;
2973 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2974 rc = ll_hsm_state_set(inode, hss);
2978 OBD_ALLOC_PTR(attr);
2980 GOTO(out, rc = -ENOMEM);
2982 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2983 attr->ia_mode |= S_IFREG;
2984 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2985 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2986 attr->ia_size = hui->hui_size;
2987 attr->ia_mtime.tv_sec = hui->hui_mtime;
2988 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2989 attr->ia_atime.tv_sec = hui->hui_atime;
2990 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2992 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2993 ATTR_UID | ATTR_GID |
2994 ATTR_MTIME | ATTR_MTIME_SET |
2995 ATTR_ATIME | ATTR_ATIME_SET;
2999 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3003 inode_unlock(inode);
3015 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3017 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3018 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3021 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3023 struct inode *inode = file_inode(file);
3025 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3026 ATTR_MTIME | ATTR_MTIME_SET |
3029 .tv_sec = lfu->lfu_atime_sec,
3030 .tv_nsec = lfu->lfu_atime_nsec,
3033 .tv_sec = lfu->lfu_mtime_sec,
3034 .tv_nsec = lfu->lfu_mtime_nsec,
3037 .tv_sec = lfu->lfu_ctime_sec,
3038 .tv_nsec = lfu->lfu_ctime_nsec,
3044 if (!capable(CAP_SYS_ADMIN))
3047 if (!S_ISREG(inode->i_mode))
3051 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3053 inode_unlock(inode);
3058 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3061 case MODE_READ_USER:
3063 case MODE_WRITE_USER:
3070 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3072 /* Used to allow the upper layers of the client to request an LDLM lock
3073 * without doing an actual read or write.
3075 * Used for ladvise lockahead to manually request specific locks.
3077 * \param[in] file file this ladvise lock request is on
3078 * \param[in] ladvise ladvise struct describing this lock request
3080 * \retval 0 success, no detailed result available (sync requests
3081 * and requests sent to the server [not handled locally]
3082 * cannot return detailed results)
3083 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3084 * see definitions for details.
3085 * \retval negative negative errno on error
3087 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3089 struct lu_env *env = NULL;
3090 struct cl_io *io = NULL;
3091 struct cl_lock *lock = NULL;
3092 struct cl_lock_descr *descr = NULL;
3093 struct dentry *dentry = file->f_path.dentry;
3094 struct inode *inode = dentry->d_inode;
3095 enum cl_lock_mode cl_mode;
3096 off_t start = ladvise->lla_start;
3097 off_t end = ladvise->lla_end;
3104 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3105 dentry, dentry->d_inode,
3106 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3109 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3111 GOTO(out, result = cl_mode);
3113 /* Get IO environment */
3114 result = cl_io_get(inode, &env, &io, &refcheck);
3118 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3121 * nothing to do for this io. This currently happens when
3122 * stripe sub-object's are not yet created.
3124 result = io->ci_result;
3125 } else if (result == 0) {
3126 lock = vvp_env_lock(env);
3127 descr = &lock->cll_descr;
3129 descr->cld_obj = io->ci_obj;
3130 /* Convert byte offsets to pages */
3131 descr->cld_start = cl_index(io->ci_obj, start);
3132 descr->cld_end = cl_index(io->ci_obj, end);
3133 descr->cld_mode = cl_mode;
3134 /* CEF_MUST is used because we do not want to convert a
3135 * lockahead request to a lockless lock */
3136 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND |
3139 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3140 descr->cld_enq_flags |= CEF_SPECULATIVE;
3142 result = cl_lock_request(env, io, lock);
3144 /* On success, we need to release the lock */
3146 cl_lock_release(env, lock);
3148 cl_io_fini(env, io);
3149 cl_env_put(env, &refcheck);
3151 /* -ECANCELED indicates a matching lock with a different extent
3152 * was already present, and -EEXIST indicates a matching lock
3153 * on exactly the same extent was already present.
3154 * We convert them to positive values for userspace to make
3155 * recognizing true errors easier.
3156 * Note we can only return these detailed results on async requests,
3157 * as sync requests look the same as i/o requests for locking. */
3158 if (result == -ECANCELED)
3159 result = LLA_RESULT_DIFFERENT;
3160 else if (result == -EEXIST)
3161 result = LLA_RESULT_SAME;
3166 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3168 static int ll_ladvise_sanity(struct inode *inode,
3169 struct llapi_lu_ladvise *ladvise)
3171 struct ll_sb_info *sbi = ll_i2sbi(inode);
3172 enum lu_ladvise_type advice = ladvise->lla_advice;
3173 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3174 * be in the first 32 bits of enum ladvise_flags */
3175 __u32 flags = ladvise->lla_peradvice_flags;
3176 /* 3 lines at 80 characters per line, should be plenty */
3179 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3182 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3183 sbi->ll_fsname, advice,
3184 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3188 /* Per-advice checks */
3190 case LU_LADVISE_LOCKNOEXPAND:
3191 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3193 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3194 "rc = %d\n", sbi->ll_fsname, flags,
3195 ladvise_names[advice], rc);
3199 case LU_LADVISE_LOCKAHEAD:
3200 /* Currently only READ and WRITE modes can be requested */
3201 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3202 ladvise->lla_lockahead_mode == 0) {
3204 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3205 "rc = %d\n", sbi->ll_fsname,
3206 ladvise->lla_lockahead_mode,
3207 ladvise_names[advice], rc);
3211 case LU_LADVISE_WILLREAD:
3212 case LU_LADVISE_DONTNEED:
3214 /* Note fall through above - These checks apply to all advices
3215 * except LOCKNOEXPAND */
3216 if (flags & ~LF_DEFAULT_MASK) {
3218 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3219 "rc = %d\n", sbi->ll_fsname, flags,
3220 ladvise_names[advice], rc);
3223 if (ladvise->lla_start >= ladvise->lla_end) {
3225 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3226 "for %s: rc = %d\n", sbi->ll_fsname,
3227 ladvise->lla_start, ladvise->lla_end,
3228 ladvise_names[advice], rc);
3240 * Give file access advices
3242 * The ladvise interface is similar to Linux fadvise() system call, except it
3243 * forwards the advices directly from Lustre client to server. The server side
3244 * codes will apply appropriate read-ahead and caching techniques for the
3245 * corresponding files.
3247 * A typical workload for ladvise is e.g. a bunch of different clients are
3248 * doing small random reads of a file, so prefetching pages into OSS cache
3249 * with big linear reads before the random IO is a net benefit. Fetching
3250 * all that data into each client cache with fadvise() may not be, due to
3251 * much more data being sent to the client.
3253 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3254 struct llapi_lu_ladvise *ladvise)
3258 struct cl_ladvise_io *lio;
3263 env = cl_env_get(&refcheck);
3265 RETURN(PTR_ERR(env));
3267 io = vvp_env_thread_io(env);
3268 io->ci_obj = ll_i2info(inode)->lli_clob;
3270 /* initialize parameters for ladvise */
3271 lio = &io->u.ci_ladvise;
3272 lio->li_start = ladvise->lla_start;
3273 lio->li_end = ladvise->lla_end;
3274 lio->li_fid = ll_inode2fid(inode);
3275 lio->li_advice = ladvise->lla_advice;
3276 lio->li_flags = flags;
3278 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3279 rc = cl_io_loop(env, io);
3283 cl_io_fini(env, io);
3284 cl_env_put(env, &refcheck);
3288 static int ll_lock_noexpand(struct file *file, int flags)
3290 struct ll_file_data *fd = file->private_data;
3292 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3297 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3300 struct fsxattr fsxattr;
3302 if (copy_from_user(&fsxattr,
3303 (const struct fsxattr __user *)arg,
3307 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3308 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3309 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3310 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3311 if (copy_to_user((struct fsxattr __user *)arg,
3312 &fsxattr, sizeof(fsxattr)))
3318 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa)
3321 * Project Quota ID state is only allowed to change from within the init
3322 * namespace. Enforce that restriction only if we are trying to change
3323 * the quota ID state. Everything else is allowed in user namespaces.
