4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
33 * Author: Peter Braam <braam@clusterfs.com>
34 * Author: Phil Schwan <phil@clusterfs.com>
35 * Author: Andreas Dilger <adilger@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <lustre_dlm.h>
40 #include <linux/pagemap.h>
41 #include <linux/file.h>
42 #include <linux/sched.h>
43 #include <linux/user_namespace.h>
44 #include <linux/uidgid.h>
45 #include <linux/falloc.h>
46 #include <linux/ktime.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <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 mutex_lock(&lli->lli_och_mutex);
322 if (fd->fd_lease_och != NULL) {
324 struct obd_client_handle *lease_och;
326 lease_och = fd->fd_lease_och;
327 fd->fd_lease_och = NULL;
328 mutex_unlock(&lli->lli_och_mutex);
330 /* Usually the lease is not released when the
331 * application crashed, we need to release here. */
332 rc = ll_lease_close(lease_och, inode, &lease_broken);
334 mutex_lock(&lli->lli_och_mutex);
336 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
337 "Clean up lease "DFID" %d/%d\n",
338 PFID(&lli->lli_fid), rc, lease_broken);
341 if (fd->fd_och != NULL) {
342 struct obd_client_handle *och;
346 mutex_unlock(&lli->lli_och_mutex);
348 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
352 /* Let's see if we have good enough OPEN lock on the file and if
353 we can skip talking to MDS */
354 if (fd->fd_omode & FMODE_WRITE) {
356 LASSERT(lli->lli_open_fd_write_count);
357 lli->lli_open_fd_write_count--;
358 } else if (fd->fd_omode & FMODE_EXEC) {
360 LASSERT(lli->lli_open_fd_exec_count);
361 lli->lli_open_fd_exec_count--;
364 LASSERT(lli->lli_open_fd_read_count);
365 lli->lli_open_fd_read_count--;
367 mutex_unlock(&lli->lli_och_mutex);
369 /* LU-4398: do not cache write open lock if the file has exec bit */
370 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
371 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
372 LDLM_IBITS, &policy, lockmode, &lockh))
373 rc = ll_md_real_close(inode, fd->fd_omode);
376 file->private_data = NULL;
377 ll_file_data_put(fd);
382 /* While this returns an error code, fput() the caller does not, so we need
383 * to make every effort to clean up all of our state here. Also, applications
384 * rarely check close errors and even if an error is returned they will not
385 * re-try the close call.
387 int ll_file_release(struct inode *inode, struct file *file)
389 struct ll_file_data *fd;
390 struct ll_sb_info *sbi = ll_i2sbi(inode);
391 struct ll_inode_info *lli = ll_i2info(inode);
392 ktime_t kstart = ktime_get();
397 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
398 PFID(ll_inode2fid(inode)), inode);
400 fd = file->private_data;
403 /* The last ref on @file, maybe not the the owner pid of statahead,
404 * because parent and child process can share the same file handle. */
405 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
406 ll_deauthorize_statahead(inode, fd);
408 if (is_root_inode(inode)) {
409 file->private_data = NULL;
410 ll_file_data_put(fd);
414 pcc_file_release(inode, file);
416 if (!S_ISDIR(inode->i_mode)) {
417 if (lli->lli_clob != NULL)
418 lov_read_and_clear_async_rc(lli->lli_clob);
419 lli->lli_async_rc = 0;
422 lli->lli_close_fd_time = ktime_get();
424 rc = ll_md_close(inode, file);
426 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
427 libcfs_debug_dumplog();
430 if (!rc && !is_root_inode(inode))
431 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
432 ktime_us_delta(ktime_get(), kstart));
436 static inline int ll_dom_readpage(void *data, struct page *page)
438 struct niobuf_local *lnb = data;
442 struct inode *inode = page2inode(page);
444 kaddr = kmap_atomic(page);
445 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
446 if (lnb->lnb_len < PAGE_SIZE)
447 memset(kaddr + lnb->lnb_len, 0,
448 PAGE_SIZE - lnb->lnb_len);
449 flush_dcache_page(page);
450 SetPageUptodate(page);
451 kunmap_atomic(kaddr);
453 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
454 if (!llcrypt_has_encryption_key(inode))
455 CDEBUG(D_SEC, "no enc key for "DFID"\n",
456 PFID(ll_inode2fid(inode)));
458 unsigned int offs = 0;
460 while (offs < PAGE_SIZE) {
461 /* decrypt only if page is not empty */
462 if (memcmp(page_address(page) + offs,
463 page_address(ZERO_PAGE(0)),
464 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
467 rc = llcrypt_decrypt_pagecache_blocks(page,
468 LUSTRE_ENCRYPTION_UNIT_SIZE,
473 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
482 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
486 struct ll_inode_info *lli = ll_i2info(inode);
487 struct cl_object *obj = lli->lli_clob;
488 struct address_space *mapping = inode->i_mapping;
490 struct niobuf_remote *rnb;
491 struct mdt_body *body;
493 unsigned long index, start;
494 struct niobuf_local lnb;
503 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
507 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
508 if (rnb == NULL || rnb->rnb_len == 0)
511 /* LU-11595: Server may return whole file and that is OK always or
512 * it may return just file tail and its offset must be aligned with
513 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
514 * smaller then offset may be not aligned and that data is just ignored.
516 if (rnb->rnb_offset & ~PAGE_MASK)
519 /* Server returns whole file or just file tail if it fills in reply
520 * buffer, in both cases total size should be equal to the file size.
522 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
523 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
524 !(inode && IS_ENCRYPTED(inode))) {
525 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
526 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
527 rnb->rnb_len, body->mbo_dom_size);
531 env = cl_env_get(&refcheck);
534 io = vvp_env_thread_io(env);
536 io->ci_ignore_layout = 1;
537 rc = cl_io_init(env, io, CIT_MISC, obj);
541 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
542 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
544 data = (char *)rnb + sizeof(*rnb);
546 lnb.lnb_file_offset = rnb->rnb_offset;
547 start = lnb.lnb_file_offset >> PAGE_SHIFT;
549 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
550 lnb.lnb_page_offset = 0;
552 struct cl_page *page;
554 lnb.lnb_data = data + (index << PAGE_SHIFT);
555 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
556 if (lnb.lnb_len > PAGE_SIZE)
557 lnb.lnb_len = PAGE_SIZE;
559 vmpage = read_cache_page(mapping, index + start,
560 ll_dom_readpage, &lnb);
561 if (IS_ERR(vmpage)) {
562 CWARN("%s: cannot fill page %lu for "DFID
563 " with data: rc = %li\n",
564 ll_i2sbi(inode)->ll_fsname, index + start,
565 PFID(lu_object_fid(&obj->co_lu)),
570 if (vmpage->mapping == NULL) {
573 /* page was truncated */
576 /* attach VM page to CL page cache */
577 page = cl_page_find(env, obj, vmpage->index, vmpage,
580 ClearPageUptodate(vmpage);
585 cl_page_export(env, page, 1);
586 cl_page_put(env, page);
590 } while (rnb->rnb_len > (index << PAGE_SHIFT));
594 cl_env_put(env, &refcheck);
599 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
600 struct lookup_intent *itp)
602 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
603 struct dentry *parent = de->d_parent;
606 struct md_op_data *op_data;
607 struct ptlrpc_request *req = NULL;
611 LASSERT(parent != NULL);
612 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
614 /* if server supports open-by-fid, or file name is invalid, don't pack
615 * name in open request */
616 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
617 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
619 len = de->d_name.len;
620 name = kmalloc(len + 1, GFP_NOFS);
625 spin_lock(&de->d_lock);
626 if (len != de->d_name.len) {
627 spin_unlock(&de->d_lock);
631 memcpy(name, de->d_name.name, len);
633 spin_unlock(&de->d_lock);
635 if (!lu_name_is_valid_2(name, len)) {
641 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
642 name, len, 0, LUSTRE_OPC_OPEN, NULL);
643 if (IS_ERR(op_data)) {
645 RETURN(PTR_ERR(op_data));
647 op_data->op_data = lmm;
648 op_data->op_data_size = lmmsize;
650 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
652 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
653 &ll_md_blocking_ast, 0);
655 ll_finish_md_op_data(op_data);
657 /* reason for keep own exit path - don`t flood log
658 * with messages with -ESTALE errors.
660 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
661 it_open_error(DISP_OPEN_OPEN, itp))
663 ll_release_openhandle(de, itp);
667 if (it_disposition(itp, DISP_LOOKUP_NEG))
668 GOTO(out, rc = -ENOENT);
670 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
671 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
672 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
676 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
678 if (!rc && itp->it_lock_mode) {
681 /* If we got a lock back and it has a LOOKUP bit set,
682 * make sure the dentry is marked as valid so we can find it.
683 * We don't need to care about actual hashing since other bits
684 * of kernel will deal with that later.
686 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
687 if (bits & MDS_INODELOCK_LOOKUP) {
688 d_lustre_revalidate(de);
689 ll_update_dir_depth(parent->d_inode, de->d_inode);
692 /* if DoM bit returned along with LAYOUT bit then there
693 * can be read-on-open data returned.