3325 if (current_user_ns() == &init_user_ns)
3328 if (ll_i2info(inode)->lli_projid != fa->fsx_projid)
3331 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3332 if (!(fa->fsx_xflags & FS_XFLAG_PROJINHERIT))
3335 if (fa->fsx_xflags & FS_XFLAG_PROJINHERIT)
3342 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3345 struct md_op_data *op_data;
3346 struct ptlrpc_request *req = NULL;
3347 struct fsxattr fsxattr;
3348 struct cl_object *obj;
3349 unsigned int inode_flags;
3352 if (copy_from_user(&fsxattr,
3353 (const struct fsxattr __user *)arg,
3357 rc = ll_ioctl_check_project(inode, &fsxattr);
3361 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3362 LUSTRE_OPC_ANY, NULL);
3363 if (IS_ERR(op_data))
3364 RETURN(PTR_ERR(op_data));
3366 inode_flags = ll_xflags_to_inode_flags(fsxattr.fsx_xflags);
3367 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3368 if (fsxattr.fsx_xflags & FS_XFLAG_PROJINHERIT)
3369 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3370 op_data->op_projid = fsxattr.fsx_projid;
3371 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3372 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3373 ptlrpc_req_finished(req);
3375 GOTO(out_fsxattr, rc);
3376 ll_update_inode_flags(inode, op_data->op_attr_flags);
3378 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3379 if (fsxattr.fsx_xflags == 0 ||
3380 fsxattr.fsx_xflags == FS_XFLAG_PROJINHERIT)
3381 GOTO(out_fsxattr, rc);
3383 obj = ll_i2info(inode)->lli_clob;
3385 struct iattr attr = { 0 };
3387 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS,
3388 fsxattr.fsx_xflags);
3392 ll_finish_md_op_data(op_data);
3396 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3399 struct inode *inode = file_inode(file);
3400 struct ll_file_data *fd = file->private_data;
3401 struct ll_inode_info *lli = ll_i2info(inode);
3402 struct obd_client_handle *och = NULL;
3403 struct split_param sp;
3404 struct pcc_param param;
3405 bool lease_broken = false;
3407 enum mds_op_bias bias = 0;
3408 struct file *layout_file = NULL;
3410 size_t data_size = 0;
3411 bool attached = false;
3416 mutex_lock(&lli->lli_och_mutex);
3417 if (fd->fd_lease_och != NULL) {
3418 och = fd->fd_lease_och;
3419 fd->fd_lease_och = NULL;
3421 mutex_unlock(&lli->lli_och_mutex);
3426 fmode = och->och_flags;
3428 switch (ioc->lil_flags) {
3429 case LL_LEASE_RESYNC_DONE:
3430 if (ioc->lil_count > IOC_IDS_MAX)
3431 GOTO(out_lease_close, rc = -EINVAL);
3433 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3434 OBD_ALLOC(data, data_size);
3436 GOTO(out_lease_close, rc = -ENOMEM);
3438 if (copy_from_user(data, (void __user *)arg, data_size))
3439 GOTO(out_lease_close, rc = -EFAULT);
3441 bias = MDS_CLOSE_RESYNC_DONE;
3443 case LL_LEASE_LAYOUT_MERGE: {
3446 if (ioc->lil_count != 1)
3447 GOTO(out_lease_close, rc = -EINVAL);
3449 arg += sizeof(*ioc);
3450 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3451 GOTO(out_lease_close, rc = -EFAULT);
3453 layout_file = fget(fd);
3455 GOTO(out_lease_close, rc = -EBADF);
3457 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3458 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3459 GOTO(out_lease_close, rc = -EPERM);
3461 data = file_inode(layout_file);
3462 bias = MDS_CLOSE_LAYOUT_MERGE;
3465 case LL_LEASE_LAYOUT_SPLIT: {
3469 if (ioc->lil_count != 2)
3470 GOTO(out_lease_close, rc = -EINVAL);
3472 arg += sizeof(*ioc);
3473 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3474 GOTO(out_lease_close, rc = -EFAULT);
3476 arg += sizeof(__u32);
3477 if (copy_from_user(&mirror_id, (void __user *)arg,
3479 GOTO(out_lease_close, rc = -EFAULT);
3481 layout_file = fget(fdv);
3483 GOTO(out_lease_close, rc = -EBADF);
3485 /* if layout_file == file, it means to destroy the mirror */
3486 sp.sp_inode = file_inode(layout_file);
3487 sp.sp_mirror_id = (__u16)mirror_id;
3489 bias = MDS_CLOSE_LAYOUT_SPLIT;
3492 case LL_LEASE_PCC_ATTACH:
3493 if (ioc->lil_count != 1)
3496 arg += sizeof(*ioc);
3497 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3499 GOTO(out_lease_close, rc2 = -EFAULT);
3501 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3503 GOTO(out_lease_close, rc2);
3506 /* Grab latest data version */
3507 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3510 GOTO(out_lease_close, rc2);
3513 bias = MDS_PCC_ATTACH;
3516 /* without close intent */
3521 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3525 rc = ll_lease_och_release(inode, file);
3534 switch (ioc->lil_flags) {
3535 case LL_LEASE_RESYNC_DONE:
3537 OBD_FREE(data, data_size);
3539 case LL_LEASE_LAYOUT_MERGE:
3540 case LL_LEASE_LAYOUT_SPLIT:
3544 case LL_LEASE_PCC_ATTACH:
3547 rc = pcc_readwrite_attach_fini(file, inode,
3548 param.pa_layout_gen,
3555 rc = ll_lease_type_from_fmode(fmode);
3559 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3562 struct inode *inode = file_inode(file);
3563 struct ll_inode_info *lli = ll_i2info(inode);
3564 struct ll_file_data *fd = file->private_data;
3565 struct obd_client_handle *och = NULL;
3566 __u64 open_flags = 0;
3572 switch (ioc->lil_mode) {
3573 case LL_LEASE_WRLCK:
3574 if (!(file->f_mode & FMODE_WRITE))
3576 fmode = FMODE_WRITE;
3578 case LL_LEASE_RDLCK:
3579 if (!(file->f_mode & FMODE_READ))
3583 case LL_LEASE_UNLCK:
3584 RETURN(ll_file_unlock_lease(file, ioc, arg));
3589 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3591 /* apply for lease */
3592 if (ioc->lil_flags & LL_LEASE_RESYNC)
3593 open_flags = MDS_OPEN_RESYNC;
3594 och = ll_lease_open(inode, file, fmode, open_flags);
3596 RETURN(PTR_ERR(och));
3598 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3599 rc = ll_lease_file_resync(och, inode, arg);
3601 ll_lease_close(och, inode, NULL);
3604 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3606 ll_lease_close(och, inode, NULL);
3612 mutex_lock(&lli->lli_och_mutex);
3613 if (fd->fd_lease_och == NULL) {
3614 fd->fd_lease_och = och;
3617 mutex_unlock(&lli->lli_och_mutex);
3619 /* impossible now that only excl is supported for now */
3620 ll_lease_close(och, inode, &lease_broken);
3626 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3628 struct ll_inode_info *lli = ll_i2info(inode);
3629 struct ll_sb_info *sbi = ll_i2sbi(inode);
3630 __u64 now = ktime_get_real_seconds();
3633 spin_lock(&lli->lli_heat_lock);
3634 heat->lh_flags = lli->lli_heat_flags;
3635 for (i = 0; i < heat->lh_count; i++)
3636 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3637 now, sbi->ll_heat_decay_weight,
3638 sbi->ll_heat_period_second);
3639 spin_unlock(&lli->lli_heat_lock);
3642 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3644 struct ll_inode_info *lli = ll_i2info(inode);
3647 spin_lock(&lli->lli_heat_lock);
3648 if (flags & LU_HEAT_FLAG_CLEAR)
3649 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3651 if (flags & LU_HEAT_FLAG_OFF)
3652 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3654 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3656 spin_unlock(&lli->lli_heat_lock);
3662 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3664 struct inode *inode = file_inode(file);
3665 struct ll_file_data *fd = file->private_data;
3669 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3670 PFID(ll_inode2fid(inode)), inode, cmd);
3671 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3673 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3674 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3678 case LL_IOC_GETFLAGS:
3679 /* Get the current value of the file flags */
3680 return put_user(fd->fd_flags, (int __user *)arg);
3681 case LL_IOC_SETFLAGS:
3682 case LL_IOC_CLRFLAGS:
3683 /* Set or clear specific file flags */
3684 /* XXX This probably needs checks to ensure the flags are
3685 * not abused, and to handle any flag side effects.