695 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
696 ll_dom_finish_open(de->d_inode, req);
700 ptlrpc_req_finished(req);
701 ll_intent_drop_lock(itp);
703 /* We did open by fid, but by the time we got to the server, the object
704 * disappeared. This is possible if the object was unlinked, but it's
705 * also possible if the object was unlinked by a rename. In the case
706 * of an object renamed over our existing one, we can't fail this open.
707 * O_CREAT also goes through this path if we had an existing dentry,
708 * and it's obviously wrong to return ENOENT for O_CREAT.
710 * Instead let's return -ESTALE, and the VFS will retry the open with
711 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
712 * revalidate, causing a lookup. This causes extra lookups in the case
713 * where we had a dentry in cache but the file is being unlinked and we
714 * lose the race with unlink, but this should be very rare.
722 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
723 struct obd_client_handle *och)
725 struct mdt_body *body;
727 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
728 och->och_open_handle = body->mbo_open_handle;
729 och->och_fid = body->mbo_fid1;
730 och->och_lease_handle.cookie = it->it_lock_handle;
731 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
732 och->och_flags = it->it_flags;
734 return md_set_open_replay_data(md_exp, och, it);
737 static int ll_local_open(struct file *file, struct lookup_intent *it,
738 struct ll_file_data *fd, struct obd_client_handle *och)
740 struct inode *inode = file_inode(file);
743 LASSERT(!file->private_data);
750 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
755 file->private_data = fd;
756 ll_readahead_init(inode, &fd->fd_ras);
757 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
758 /* turn off the kernel's read-ahead */
759 file->f_ra.ra_pages = 0;
761 /* ll_cl_context initialize */
762 rwlock_init(&fd->fd_lock);
763 INIT_LIST_HEAD(&fd->fd_lccs);
768 void ll_track_file_opens(struct inode *inode)
770 struct ll_inode_info *lli = ll_i2info(inode);
771 struct ll_sb_info *sbi = ll_i2sbi(inode);
773 /* do not skew results with delays from never-opened inodes */
774 if (ktime_to_ns(lli->lli_close_fd_time))
775 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
776 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
778 if (ktime_after(ktime_get(),
779 ktime_add_ms(lli->lli_close_fd_time,
780 sbi->ll_oc_max_ms))) {
781 lli->lli_open_fd_count = 1;
782 lli->lli_close_fd_time = ns_to_ktime(0);
784 lli->lli_open_fd_count++;
787 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
788 lli->lli_open_fd_count);
791 /* Open a file, and (for the very first open) create objects on the OSTs at
792 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
793 * creation or open until ll_lov_setstripe() ioctl is called.
795 * If we already have the stripe MD locally then we don't request it in
796 * md_open(), by passing a lmm_size = 0.
798 * It is up to the application to ensure no other processes open this file
799 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
800 * used. We might be able to avoid races of that sort by getting lli_open_sem
801 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
802 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
804 int ll_file_open(struct inode *inode, struct file *file)
806 struct ll_inode_info *lli = ll_i2info(inode);
807 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
808 .it_flags = file->f_flags };
809 struct obd_client_handle **och_p = NULL;
810 __u64 *och_usecount = NULL;
811 struct ll_file_data *fd;
812 ktime_t kstart = ktime_get();
816 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
817 PFID(ll_inode2fid(inode)), inode, file->f_flags);
819 it = file->private_data; /* XXX: compat macro */
820 file->private_data = NULL; /* prevent ll_local_open assertion */
822 if (S_ISREG(inode->i_mode)) {
823 rc = llcrypt_file_open(inode, file);
825 GOTO(out_nofiledata, rc);
828 fd = ll_file_data_get();
830 GOTO(out_nofiledata, rc = -ENOMEM);
833 if (S_ISDIR(inode->i_mode))
834 ll_authorize_statahead(inode, fd);
836 ll_track_file_opens(inode);
837 if (is_root_inode(inode)) {
838 file->private_data = fd;
842 if (!it || !it->it_disposition) {
843 /* Convert f_flags into access mode. We cannot use file->f_mode,
844 * because everything but O_ACCMODE mask was stripped from
846 if ((oit.it_flags + 1) & O_ACCMODE)
848 if (file->f_flags & O_TRUNC)
849 oit.it_flags |= FMODE_WRITE;
851 /* kernel only call f_op->open in dentry_open. filp_open calls
852 * dentry_open after call to open_namei that checks permissions.
853 * Only nfsd_open call dentry_open directly without checking
854 * permissions and because of that this code below is safe.