3687 if (get_user(flags, (int __user *) arg))
3690 if (cmd == LL_IOC_SETFLAGS) {
3691 if ((flags & LL_FILE_IGNORE_LOCK) &&
3692 !(file->f_flags & O_DIRECT)) {
3693 CERROR("%s: unable to disable locking on "
3694 "non-O_DIRECT file\n", current->comm);
3698 fd->fd_flags |= flags;
3700 fd->fd_flags &= ~flags;
3703 case LL_IOC_LOV_SETSTRIPE:
3704 case LL_IOC_LOV_SETSTRIPE_NEW:
3705 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3706 case LL_IOC_LOV_SETEA:
3707 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3708 case LL_IOC_LOV_SWAP_LAYOUTS: {
3710 struct lustre_swap_layouts lsl;
3712 if (copy_from_user(&lsl, (char __user *)arg,
3713 sizeof(struct lustre_swap_layouts)))
3716 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3719 file2 = fget(lsl.sl_fd);
3723 /* O_WRONLY or O_RDWR */
3724 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3725 GOTO(out, rc = -EPERM);
3727 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3728 struct inode *inode2;
3729 struct ll_inode_info *lli;
3730 struct obd_client_handle *och = NULL;
3732 lli = ll_i2info(inode);
3733 mutex_lock(&lli->lli_och_mutex);
3734 if (fd->fd_lease_och != NULL) {
3735 och = fd->fd_lease_och;
3736 fd->fd_lease_och = NULL;
3738 mutex_unlock(&lli->lli_och_mutex);
3740 GOTO(out, rc = -ENOLCK);
3741 inode2 = file_inode(file2);
3742 rc = ll_swap_layouts_close(och, inode, inode2);
3744 rc = ll_swap_layouts(file, file2, &lsl);
3750 case LL_IOC_LOV_GETSTRIPE:
3751 case LL_IOC_LOV_GETSTRIPE_NEW:
3752 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3753 case FS_IOC_GETFLAGS:
3754 case FS_IOC_SETFLAGS:
3755 RETURN(ll_iocontrol(inode, file, cmd, arg));
3756 case FSFILT_IOC_GETVERSION:
3757 case FS_IOC_GETVERSION:
3758 RETURN(put_user(inode->i_generation, (int __user *)arg));
3759 /* We need to special case any other ioctls we want to handle,
3760 * to send them to the MDS/OST as appropriate and to properly
3761 * network encode the arg field. */
3762 case FS_IOC_SETVERSION:
3765 case LL_IOC_GROUP_LOCK:
3766 RETURN(ll_get_grouplock(inode, file, arg));
3767 case LL_IOC_GROUP_UNLOCK:
3768 RETURN(ll_put_grouplock(inode, file, arg));
3769 case IOC_OBD_STATFS:
3770 RETURN(ll_obd_statfs(inode, (void __user *)arg));
3772 case LL_IOC_FLUSHCTX:
3773 RETURN(ll_flush_ctx(inode));
3774 case LL_IOC_PATH2FID: {
3775 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
3776 sizeof(struct lu_fid)))
3781 case LL_IOC_GETPARENT:
3782 RETURN(ll_getparent(file, (struct getparent __user *)arg));
3784 case OBD_IOC_FID2PATH:
3785 RETURN(ll_fid2path(inode, (void __user *)arg));
3786 case LL_IOC_DATA_VERSION: {
3787 struct ioc_data_version idv;
3790 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
3793 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
3794 rc = ll_ioc_data_version(inode, &idv);
3797 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
3803 case LL_IOC_GET_MDTIDX: {
3806 mdtidx = ll_get_mdt_idx(inode);
3810 if (put_user((int)mdtidx, (int __user *)arg))
3815 case OBD_IOC_GETNAME_OLD:
3817 case OBD_IOC_GETDTNAME:
3819 case OBD_IOC_GETMDNAME:
3820 RETURN(ll_get_obd_name(inode, cmd, arg));
3821 case LL_IOC_HSM_STATE_GET: {
3822 struct md_op_data *op_data;
3823 struct hsm_user_state *hus;
3830 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3831 LUSTRE_OPC_ANY, hus);
3832 if (IS_ERR(op_data)) {
3834 RETURN(PTR_ERR(op_data));
3837 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3840 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
3843 ll_finish_md_op_data(op_data);
3847 case LL_IOC_HSM_STATE_SET: {
3848 struct hsm_state_set *hss;
3855 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
3860 rc = ll_hsm_state_set(inode, hss);
3865 case LL_IOC_HSM_ACTION: {
3866 struct md_op_data *op_data;
3867 struct hsm_current_action *hca;
3874 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3875 LUSTRE_OPC_ANY, hca);
3876 if (IS_ERR(op_data)) {
3878 RETURN(PTR_ERR(op_data));
3881 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3884 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
3887 ll_finish_md_op_data(op_data);
3891 case LL_IOC_SET_LEASE_OLD: {
3892 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
3894 RETURN(ll_file_set_lease(file, &ioc, 0));
3896 case LL_IOC_SET_LEASE: {
3897 struct ll_ioc_lease ioc;
3899 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
3902 RETURN(ll_file_set_lease(file, &ioc, arg));
3904 case LL_IOC_GET_LEASE: {
3905 struct ll_inode_info *lli = ll_i2info(inode);
3906 struct ldlm_lock *lock = NULL;
3909 mutex_lock(&lli->lli_och_mutex);
3910 if (fd->fd_lease_och != NULL) {
3911 struct obd_client_handle *och = fd->fd_lease_och;
3913 lock = ldlm_handle2lock(&och->och_lease_handle);
3915 lock_res_and_lock(lock);
3916 if (!ldlm_is_cancel(lock))
3917 fmode = och->och_flags;
3919 unlock_res_and_lock(lock);
3920 LDLM_LOCK_PUT(lock);
3923 mutex_unlock(&lli->lli_och_mutex);
3925 RETURN(ll_lease_type_from_fmode(fmode));
3927 case LL_IOC_HSM_IMPORT: {
3928 struct hsm_user_import *hui;
3934 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
3939 rc = ll_hsm_import(inode, file, hui);
3944 case LL_IOC_FUTIMES_3: {
3945 struct ll_futimes_3 lfu;
3947 if (copy_from_user(&lfu,
3948 (const struct ll_futimes_3 __user *)arg,
3952 RETURN(ll_file_futimes_3(file, &lfu));
3954 case LL_IOC_LADVISE: {
3955 struct llapi_ladvise_hdr *k_ladvise_hdr;
3956 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
3959 int alloc_size = sizeof(*k_ladvise_hdr);
3962 u_ladvise_hdr = (void __user *)arg;
3963 OBD_ALLOC_PTR(k_ladvise_hdr);
3964 if (k_ladvise_hdr == NULL)
3967 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3968 GOTO(out_ladvise, rc = -EFAULT);
3970 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
3971 k_ladvise_hdr->lah_count < 1)
3972 GOTO(out_ladvise, rc = -EINVAL);
3974 num_advise = k_ladvise_hdr->lah_count;
3975 if (num_advise >= LAH_COUNT_MAX)
3976 GOTO(out_ladvise, rc = -EFBIG);
3978 OBD_FREE_PTR(k_ladvise_hdr);
3979 alloc_size = offsetof(typeof(*k_ladvise_hdr),
3980 lah_advise[num_advise]);
3981 OBD_ALLOC(k_ladvise_hdr, alloc_size);
3982 if (k_ladvise_hdr == NULL)
3986 * TODO: submit multiple advices to one server in a single RPC
3988 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3989 GOTO(out_ladvise, rc = -EFAULT);
3991 for (i = 0; i < num_advise; i++) {
3992 struct llapi_lu_ladvise *k_ladvise =
3993 &k_ladvise_hdr->lah_advise[i];
3994 struct llapi_lu_ladvise __user *u_ladvise =
3995 &u_ladvise_hdr->lah_advise[i];
3997 rc = ll_ladvise_sanity(inode, k_ladvise);
3999 GOTO(out_ladvise, rc);
4001 switch (k_ladvise->lla_advice) {
4002 case LU_LADVISE_LOCKNOEXPAND:
4003 rc = ll_lock_noexpand(file,
4004 k_ladvise->lla_peradvice_flags);
4005 GOTO(out_ladvise, rc);
4006 case LU_LADVISE_LOCKAHEAD:
4008 rc = ll_file_lock_ahead(file, k_ladvise);
4011 GOTO(out_ladvise, rc);
4014 &u_ladvise->lla_lockahead_result))
4015 GOTO(out_ladvise, rc = -EFAULT);
4018 rc = ll_ladvise(inode, file,
4019 k_ladvise_hdr->lah_flags,
4022 GOTO(out_ladvise, rc);
4029 OBD_FREE(k_ladvise_hdr, alloc_size);
4032 case LL_IOC_FLR_SET_MIRROR: {
4033 /* mirror I/O must be direct to avoid polluting page cache
4035 if (!(file->f_flags & O_DIRECT))
4038 fd->fd_designated_mirror = (__u32)arg;
4041 case FS_IOC_FSGETXATTR:
4042 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
4043 case FS_IOC_FSSETXATTR:
4044 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
4046 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
4047 case LL_IOC_HEAT_GET: {
4048 struct lu_heat uheat;
4049 struct lu_heat *heat;
4052 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
4055 if (uheat.lh_count > OBD_HEAT_COUNT)
4056 uheat.lh_count = OBD_HEAT_COUNT;
4058 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4059 OBD_ALLOC(heat, size);
4063 heat->lh_count = uheat.lh_count;
4064 ll_heat_get(inode, heat);
4065 rc = copy_to_user((char __user *)arg, heat, size);
4066 OBD_FREE(heat, size);
4067 RETURN(rc ? -EFAULT : 0);
4069 case LL_IOC_HEAT_SET: {
4072 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
4075 rc = ll_heat_set(inode, flags);
4078 case LL_IOC_PCC_DETACH: {
4079 struct lu_pcc_detach *detach;
4081 OBD_ALLOC_PTR(detach);
4085 if (copy_from_user(detach,
4086 (const struct lu_pcc_detach __user *)arg,
4088 GOTO(out_detach_free, rc = -EFAULT);
4090 if (!S_ISREG(inode->i_mode))
4091 GOTO(out_detach_free, rc = -EINVAL);
4093 if (!inode_owner_or_capable(inode))
4094 GOTO(out_detach_free, rc = -EPERM);
4096 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4098 OBD_FREE_PTR(detach);
4101 case LL_IOC_PCC_STATE: {
4102 struct lu_pcc_state __user *ustate =
4103 (struct lu_pcc_state __user *)arg;
4104 struct lu_pcc_state *state;
4106 OBD_ALLOC_PTR(state);
4110 if (copy_from_user(state, ustate, sizeof(*state)))
4111 GOTO(out_state, rc = -EFAULT);
4113 rc = pcc_ioctl_state(file, inode, state);
4115 GOTO(out_state, rc);
4117 if (copy_to_user(ustate, state, sizeof(*state)))
4118 GOTO(out_state, rc = -EFAULT);
4121 OBD_FREE_PTR(state);
4124 #ifdef HAVE_LUSTRE_CRYPTO
4125 case LL_IOC_SET_ENCRYPTION_POLICY:
4126 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4128 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
4129 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
4130 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4132 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
4133 case LL_IOC_ADD_ENCRYPTION_KEY:
4134 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4136 return llcrypt_ioctl_add_key(file, (void __user *)arg);
4137 case LL_IOC_REMOVE_ENCRYPTION_KEY:
4138 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4140 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
4141 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
4142 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4144 return llcrypt_ioctl_remove_key_all_users(file,
4145 (void __user *)arg);
4146 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
4147 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4149 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
4152 case LL_IOC_UNLOCK_FOREIGN: {
4153 struct dentry *dentry = file_dentry(file);
4155 /* if not a foreign symlink do nothing */
4156 if (ll_foreign_is_removable(dentry, true)) {
4158 "prevent unlink of non-foreign file ("DFID")\n",
4159 PFID(ll_inode2fid(inode)));
4160 RETURN(-EOPNOTSUPP);
4166 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
4167 (void __user *)arg));
4171 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4173 struct inode *inode = file_inode(file);
4176 struct cl_lseek_io *lsio;
4183 env = cl_env_get(&refcheck);
4185 RETURN(PTR_ERR(env));
4187 io = vvp_env_thread_io(env);
4188 io->ci_obj = ll_i2info(inode)->lli_clob;
4189 ll_io_set_mirror(io, file);
4191 lsio = &io->u.ci_lseek;
4192 lsio->ls_start = offset;
4193 lsio->ls_whence = whence;
4194 lsio->ls_result = -ENXIO;
4197 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4199 struct vvp_io *vio = vvp_env_io(env);
4201 vio->vui_fd = file->private_data;
4202 rc = cl_io_loop(env, io);
4206 retval = rc ? : lsio->ls_result;
4207 cl_io_fini(env, io);
4208 } while (unlikely(io->ci_need_restart));
4210 cl_env_put(env, &refcheck);
4215 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4217 struct inode *inode = file_inode(file);
4218 loff_t retval = offset, eof = 0;
4219 ktime_t kstart = ktime_get();
4223 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4224 PFID(ll_inode2fid(inode)), inode, retval, retval,
4227 if (origin == SEEK_END) {
4228 retval = ll_glimpse_size(inode);
4231 eof = i_size_read(inode);
4234 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4238 /* flush local cache first if any */
4239 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4242 retval = ll_lseek(file, offset, origin);
4245 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4247 retval = generic_file_llseek_size(file, offset, origin,
4248 ll_file_maxbytes(inode), eof);
4251 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4252 ktime_us_delta(ktime_get(), kstart));
4256 static int ll_flush(struct file *file, fl_owner_t id)
4258 struct inode *inode = file_inode(file);
4259 struct ll_inode_info *lli = ll_i2info(inode);
4260 struct ll_file_data *fd = file->private_data;
4263 LASSERT(!S_ISDIR(inode->i_mode));
4265 /* catch async errors that were recorded back when async writeback
4266 * failed for pages in this mapping. */
4267 rc = lli->lli_async_rc;
4268 lli->lli_async_rc = 0;
4269 if (lli->lli_clob != NULL) {
4270 err = lov_read_and_clear_async_rc(lli->lli_clob);
4275 /* The application has been told write failure already.
4276 * Do not report failure again. */
4277 if (fd->fd_write_failed)
4279 return rc ? -EIO : 0;
4283 * Called to make sure a portion of file has been written out.
4284 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4286 * Return how many pages have been written.
4288 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4289 enum cl_fsync_mode mode, int ignore_layout)
4293 struct cl_fsync_io *fio;
4298 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4299 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4302 env = cl_env_get(&refcheck);
4304 RETURN(PTR_ERR(env));
4306 io = vvp_env_thread_io(env);
4307 io->ci_obj = ll_i2info(inode)->lli_clob;
4308 io->ci_ignore_layout = ignore_layout;
4310 /* initialize parameters for sync */
4311 fio = &io->u.ci_fsync;
4312 fio->fi_start = start;
4314 fio->fi_fid = ll_inode2fid(inode);
4315 fio->fi_mode = mode;
4316 fio->fi_nr_written = 0;
4318 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4319 result = cl_io_loop(env, io);
4321 result = io->ci_result;
4323 result = fio->fi_nr_written;
4324 cl_io_fini(env, io);
4325 cl_env_put(env, &refcheck);
4331 * When dentry is provided (the 'else' case), file_dentry() may be
4332 * null and dentry must be used directly rather than pulled from
4333 * file_dentry() as is done otherwise.
4336 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4338 struct dentry *dentry = file_dentry(file);
4339 struct inode *inode = dentry->d_inode;
4340 struct ll_inode_info *lli = ll_i2info(inode);
4341 struct ptlrpc_request *req;
4342 ktime_t kstart = ktime_get();
4348 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4349 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4351 /* fsync's caller has already called _fdata{sync,write}, we want
4352 * that IO to finish before calling the osc and mdc sync methods */
4353 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4356 /* catch async errors that were recorded back when async writeback
4357 * failed for pages in this mapping. */
4358 if (!S_ISDIR(inode->i_mode)) {
4359 err = lli->lli_async_rc;
4360 lli->lli_async_rc = 0;
4363 if (lli->lli_clob != NULL) {
4364 err = lov_read_and_clear_async_rc(lli->lli_clob);
4370 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4374 ptlrpc_req_finished(req);
4376 if (S_ISREG(inode->i_mode)) {
4377 struct ll_file_data *fd = file->private_data;
4380 /* Sync metadata on MDT first, and then sync the cached data
4383 err = pcc_fsync(file, start, end, datasync, &cached);
4385 err = cl_sync_file_range(inode, start, end,
4387 if (rc == 0 && err < 0)
4390 fd->fd_write_failed = true;
4392 fd->fd_write_failed = false;
4395 inode_unlock(inode);
4398 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4399 ktime_us_delta(ktime_get(), kstart));
4404 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4406 struct inode *inode = file_inode(file);
4407 struct ll_sb_info *sbi = ll_i2sbi(inode);
4408 struct ldlm_enqueue_info einfo = {
4409 .ei_type = LDLM_FLOCK,
4410 .ei_cb_cp = ldlm_flock_completion_ast,
4411 .ei_cbdata = file_lock,
4413 struct md_op_data *op_data;
4414 struct lustre_handle lockh = { 0 };
4415 union ldlm_policy_data flock = { { 0 } };
4416 int fl_type = file_lock->fl_type;
4417 ktime_t kstart = ktime_get();
4423 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4424 PFID(ll_inode2fid(inode)), file_lock);
4426 if (file_lock->fl_flags & FL_FLOCK) {
4427 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4428 /* flocks are whole-file locks */
4429 flock.l_flock.end = OFFSET_MAX;
4430 /* For flocks owner is determined by the local file desctiptor*/
4431 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4432 } else if (file_lock->fl_flags & FL_POSIX) {
4433 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4434 flock.l_flock.start = file_lock->fl_start;
4435 flock.l_flock.end = file_lock->fl_end;
4439 flock.l_flock.pid = file_lock->fl_pid;
4441 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4442 /* Somewhat ugly workaround for svc lockd.