856 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
857 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
859 /* We do not want O_EXCL here, presumably we opened the file
860 * already? XXX - NFS implications? */
861 oit.it_flags &= ~O_EXCL;
863 /* bug20584, if "it_flags" contains O_CREAT, the file will be
864 * created if necessary, then "IT_CREAT" should be set to keep
865 * consistent with it */
866 if (oit.it_flags & O_CREAT)
867 oit.it_op |= IT_CREAT;
873 /* Let's see if we have file open on MDS already. */
874 if (it->it_flags & FMODE_WRITE) {
875 och_p = &lli->lli_mds_write_och;
876 och_usecount = &lli->lli_open_fd_write_count;
877 } else if (it->it_flags & FMODE_EXEC) {
878 och_p = &lli->lli_mds_exec_och;
879 och_usecount = &lli->lli_open_fd_exec_count;
881 och_p = &lli->lli_mds_read_och;
882 och_usecount = &lli->lli_open_fd_read_count;
885 mutex_lock(&lli->lli_och_mutex);
886 if (*och_p) { /* Open handle is present */
887 if (it_disposition(it, DISP_OPEN_OPEN)) {
888 /* Well, there's extra open request that we do not need,
889 * let's close it somehow. This will decref request. */
890 rc = it_open_error(DISP_OPEN_OPEN, it);
892 mutex_unlock(&lli->lli_och_mutex);
893 GOTO(out_openerr, rc);
896 ll_release_openhandle(file_dentry(file), it);
900 rc = ll_local_open(file, it, fd, NULL);
903 mutex_unlock(&lli->lli_och_mutex);
904 GOTO(out_openerr, rc);
907 LASSERT(*och_usecount == 0);
908 if (!it->it_disposition) {
909 struct dentry *dentry = file_dentry(file);
910 struct ll_sb_info *sbi = ll_i2sbi(inode);
911 struct ll_dentry_data *ldd;
913 /* We cannot just request lock handle now, new ELC code
914 * means that one of other OPEN locks for this file
915 * could be cancelled, and since blocking ast handler
916 * would attempt to grab och_mutex as well, that would
917 * result in a deadlock
919 mutex_unlock(&lli->lli_och_mutex);
921 * Normally called under two situations:
923 * 2. A race/condition on MDS resulting in no open
924 * handle to be returned from LOOKUP|OPEN request,
925 * for example if the target entry was a symlink.
927 * In NFS path we know there's pathologic behavior
928 * so we always enable open lock caching when coming
929 * from there. It's detected by setting a flag in
932 * After reaching number of opens of this inode
933 * we always ask for an open lock on it to handle
934 * bad userspace actors that open and close files
935 * in a loop for absolutely no good reason
938 ldd = ll_d2d(dentry);
939 if (filename_is_volatile(dentry->d_name.name,
942 /* There really is nothing here, but this
943 * make this more readable I think.
944 * We do not want openlock for volatile
945 * files under any circumstances
947 } else if (ldd && ldd->lld_nfs_dentry) {
948 /* NFS path. This also happens to catch
949 * open by fh files I guess
951 it->it_flags |= MDS_OPEN_LOCK;
952 /* clear the flag for future lookups */
953 ldd->lld_nfs_dentry = 0;
954 } else if (sbi->ll_oc_thrsh_count > 0) {
955 /* Take MDS_OPEN_LOCK with many opens */
956 if (lli->lli_open_fd_count >=
957 sbi->ll_oc_thrsh_count)
958 it->it_flags |= MDS_OPEN_LOCK;
960 /* If this is open after we just closed */
961 else if (ktime_before(ktime_get(),
962 ktime_add_ms(lli->lli_close_fd_time,
963 sbi->ll_oc_thrsh_ms)))
964 it->it_flags |= MDS_OPEN_LOCK;
968 * Always specify MDS_OPEN_BY_FID because we don't want
969 * to get file with different fid.