4443 * lockd installs custom fl_lmops->lm_compare_owner that checks
4444 * for the fl_owner to be the same (which it always is on local node
4445 * I guess between lockd processes) and then compares pid.
4446 * As such we assign pid to the owner field to make it all work,
4447 * conflict with normal locks is unlikely since pid space and
4448 * pointer space for current->files are not intersecting */
4449 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4450 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4455 einfo.ei_mode = LCK_PR;
4458 /* An unlock request may or may not have any relation to
4459 * existing locks so we may not be able to pass a lock handle
4460 * via a normal ldlm_lock_cancel() request. The request may even
4461 * unlock a byte range in the middle of an existing lock. In
4462 * order to process an unlock request we need all of the same
4463 * information that is given with a normal read or write record
4464 * lock request. To avoid creating another ldlm unlock (cancel)
4465 * message we'll treat a LCK_NL flock request as an unlock. */
4466 einfo.ei_mode = LCK_NL;
4469 einfo.ei_mode = LCK_PW;
4472 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4487 flags = LDLM_FL_BLOCK_NOWAIT;
4493 flags = LDLM_FL_TEST_LOCK;
4496 CERROR("unknown fcntl lock command: %d\n", cmd);
4500 /* Save the old mode so that if the mode in the lock changes we
4501 * can decrement the appropriate reader or writer refcount. */
4502 file_lock->fl_type = einfo.ei_mode;
4504 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4505 LUSTRE_OPC_ANY, NULL);
4506 if (IS_ERR(op_data))
4507 RETURN(PTR_ERR(op_data));
4509 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4510 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4511 flock.l_flock.pid, flags, einfo.ei_mode,
4512 flock.l_flock.start, flock.l_flock.end);
4514 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4517 /* Restore the file lock type if not TEST lock. */
4518 if (!(flags & LDLM_FL_TEST_LOCK))
4519 file_lock->fl_type = fl_type;
4521 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4522 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4523 !(flags & LDLM_FL_TEST_LOCK))
4524 rc2 = locks_lock_file_wait(file, file_lock);
4526 if ((file_lock->fl_flags & FL_FLOCK) &&
4527 (rc == 0 || file_lock->fl_type == F_UNLCK))
4528 rc2 = flock_lock_file_wait(file, file_lock);
4529 if ((file_lock->fl_flags & FL_POSIX) &&
4530 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4531 !(flags & LDLM_FL_TEST_LOCK))
4532 rc2 = posix_lock_file_wait(file, file_lock);
4533 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4535 if (rc2 && file_lock->fl_type != F_UNLCK) {
4536 einfo.ei_mode = LCK_NL;
4537 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4542 ll_finish_md_op_data(op_data);
4545 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4546 ktime_us_delta(ktime_get(), kstart));
4550 int ll_get_fid_by_name(struct inode *parent, const char *name,
4551 int namelen, struct lu_fid *fid,
4552 struct inode **inode)
4554 struct md_op_data *op_data = NULL;
4555 struct mdt_body *body;
4556 struct ptlrpc_request *req;
4560 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4561 LUSTRE_OPC_ANY, NULL);
4562 if (IS_ERR(op_data))
4563 RETURN(PTR_ERR(op_data));
4565 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4566 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4567 ll_finish_md_op_data(op_data);
4571 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4573 GOTO(out_req, rc = -EFAULT);
4575 *fid = body->mbo_fid1;
4578 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
4580 ptlrpc_req_finished(req);
4584 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4587 struct dentry *dchild = NULL;
4588 struct inode *child_inode = NULL;
4589 struct md_op_data *op_data;
4590 struct ptlrpc_request *request = NULL;
4591 struct obd_client_handle *och = NULL;
4593 struct mdt_body *body;
4594 __u64 data_version = 0;
4595 size_t namelen = strlen(name);
4596 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4600 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4601 PFID(ll_inode2fid(parent)), name,
4602 lum->lum_stripe_offset, lum->lum_stripe_count);
4604 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4605 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4606 lustre_swab_lmv_user_md(lum);
4608 /* Get child FID first */
4609 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4612 dchild = d_lookup(file_dentry(file), &qstr);
4614 if (dchild->d_inode)
4615 child_inode = igrab(dchild->d_inode);
4620 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4629 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4630 OBD_CONNECT2_DIR_MIGRATE)) {
4631 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4632 ll_dir_striped(child_inode)) {
4633 CERROR("%s: MDT doesn't support stripe directory "
4634 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4635 GOTO(out_iput, rc = -EOPNOTSUPP);
4640 * lfs migrate command needs to be blocked on the client
4641 * by checking the migrate FID against the FID of the
4644 if (is_root_inode(child_inode))
4645 GOTO(out_iput, rc = -EINVAL);
4647 if (IS_ENCRYPTED(child_inode)) {
4648 rc = llcrypt_get_encryption_info(child_inode);
4651 if (!llcrypt_has_encryption_key(child_inode)) {
4652 CDEBUG(D_SEC, "no enc key for "DFID"\n",
4653 PFID(ll_inode2fid(child_inode)));
4654 GOTO(out_iput, rc = -ENOKEY);
4658 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4659 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4660 if (IS_ERR(op_data))
4661 GOTO(out_iput, rc = PTR_ERR(op_data));
4663 inode_lock(child_inode);
4664 op_data->op_fid3 = *ll_inode2fid(child_inode);
4665 if (!fid_is_sane(&op_data->op_fid3)) {
4666 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4667 ll_i2sbi(parent)->ll_fsname, name,
4668 PFID(&op_data->op_fid3));
4669 GOTO(out_unlock, rc = -EINVAL);
4672 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4673 op_data->op_data = lum;
4674 op_data->op_data_size = lumlen;
4677 if (S_ISREG(child_inode->i_mode)) {
4678 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4682 GOTO(out_unlock, rc);
4685 rc = ll_data_version(child_inode, &data_version,
4688 GOTO(out_close, rc);
4690 op_data->op_open_handle = och->och_open_handle;
4691 op_data->op_data_version = data_version;
4692 op_data->op_lease_handle = och->och_lease_handle;
4693 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4695 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4696 och->och_mod->mod_open_req->rq_replay = 0;
4697 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4700 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name, namelen,
4701 name, namelen, &request);
4703 LASSERT(request != NULL);
4704 ll_update_times(request, parent);
4707 if (rc == 0 || rc == -EAGAIN) {
4708 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4709 LASSERT(body != NULL);
4711 /* If the server does release layout lock, then we cleanup
4712 * the client och here, otherwise release it in out_close: */
4713 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4714 obd_mod_put(och->och_mod);
4715 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4717 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4723 if (request != NULL) {
4724 ptlrpc_req_finished(request);
4728 /* Try again if the lease has cancelled. */
4729 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4734 ll_lease_close(och, child_inode, NULL);
4736 clear_nlink(child_inode);
4738 inode_unlock(child_inode);
4739 ll_finish_md_op_data(op_data);
4746 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4748 struct ll_file_data *fd = file->private_data;
4752 * In order to avoid flood of warning messages, only print one message
4753 * for one file. And the entire message rate on the client is limited
4754 * by CDEBUG_LIMIT too.
4756 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
4757 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
4758 CDEBUG_LIMIT(D_TTY | D_CONSOLE,
4759 "flock disabled, mount with '-o [local]flock' to enable\r\n");
4765 * test if some locks matching bits and l_req_mode are acquired
4766 * - bits can be in different locks
4767 * - if found clear the common lock bits in *bits
4768 * - the bits not found, are kept in *bits
4770 * \param bits [IN] searched lock bits [IN]
4771 * \param l_req_mode [IN] searched lock mode
4772 * \retval boolean, true iff all bits are found
4774 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
4776 struct lustre_handle lockh;
4777 union ldlm_policy_data policy;
4778 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
4779 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
4788 fid = &ll_i2info(inode)->lli_fid;
4789 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
4790 ldlm_lockname[mode]);
4792 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
4793 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
4794 policy.l_inodebits.bits = *bits & BIT(i);
4795 if (policy.l_inodebits.bits == 0)
4798 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
4799 &policy, mode, &lockh)) {
4800 struct ldlm_lock *lock;
4802 lock = ldlm_handle2lock(&lockh);
4805 ~(lock->l_policy_data.l_inodebits.bits);
4806 LDLM_LOCK_PUT(lock);
4808 *bits &= ~policy.l_inodebits.bits;
4815 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
4816 struct lustre_handle *lockh, __u64 flags,
4817 enum ldlm_mode mode)
4819 union ldlm_policy_data policy = { .l_inodebits = { bits } };
4824 fid = &ll_i2info(inode)->lli_fid;
4825 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
4827 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
4828 fid, LDLM_IBITS, &policy, mode, lockh);
4833 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
4835 /* Already unlinked. Just update nlink and return success */
4836 if (rc == -ENOENT) {
4838 /* If it is striped directory, and there is bad stripe
4839 * Let's revalidate the dentry again, instead of returning
4841 if (ll_dir_striped(inode))
4844 /* This path cannot be hit for regular files unless in
4845 * case of obscure races, so no need to to validate
4847 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
4849 } else if (rc != 0) {
4850 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
4851 "%s: revalidate FID "DFID" error: rc = %d\n",
4852 ll_i2sbi(inode)->ll_fsname,
4853 PFID(ll_inode2fid(inode)), rc);
4859 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
4861 struct inode *parent;
4862 struct inode *inode = dentry->d_inode;
4863 struct obd_export *exp = ll_i2mdexp(inode);
4864 struct lookup_intent oit = {
4867 struct ptlrpc_request *req = NULL;
4868 struct md_op_data *op_data;
4869 const char *name = NULL;
4874 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
4875 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
4877 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
4878 parent = dentry->d_parent->d_inode;
4879 name = dentry->d_name.name;
4880 namelen = dentry->d_name.len;
4885 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
4886 LUSTRE_OPC_ANY, NULL);
4887 if (IS_ERR(op_data))
4888 RETURN(PTR_ERR(op_data));
4890 /* Call getattr by fid */
4891 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
4892 op_data->op_flags = MF_GETATTR_BY_FID;
4893 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
4894 ll_finish_md_op_data(op_data);
4896 rc = ll_inode_revalidate_fini(inode, rc);
4900 rc = ll_revalidate_it_finish(req, &oit, dentry);
4902 ll_intent_release(&oit);
4906 /* Unlinked? Unhash dentry, so it is not picked up later by
4907 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
4908 * here to preserve get_cwd functionality on 2.6.
4910 if (!dentry->d_inode->i_nlink)
4911 d_lustre_invalidate(dentry);
4913 ll_lookup_finish_locks(&oit, dentry);
4915 ptlrpc_req_finished(req);
4920 static int ll_merge_md_attr(struct inode *inode)
4922 struct ll_inode_info *lli = ll_i2info(inode);
4923 struct cl_attr attr = { 0 };
4926 LASSERT(lli->lli_lsm_md != NULL);
4928 if (!lmv_dir_striped(lli->lli_lsm_md))
4931 down_read(&lli->lli_lsm_sem);
4932 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
4933 &attr, ll_md_blocking_ast);
4934 up_read(&lli->lli_lsm_sem);
4938 set_nlink(inode, attr.cat_nlink);
4939 inode->i_blocks = attr.cat_blocks;
4940 i_size_write(inode, attr.cat_size);
4942 ll_i2info(inode)->lli_atime = attr.cat_atime;
4943 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
4944 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
4949 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
4950 unsigned int flags, bool foreign)
4952 struct inode *inode = de->d_inode;
4953 struct ll_sb_info *sbi = ll_i2sbi(inode);
4954 struct ll_inode_info *lli = ll_i2info(inode);
4955 struct inode *dir = de->d_parent->d_inode;
4956 bool need_glimpse = true;
4957 ktime_t kstart = ktime_get();
4960 /* The OST object(s) determine the file size, blocks and mtime. */
4961 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
4962 request_mask & STATX_MTIME))
4963 need_glimpse = false;
4965 if (dentry_may_statahead(dir, de))
4966 ll_start_statahead(dir, de, need_glimpse &&
4967 !(flags & AT_STATX_DONT_SYNC));
4969 if (flags & AT_STATX_DONT_SYNC)
4970 GOTO(fill_attr, rc = 0);
4972 rc = ll_inode_revalidate(de, IT_GETATTR);
4976 /* foreign file/dir are always of zero length, so don't
4977 * need to validate size.
4979 if (S_ISREG(inode->i_mode) && !foreign) {
4983 GOTO(fill_attr, rc);
4985 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
4986 if (cached && rc < 0)
4990 GOTO(fill_attr, rc);
4993 * If the returned attr is masked with OBD_MD_FLSIZE &
4994 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
4995 * or blocks obtained from MDT is strictly correct, and the file
4996 * is usually not being modified by clients, and the [a|m|c]time
4997 * got from MDT is also strictly correct.
4998 * Under this circumstance, it does not need to send glimpse
4999 * RPCs to OSTs for file attributes such as the size and blocks.
5001 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5002 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5003 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5004 inode->i_mtime.tv_sec = lli->lli_mtime;
5005 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5006 inode->i_atime.tv_sec = lli->lli_atime;
5007 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5008 inode->i_ctime.tv_sec = lli->lli_ctime;
5009 GOTO(fill_attr, rc);
5012 /* In case of restore, the MDT has the right size and has
5013 * already send it back without granting the layout lock,
5014 * inode is up-to-date so glimpse is useless.
5015 * Also to glimpse we need the layout, in case of a running
5016 * restore the MDT holds the layout lock so the glimpse will
5017 * block up to the end of restore (getattr will block)
5019 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5020 rc = ll_glimpse_size(inode);
5025 /* If object isn't regular a file then don't validate size. */
5026 /* foreign dir is not striped dir */
5027 if (ll_dir_striped(inode) && !foreign) {
5028 rc = ll_merge_md_attr(inode);
5033 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5034 inode->i_atime.tv_sec = lli->lli_atime;
5035 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5036 inode->i_mtime.tv_sec = lli->lli_mtime;
5037 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5038 inode->i_ctime.tv_sec = lli->lli_ctime;
5042 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5044 if (ll_need_32bit_api(sbi)) {
5045 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5046 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5047 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5049 stat->ino = inode->i_ino;
5050 stat->dev = inode->i_sb->s_dev;
5051 stat->rdev = inode->i_rdev;
5054 /* foreign symlink to be exposed as a real symlink */
5056 stat->mode = inode->i_mode;
5058 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5060 stat->uid = inode->i_uid;
5061 stat->gid = inode->i_gid;
5062 stat->atime = inode->i_atime;
5063 stat->mtime = inode->i_mtime;
5064 stat->ctime = inode->i_ctime;
5065 /* stat->blksize is used to tell about preferred IO size */
5066 if (sbi->ll_stat_blksize)
5067 stat->blksize = sbi->ll_stat_blksize;
5068 else if (S_ISREG(inode->i_mode))
5069 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5070 LL_MAX_BLKSIZE_BITS);
5072 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5074 stat->nlink = inode->i_nlink;
5075 stat->size = i_size_read(inode);
5076 stat->blocks = inode->i_blocks;
5078 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
5079 if (flags & AT_STATX_DONT_SYNC) {
5080 if (stat->size == 0 &&
5081 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5082 stat->size = lli->lli_lazysize;
5083 if (stat->blocks == 0 &&
5084 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5085 stat->blocks = lli->lli_lazyblocks;
5088 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5089 stat->result_mask |= STATX_BTIME;
5090 stat->btime.tv_sec = lli->lli_btime;
5093 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5094 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5095 stat->result_mask &= request_mask;
5098 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5099 ktime_us_delta(ktime_get(), kstart));
5104 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
5105 int ll_getattr(const struct path *path, struct kstat *stat,
5106 u32 request_mask, unsigned int flags)
5108 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5112 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5114 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5115 AT_STATX_SYNC_AS_STAT, false);
5119 int cl_falloc(struct inode *inode, int mode, loff_t offset, loff_t len)
5125 loff_t size = i_size_read(inode);
5129 env = cl_env_get(&refcheck);
5131 RETURN(PTR_ERR(env));
5133 io = vvp_env_thread_io(env);
5134 io->ci_obj = ll_i2info(inode)->lli_clob;
5135 io->ci_verify_layout = 1;
5136 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5137 io->u.ci_setattr.sa_falloc_mode = mode;
5138 io->u.ci_setattr.sa_falloc_offset = offset;
5139 io->u.ci_setattr.sa_falloc_end = offset + len;
5140 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5141 if (io->u.ci_setattr.sa_falloc_end > size) {
5142 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5144 /* Check new size against VFS/VM file size limit and rlimit */
5145 rc = inode_newsize_ok(inode, newsize);
5148 if (newsize > ll_file_maxbytes(inode)) {
5149 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5150 (unsigned long long)newsize,
5151 ll_file_maxbytes(inode));
5158 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5160 rc = cl_io_loop(env, io);
5163 cl_io_fini(env, io);
5164 } while (unlikely(io->ci_need_restart));
5167 cl_env_put(env, &refcheck);
5171 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5173 struct inode *inode = filp->f_path.dentry->d_inode;
5176 if (offset < 0 || len <= 0)
5179 * Encrypted inodes can't handle collapse range or zero range or insert
5180 * range since we would need to re-encrypt blocks with a different IV or
5181 * XTS tweak (which are based on the logical block number).