971 it->it_flags |= MDS_OPEN_BY_FID;
972 rc = ll_intent_file_open(dentry, NULL, 0, it);
974 GOTO(out_openerr, rc);
978 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
980 GOTO(out_och_free, rc = -ENOMEM);
984 /* md_intent_lock() didn't get a request ref if there was an
985 * open error, so don't do cleanup on the request here
987 /* XXX (green): Should not we bail out on any error here, not
988 * just open error? */
989 rc = it_open_error(DISP_OPEN_OPEN, it);
991 GOTO(out_och_free, rc);
993 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
994 "inode %p: disposition %x, status %d\n", inode,
995 it_disposition(it, ~0), it->it_status);
997 rc = ll_local_open(file, it, fd, *och_p);
999 GOTO(out_och_free, rc);
1002 rc = pcc_file_open(inode, file);
1004 GOTO(out_och_free, rc);
1006 mutex_unlock(&lli->lli_och_mutex);
1010 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1011 different kind of OPEN lock for this same inode gets cancelled
1012 by ldlm_cancel_lru */
1013 if (!S_ISREG(inode->i_mode))
1014 GOTO(out_och_free, rc);
1015 cl_lov_delay_create_clear(&file->f_flags);
1016 GOTO(out_och_free, rc);
1020 if (och_p && *och_p) {
1021 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1022 *och_p = NULL; /* OBD_FREE writes some magic there */
1025 mutex_unlock(&lli->lli_och_mutex);
1028 if (lli->lli_opendir_key == fd)
1029 ll_deauthorize_statahead(inode, fd);
1032 ll_file_data_put(fd);
1034 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1035 ktime_us_delta(ktime_get(), kstart));
1039 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1040 ptlrpc_req_finished(it->it_request);
1041 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1047 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1048 struct ldlm_lock_desc *desc, void *data, int flag)
1051 struct lustre_handle lockh;
1055 case LDLM_CB_BLOCKING:
1056 ldlm_lock2handle(lock, &lockh);
1057 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1059 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1063 case LDLM_CB_CANCELING:
1071 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1072 * and save it as fd->fd_och so as to force client to reopen the file even
1073 * if it has an open lock in cache already.
1075 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1076 struct lustre_handle *old_open_handle)
1078 struct ll_inode_info *lli = ll_i2info(inode);
1079 struct ll_file_data *fd = file->private_data;
1080 struct obd_client_handle **och_p;
1081 __u64 *och_usecount;
1085 /* Get the openhandle of the file */
1086 mutex_lock(&lli->lli_och_mutex);
1087 if (fd->fd_lease_och != NULL)
1088 GOTO(out_unlock, rc = -EBUSY);
1090 if (fd->fd_och == NULL) {
1091 if (file->f_mode & FMODE_WRITE) {
1092 LASSERT(lli->lli_mds_write_och != NULL);
1093 och_p = &lli->lli_mds_write_och;
1094 och_usecount = &lli->lli_open_fd_write_count;
1096 LASSERT(lli->lli_mds_read_och != NULL);
1097 och_p = &lli->lli_mds_read_och;
1098 och_usecount = &lli->lli_open_fd_read_count;
1101 if (*och_usecount > 1)
1102 GOTO(out_unlock, rc = -EBUSY);
1104 fd->fd_och = *och_p;
1109 *old_open_handle = fd->fd_och->och_open_handle;
1113 mutex_unlock(&lli->lli_och_mutex);
1118 * Release ownership on lli_mds_*_och when putting back a file lease.
1120 static int ll_lease_och_release(struct inode *inode, struct file *file)
1122 struct ll_inode_info *lli = ll_i2info(inode);
1123 struct ll_file_data *fd = file->private_data;
1124 struct obd_client_handle **och_p;
1125 struct obd_client_handle *old_och = NULL;
1126 __u64 *och_usecount;
1130 mutex_lock(&lli->lli_och_mutex);
1131 if (file->f_mode & FMODE_WRITE) {
1132 och_p = &lli->lli_mds_write_och;
1133 och_usecount = &lli->lli_open_fd_write_count;
1135 och_p = &lli->lli_mds_read_och;
1136 och_usecount = &lli->lli_open_fd_read_count;
1139 /* The file may have been open by another process (broken lease) so
1140 * *och_p is not NULL. In this case we should simply increase usecount
1143 if (*och_p != NULL) {
1144 old_och = fd->fd_och;
1147 *och_p = fd->fd_och;
1151 mutex_unlock(&lli->lli_och_mutex);
1153 if (old_och != NULL)
1154 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1160 * Acquire a lease and open the file.