5182 * Similar to what ext4 does.
5184 if (IS_ENCRYPTED(inode) &&
5185 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5186 FALLOC_FL_ZERO_RANGE)))
5187 RETURN(-EOPNOTSUPP);
5190 * mode == 0 (which is standard prealloc) and PUNCH is supported
5191 * Rest of mode options are not supported yet.
5193 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5194 RETURN(-EOPNOTSUPP);
5196 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5198 rc = cl_falloc(inode, mode, offset, len);
5200 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5201 * used by Lustre in several places. Retuning it here would
5202 * confuse applications that explicity test for EOPNOTSUPP
5203 * (95) and fall back to ftruncate().
5205 if (rc == -ENOTSUPP)
5211 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5212 __u64 start, __u64 len)
5216 struct fiemap *fiemap;
5217 unsigned int extent_count = fieinfo->fi_extents_max;
5219 num_bytes = sizeof(*fiemap) + (extent_count *
5220 sizeof(struct fiemap_extent));
5221 OBD_ALLOC_LARGE(fiemap, num_bytes);
5226 fiemap->fm_flags = fieinfo->fi_flags;
5227 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5228 fiemap->fm_start = start;
5229 fiemap->fm_length = len;
5230 if (extent_count > 0 &&
5231 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5232 sizeof(struct fiemap_extent)) != 0)
5233 GOTO(out, rc = -EFAULT);
5235 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5237 if (IS_ENCRYPTED(inode)) {
5240 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5241 fiemap->fm_extents[i].fe_flags |=
5242 FIEMAP_EXTENT_DATA_ENCRYPTED |
5243 FIEMAP_EXTENT_ENCODED;
5246 fieinfo->fi_flags = fiemap->fm_flags;
5247 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5248 if (extent_count > 0 &&
5249 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5250 fiemap->fm_mapped_extents *
5251 sizeof(struct fiemap_extent)) != 0)
5252 GOTO(out, rc = -EFAULT);
5254 OBD_FREE_LARGE(fiemap, num_bytes);
5258 int ll_inode_permission(struct inode *inode, int mask)
5261 struct ll_sb_info *sbi;
5262 struct root_squash_info *squash;
5263 struct cred *cred = NULL;
5264 const struct cred *old_cred = NULL;
5265 bool squash_id = false;
5266 ktime_t kstart = ktime_get();
5270 if (mask & MAY_NOT_BLOCK)
5274 * as root inode are NOT getting validated in lookup operation,
5275 * need to do it before permission check.
5278 if (is_root_inode(inode)) {
5279 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
5284 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5285 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5287 /* squash fsuid/fsgid if needed */
5288 sbi = ll_i2sbi(inode);
5289 squash = &sbi->ll_squash;
5290 if (unlikely(squash->rsi_uid != 0 &&
5291 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5292 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
5296 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5297 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5298 squash->rsi_uid, squash->rsi_gid);
5300 /* update current process's credentials
5301 * and FS capability */
5302 cred = prepare_creds();
5306 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5307 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5308 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5309 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5311 old_cred = override_creds(cred);
5314 rc = generic_permission(inode, mask);
5315 /* restore current process's credentials and FS capability */
5317 revert_creds(old_cred);
5322 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5323 ktime_us_delta(ktime_get(), kstart));
5328 /* -o localflock - only provides locally consistent flock locks */
5329 static const struct file_operations ll_file_operations = {
5330 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5331 # ifdef HAVE_SYNC_READ_WRITE
5332 .read = new_sync_read,
5333 .write = new_sync_write,
5335 .read_iter = ll_file_read_iter,
5336 .write_iter = ll_file_write_iter,
5337 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5338 .read = ll_file_read,
5339 .aio_read = ll_file_aio_read,
5340 .write = ll_file_write,
5341 .aio_write = ll_file_aio_write,
5342 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5343 .unlocked_ioctl = ll_file_ioctl,
5344 .open = ll_file_open,
5345 .release = ll_file_release,
5346 .mmap = ll_file_mmap,
5347 .llseek = ll_file_seek,
5348 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5349 .splice_read = generic_file_splice_read,
5351 .splice_read = pcc_file_splice_read,
5355 .fallocate = ll_fallocate,
5358 static const struct file_operations ll_file_operations_flock = {
5359 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5360 # ifdef HAVE_SYNC_READ_WRITE
5361 .read = new_sync_read,
5362 .write = new_sync_write,
5363 # endif /* HAVE_SYNC_READ_WRITE */
5364 .read_iter = ll_file_read_iter,
5365 .write_iter = ll_file_write_iter,
5366 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5367 .read = ll_file_read,
5368 .aio_read = ll_file_aio_read,
5369 .write = ll_file_write,
5370 .aio_write = ll_file_aio_write,
5371 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5372 .unlocked_ioctl = ll_file_ioctl,
5373 .open = ll_file_open,
5374 .release = ll_file_release,
5375 .mmap = ll_file_mmap,
5376 .llseek = ll_file_seek,
5377 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5378 .splice_read = generic_file_splice_read,
5380 .splice_read = pcc_file_splice_read,
5384 .flock = ll_file_flock,
5385 .lock = ll_file_flock,
5386 .fallocate = ll_fallocate,
5389 /* These are for -o noflock - to return ENOSYS on flock calls */
5390 static const struct file_operations ll_file_operations_noflock = {
5391 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5392 # ifdef HAVE_SYNC_READ_WRITE
5393 .read = new_sync_read,
5394 .write = new_sync_write,
5395 # endif /* HAVE_SYNC_READ_WRITE */
5396 .read_iter = ll_file_read_iter,
5397 .write_iter = ll_file_write_iter,
5398 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5399 .read = ll_file_read,
5400 .aio_read = ll_file_aio_read,
5401 .write = ll_file_write,
5402 .aio_write = ll_file_aio_write,
5403 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5404 .unlocked_ioctl = ll_file_ioctl,
5405 .open = ll_file_open,
5406 .release = ll_file_release,
5407 .mmap = ll_file_mmap,
5408 .llseek = ll_file_seek,
5409 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5410 .splice_read = generic_file_splice_read,
5412 .splice_read = pcc_file_splice_read,
5416 .flock = ll_file_noflock,
5417 .lock = ll_file_noflock,
5418 .fallocate = ll_fallocate,
5421 const struct inode_operations ll_file_inode_operations = {
5422 .setattr = ll_setattr,
5423 .getattr = ll_getattr,
5424 .permission = ll_inode_permission,
5425 #ifdef HAVE_IOP_XATTR
5426 .setxattr = ll_setxattr,
5427 .getxattr = ll_getxattr,
5428 .removexattr = ll_removexattr,
5430 .listxattr = ll_listxattr,
5431 .fiemap = ll_fiemap,
5432 .get_acl = ll_get_acl,
5433 #ifdef HAVE_IOP_SET_ACL
5434 .set_acl = ll_set_acl,
5438 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5440 const struct file_operations *fops = &ll_file_operations_noflock;
5442 if (sbi->ll_flags & LL_SBI_FLOCK)
5443 fops = &ll_file_operations_flock;
5444 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
5445 fops = &ll_file_operations;
5450 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5452 struct ll_inode_info *lli = ll_i2info(inode);
5453 struct cl_object *obj = lli->lli_clob;
5462 env = cl_env_get(&refcheck);
5464 RETURN(PTR_ERR(env));
5466 rc = cl_conf_set(env, lli->lli_clob, conf);
5470 if (conf->coc_opc == OBJECT_CONF_SET) {
5471 struct ldlm_lock *lock = conf->coc_lock;
5472 struct cl_layout cl = {
5476 LASSERT(lock != NULL);
5477 LASSERT(ldlm_has_layout(lock));
5479 /* it can only be allowed to match after layout is
5480 * applied to inode otherwise false layout would be
5481 * seen. Applying layout shoud happen before dropping
5482 * the intent lock. */
5483 ldlm_lock_allow_match(lock);
5485 rc = cl_object_layout_get(env, obj, &cl);