1162 static struct obd_client_handle *
1163 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1166 struct lookup_intent it = { .it_op = IT_OPEN };
1167 struct ll_sb_info *sbi = ll_i2sbi(inode);
1168 struct md_op_data *op_data;
1169 struct ptlrpc_request *req = NULL;
1170 struct lustre_handle old_open_handle = { 0 };
1171 struct obd_client_handle *och = NULL;
1176 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1177 RETURN(ERR_PTR(-EINVAL));
1180 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1181 RETURN(ERR_PTR(-EPERM));
1183 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1185 RETURN(ERR_PTR(rc));
1190 RETURN(ERR_PTR(-ENOMEM));
1192 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1193 LUSTRE_OPC_ANY, NULL);
1194 if (IS_ERR(op_data))
1195 GOTO(out, rc = PTR_ERR(op_data));
1197 /* To tell the MDT this openhandle is from the same owner */
1198 op_data->op_open_handle = old_open_handle;
1200 it.it_flags = fmode | open_flags;
1201 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1202 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1203 &ll_md_blocking_lease_ast,
1204 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1205 * it can be cancelled which may mislead applications that the lease is
1207 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1208 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1209 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1210 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1211 ll_finish_md_op_data(op_data);
1212 ptlrpc_req_finished(req);
1214 GOTO(out_release_it, rc);
1216 if (it_disposition(&it, DISP_LOOKUP_NEG))
1217 GOTO(out_release_it, rc = -ENOENT);
1219 rc = it_open_error(DISP_OPEN_OPEN, &it);
1221 GOTO(out_release_it, rc);
1223 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1224 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1226 GOTO(out_release_it, rc);
1228 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1229 GOTO(out_close, rc = -EOPNOTSUPP);
1231 /* already get lease, handle lease lock */
1232 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1233 if (!it.it_lock_mode ||
1234 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1235 /* open lock must return for lease */
1236 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1237 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1239 GOTO(out_close, rc = -EPROTO);
1242 ll_intent_release(&it);
1246 /* Cancel open lock */
1247 if (it.it_lock_mode != 0) {
1248 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1250 it.it_lock_mode = 0;
1251 och->och_lease_handle.cookie = 0ULL;
1253 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1255 CERROR("%s: error closing file "DFID": %d\n",
1256 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1257 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1259 ll_intent_release(&it);
1263 RETURN(ERR_PTR(rc));
1267 * Check whether a layout swap can be done between two inodes.
1269 * \param[in] inode1 First inode to check
1270 * \param[in] inode2 Second inode to check
1272 * \retval 0 on success, layout swap can be performed between both inodes
1273 * \retval negative error code if requirements are not met
1275 static int ll_check_swap_layouts_validity(struct inode *inode1,
1276 struct inode *inode2)
1278 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1281 if (inode_permission(inode1, MAY_WRITE) ||
1282 inode_permission(inode2, MAY_WRITE))
1285 if (inode1->i_sb != inode2->i_sb)
1291 static int ll_swap_layouts_close(struct obd_client_handle *och,
1292 struct inode *inode, struct inode *inode2)
1294 const struct lu_fid *fid1 = ll_inode2fid(inode);
1295 const struct lu_fid *fid2;
1299 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1300 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1302 rc = ll_check_swap_layouts_validity(inode, inode2);
1304 GOTO(out_free_och, rc);
1306 /* We now know that inode2 is a lustre inode */
1307 fid2 = ll_inode2fid(inode2);
1309 rc = lu_fid_cmp(fid1, fid2);
1311 GOTO(out_free_och, rc = -EINVAL);
1313 /* Close the file and {swap,merge} layouts between inode & inode2.
1314 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1315 * because we still need it to pack l_remote_handle to MDT. */
1316 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1319 och = NULL; /* freed in ll_close_inode_openhandle() */
1329 * Release lease and close the file.
1330 * It will check if the lease has ever broken.
1332 static int ll_lease_close_intent(struct obd_client_handle *och,
1333 struct inode *inode,
1334 bool *lease_broken, enum mds_op_bias bias,
1337 struct ldlm_lock *lock;
1338 bool cancelled = true;
1342 lock = ldlm_handle2lock(&och->och_lease_handle);
1344 lock_res_and_lock(lock);
1345 cancelled = ldlm_is_cancel(lock);
1346 unlock_res_and_lock(lock);
1347 LDLM_LOCK_PUT(lock);
1350 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1351 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1353 if (lease_broken != NULL)
1354 *lease_broken = cancelled;
1356 if (!cancelled && !bias)
1357 ldlm_cli_cancel(&och->och_lease_handle, 0);
1359 if (cancelled) { /* no need to excute intent */
1364 rc = ll_close_inode_openhandle(inode, och, bias, data);
1368 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1371 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1375 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1377 static int ll_lease_file_resync(struct obd_client_handle *och,
1378 struct inode *inode, unsigned long arg)
1380 struct ll_sb_info *sbi = ll_i2sbi(inode);
1381 struct md_op_data *op_data;
1382 struct ll_ioc_lease_id ioc;
1383 __u64 data_version_unused;
1387 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1388 LUSTRE_OPC_ANY, NULL);
1389 if (IS_ERR(op_data))
1390 RETURN(PTR_ERR(op_data));
1392 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1396 /* before starting file resync, it's necessary to clean up page cache
1397 * in client memory, otherwise once the layout version is increased,
1398 * writing back cached data will be denied the OSTs. */
1399 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1403 op_data->op_lease_handle = och->och_lease_handle;
1404 op_data->op_mirror_id = ioc.lil_mirror_id;
1405 rc = md_file_resync(sbi->ll_md_exp, op_data);
1411 ll_finish_md_op_data(op_data);
1415 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1417 struct ll_inode_info *lli = ll_i2info(inode);
1418 struct cl_object *obj = lli->lli_clob;
1419 struct cl_attr *attr = vvp_env_thread_attr(env);
1427 ll_inode_size_lock(inode);
1429 /* Merge timestamps the most recently obtained from MDS with
1430 * timestamps obtained from OSTs.