5490 DFID": layout version change: %u -> %u\n",
5491 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5493 ll_layout_version_set(lli, cl.cl_layout_gen);
5497 cl_env_put(env, &refcheck);
5499 RETURN(rc < 0 ? rc : 0);
5502 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5503 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5506 struct ll_sb_info *sbi = ll_i2sbi(inode);
5507 struct ptlrpc_request *req;
5514 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5515 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5516 lock->l_lvb_data, lock->l_lvb_len);
5518 if (lock->l_lvb_data != NULL)
5521 /* if layout lock was granted right away, the layout is returned
5522 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5523 * blocked and then granted via completion ast, we have to fetch
5524 * layout here. Please note that we can't use the LVB buffer in
5525 * completion AST because it doesn't have a large enough buffer */
5526 rc = ll_get_default_mdsize(sbi, &lmmsize);
5530 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5531 XATTR_NAME_LOV, lmmsize, &req);
5534 GOTO(out, rc = 0); /* empty layout */
5541 if (lmmsize == 0) /* empty layout */
5544 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5546 GOTO(out, rc = -EFAULT);
5548 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5549 if (lvbdata == NULL)
5550 GOTO(out, rc = -ENOMEM);
5552 memcpy(lvbdata, lmm, lmmsize);
5553 lock_res_and_lock(lock);
5554 if (unlikely(lock->l_lvb_data == NULL)) {
5555 lock->l_lvb_type = LVB_T_LAYOUT;
5556 lock->l_lvb_data = lvbdata;
5557 lock->l_lvb_len = lmmsize;
5560 unlock_res_and_lock(lock);
5563 OBD_FREE_LARGE(lvbdata, lmmsize);
5568 ptlrpc_req_finished(req);
5573 * Apply the layout to the inode. Layout lock is held and will be released
5576 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5577 struct inode *inode)
5579 struct ll_inode_info *lli = ll_i2info(inode);
5580 struct ll_sb_info *sbi = ll_i2sbi(inode);
5581 struct ldlm_lock *lock;
5582 struct cl_object_conf conf;
5585 bool wait_layout = false;
5588 LASSERT(lustre_handle_is_used(lockh));
5590 lock = ldlm_handle2lock(lockh);
5591 LASSERT(lock != NULL);
5592 LASSERT(ldlm_has_layout(lock));
5594 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5595 PFID(&lli->lli_fid), inode);
5597 /* in case this is a caching lock and reinstate with new inode */
5598 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5600 lock_res_and_lock(lock);
5601 lvb_ready = ldlm_is_lvb_ready(lock);
5602 unlock_res_and_lock(lock);
5604 /* checking lvb_ready is racy but this is okay. The worst case is
5605 * that multi processes may configure the file on the same time. */
5609 rc = ll_layout_fetch(inode, lock);
5613 /* for layout lock, lmm is stored in lock's lvb.
5614 * lvb_data is immutable if the lock is held so it's safe to access it
5617 * set layout to file. Unlikely this will fail as old layout was
5618 * surely eliminated */
5619 memset(&conf, 0, sizeof conf);
5620 conf.coc_opc = OBJECT_CONF_SET;
5621 conf.coc_inode = inode;
5622 conf.coc_lock = lock;
5623 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5624 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5625 rc = ll_layout_conf(inode, &conf);
5627 /* refresh layout failed, need to wait */
5628 wait_layout = rc == -EBUSY;
5631 LDLM_LOCK_PUT(lock);
5632 ldlm_lock_decref(lockh, mode);
5634 /* wait for IO to complete if it's still being used. */
5636 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5637 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5639 memset(&conf, 0, sizeof conf);
5640 conf.coc_opc = OBJECT_CONF_WAIT;
5641 conf.coc_inode = inode;
5642 rc = ll_layout_conf(inode, &conf);
5646 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5647 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5653 * Issue layout intent RPC to MDS.
5654 * \param inode [in] file inode
5655 * \param intent [in] layout intent
5657 * \retval 0 on success
5658 * \retval < 0 error code
5660 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5662 struct ll_inode_info *lli = ll_i2info(inode);
5663 struct ll_sb_info *sbi = ll_i2sbi(inode);
5664 struct md_op_data *op_data;
5665 struct lookup_intent it;
5666 struct ptlrpc_request *req;
5670 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5671 0, 0, LUSTRE_OPC_ANY, NULL);
5672 if (IS_ERR(op_data))
5673 RETURN(PTR_ERR(op_data));
5675 op_data->op_data = intent;
5676 op_data->op_data_size = sizeof(*intent);
5678 memset(&it, 0, sizeof(it));
5679 it.it_op = IT_LAYOUT;
5680 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5681 intent->li_opc == LAYOUT_INTENT_TRUNC)
5682 it.it_flags = FMODE_WRITE;
5684 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5685 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5687 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5688 &ll_md_blocking_ast, 0);
5689 if (it.it_request != NULL)
5690 ptlrpc_req_finished(it.it_request);
5691 it.it_request = NULL;
5693 ll_finish_md_op_data(op_data);
5695 /* set lock data in case this is a new lock */
5697 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5699 ll_intent_drop_lock(&it);
5705 * This function checks if there exists a LAYOUT lock on the client side,
5706 * or enqueues it if it doesn't have one in cache.
5708 * This function will not hold layout lock so it may be revoked any time after
5709 * this function returns. Any operations depend on layout should be redone
5712 * This function should be called before lov_io_init() to get an uptodate
5713 * layout version, the caller should save the version number and after IO
5714 * is finished, this function should be called again to verify that layout
5715 * is not changed during IO time.
5717 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5719 struct ll_inode_info *lli = ll_i2info(inode);
5720 struct ll_sb_info *sbi = ll_i2sbi(inode);
5721 struct lustre_handle lockh;
5722 struct layout_intent intent = {
5723 .li_opc = LAYOUT_INTENT_ACCESS,
5725 enum ldlm_mode mode;
5729 *gen = ll_layout_version_get(lli);
5730 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
5734 LASSERT(fid_is_sane(ll_inode2fid(inode)));
5735 LASSERT(S_ISREG(inode->i_mode));
5737 /* take layout lock mutex to enqueue layout lock exclusively. */
5738 mutex_lock(&lli->lli_layout_mutex);
5741 /* mostly layout lock is caching on the local side, so try to
5742 * match it before grabbing layout lock mutex. */
5743 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
5744 LCK_CR | LCK_CW | LCK_PR |
5746 if (mode != 0) { /* hit cached lock */
5747 rc = ll_layout_lock_set(&lockh, mode, inode);
5753 rc = ll_layout_intent(inode, &intent);
5759 *gen = ll_layout_version_get(lli);
5760 mutex_unlock(&lli->lli_layout_mutex);
5766 * Issue layout intent RPC indicating where in a file an IO is about to write.
5768 * \param[in] inode file inode.
5769 * \param[in] ext write range with start offset of fille in bytes where
5770 * an IO is about to write, and exclusive end offset in
5773 * \retval 0 on success
5774 * \retval < 0 error code
5776 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
5777 struct lu_extent *ext)
5779 struct layout_intent intent = {
5781 .li_extent.e_start = ext->e_start,
5782 .li_extent.e_end = ext->e_end,
5787 rc = ll_layout_intent(inode, &intent);
5793 * This function send a restore request to the MDT
5795 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
5797 struct hsm_user_request *hur;
5801 len = sizeof(struct hsm_user_request) +
5802 sizeof(struct hsm_user_item);
5803 OBD_ALLOC(hur, len);
5807 hur->hur_request.hr_action = HUA_RESTORE;
5808 hur->hur_request.hr_archive_id = 0;
5809 hur->hur_request.hr_flags = 0;
5810 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
5811 sizeof(hur->hur_user_item[0].hui_fid));
5812 hur->hur_user_item[0].hui_extent.offset = offset;
5813 hur->hur_user_item[0].hui_extent.length = length;
5814 hur->hur_request.hr_itemcount = 1;
5815 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,