1432 * Do not overwrite atime of inode because it may be refreshed
1433 * by file_accessed() function. If the read was served by cache
1434 * data, there is no RPC to be sent so that atime may not be
1435 * transferred to OSTs at all. MDT only updates atime at close time
1436 * if it's at least 'mdd.*.atime_diff' older.
1437 * All in all, the atime in Lustre does not strictly comply with
1438 * POSIX. Solving this problem needs to send an RPC to MDT for each
1439 * read, this will hurt performance.
1441 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1442 inode->i_atime.tv_sec < lli->lli_atime)
1443 inode->i_atime.tv_sec = lli->lli_atime;
1445 inode->i_mtime.tv_sec = lli->lli_mtime;
1446 inode->i_ctime.tv_sec = lli->lli_ctime;
1448 mtime = inode->i_mtime.tv_sec;
1449 atime = inode->i_atime.tv_sec;
1450 ctime = inode->i_ctime.tv_sec;
1452 cl_object_attr_lock(obj);
1453 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1456 rc = cl_object_attr_get(env, obj, attr);
1457 cl_object_attr_unlock(obj);
1460 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1462 if (atime < attr->cat_atime)
1463 atime = attr->cat_atime;
1465 if (ctime < attr->cat_ctime)
1466 ctime = attr->cat_ctime;
1468 if (mtime < attr->cat_mtime)
1469 mtime = attr->cat_mtime;
1471 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1472 PFID(&lli->lli_fid), attr->cat_size);
1474 i_size_write(inode, attr->cat_size);
1475 inode->i_blocks = attr->cat_blocks;
1477 inode->i_mtime.tv_sec = mtime;
1478 inode->i_atime.tv_sec = atime;
1479 inode->i_ctime.tv_sec = ctime;
1482 ll_inode_size_unlock(inode);
1488 * Set designated mirror for I/O.
1490 * So far only read, write, and truncated can support to issue I/O to
1491 * designated mirror.
1493 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1495 struct ll_file_data *fd = file->private_data;
1497 /* clear layout version for generic(non-resync) I/O in case it carries
1498 * stale layout version due to I/O restart */
1499 io->ci_layout_version = 0;
1501 /* FLR: disable non-delay for designated mirror I/O because obviously
1502 * only one mirror is available */
1503 if (fd->fd_designated_mirror > 0) {
1505 io->ci_designated_mirror = fd->fd_designated_mirror;
1506 io->ci_layout_version = fd->fd_layout_version;
1509 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1510 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1513 static bool file_is_noatime(const struct file *file)
1515 const struct vfsmount *mnt = file->f_path.mnt;
1516 const struct inode *inode = file_inode((struct file *)file);
1518 /* Adapted from file_accessed() and touch_atime().*/
1519 if (file->f_flags & O_NOATIME)
1522 if (inode->i_flags & S_NOATIME)
1525 if (IS_NOATIME(inode))
1528 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1531 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1534 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1540 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1541 struct vvp_io_args *args)
1543 struct inode *inode = file_inode(file);
1544 struct ll_file_data *fd = file->private_data;
1546 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1547 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1549 if (iot == CIT_WRITE) {
1550 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1551 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1552 file->f_flags & O_DIRECT ||
1554 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1555 io->u.ci_wr.wr_sync |= !!(args &&
1556 (args->u.normal.via_iocb->ki_flags &
1561 io->ci_obj = ll_i2info(inode)->lli_clob;
1562 io->ci_lockreq = CILR_MAYBE;
1563 if (ll_file_nolock(file)) {
1564 io->ci_lockreq = CILR_NEVER;
1565 io->ci_no_srvlock = 1;
1566 } else if (file->f_flags & O_APPEND) {
1567 io->ci_lockreq = CILR_MANDATORY;
1569 io->ci_noatime = file_is_noatime(file);
1570 io->ci_async_readahead = false;
1572 /* FLR: only use non-delay I/O for read as there is only one
1573 * avaliable mirror for write. */
1574 io->ci_ndelay = !(iot == CIT_WRITE);
1576 ll_io_set_mirror(io, file);
1579 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1582 struct ll_inode_info *lli = ll_i2info(inode);
1583 struct ll_sb_info *sbi = ll_i2sbi(inode);
1584 enum obd_heat_type sample_type;
1585 enum obd_heat_type iobyte_type;
1586 __u64 now = ktime_get_real_seconds();
1588 if (!ll_sbi_has_file_heat(sbi) ||
1589 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1592 if (iot == CIT_READ) {
1593 sample_type = OBD_HEAT_READSAMPLE;
1594 iobyte_type = OBD_HEAT_READBYTE;
1595 } else if (iot == CIT_WRITE) {
1596 sample_type = OBD_HEAT_WRITESAMPLE;
1597 iobyte_type = OBD_HEAT_WRITEBYTE;
1602 spin_lock(&lli->lli_heat_lock);
1603 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1604 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1605 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1606 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1607 spin_unlock(&lli->lli_heat_lock);
1611 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1612 struct file *file, enum cl_io_type iot,
1613 loff_t *ppos, size_t count)
1615 struct vvp_io *vio = vvp_env_io(env);
1616 struct inode *inode = file_inode(file);
1617 struct ll_inode_info *lli = ll_i2info(inode);
1618 struct ll_sb_info *sbi = ll_i2sbi(inode);
1619 struct ll_file_data *fd = file->private_data;
1620 struct range_lock range;
1621 bool range_locked = false;
1626 unsigned int retried = 0, dio_lock = 0;
1627 bool is_aio = false;
1628 bool is_parallel_dio = false;
1629 struct cl_dio_aio *ci_aio = NULL;
1631 bool partial_io = false;
1632 size_t max_io_pages, max_cached_pages;
1636 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1637 file_dentry(file)->d_name.name,
1638 iot == CIT_READ ? "read" : "write", *ppos, count);
1640 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1641 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1642 if (max_io_pages > (max_cached_pages >> 2))
1643 max_io_pages = max_cached_pages >> 2;
1645 io = vvp_env_thread_io(env);
1646 if (file->f_flags & O_DIRECT) {
1647 if (!is_sync_kiocb(args->u.normal.via_iocb))
1650 /* the kernel does not support AIO on pipes, and parallel DIO
1651 * uses part of the AIO path, so we must not do parallel dio
1654 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1657 if (!ll_sbi_has_parallel_dio(sbi))
1658 is_parallel_dio = false;
1660 ci_aio = cl_aio_alloc(args->u.normal.via_iocb,
1661 ll_i2info(inode)->lli_clob);
1663 GOTO(out, rc = -ENOMEM);
1668 * IO block size need be aware of cached page limit, otherwise
1669 * if we have small max_cached_mb but large block IO issued, io
1670 * could not be finished and blocked whole client.
1672 if (file->f_flags & O_DIRECT)
1675 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1676 partial_io = per_bytes < count;
1677 io = vvp_env_thread_io(env);
1678 ll_io_init(io, file, iot, args);
1679 io->ci_aio = ci_aio;
1680 io->ci_dio_lock = dio_lock;
1681 io->ci_ndelay_tried = retried;
1682 io->ci_parallel_dio = is_parallel_dio;
1684 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1685 if (file->f_flags & O_APPEND)
1686 range_lock_init(&range, 0, LUSTRE_EOF);
1688 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1690 vio->vui_fd = file->private_data;
1691 vio->vui_iter = args->u.normal.via_iter;
1692 vio->vui_iocb = args->u.normal.via_iocb;
1693 /* Direct IO reads must also take range lock,
1694 * or multiple reads will try to work on the same pages
1695 * See LU-6227 for details.
1697 if (((iot == CIT_WRITE) ||
1698 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1699 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1700 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1702 rc = range_lock(&lli->lli_write_tree, &range);
1706 range_locked = true;
1709 ll_cl_add(file, env, io, LCC_RW);
1710 rc = cl_io_loop(env, io);
1711 ll_cl_remove(file, env);
1713 if (range_locked && !is_parallel_dio) {
1714 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1716 range_unlock(&lli->lli_write_tree, &range);
1717 range_locked = false;
1720 /* cl_io_rw_init() handled IO */
1724 /* N/B: parallel DIO may be disabled during i/o submission;
1725 * if that occurs, async RPCs are resolved before we get here, and this
1726 * wait call completes immediately.
1728 if (is_parallel_dio) {
1729 struct cl_sync_io *anchor = &io->ci_aio->cda_sync;
1731 /* for dio, EIOCBQUEUED is an implementation detail,
1732 * and we don't return it to userspace
1734 if (rc == -EIOCBQUEUED)
1737 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1742 range_unlock(&lli->lli_write_tree, &range);
1743 range_locked = false;
1748 * In order to move forward AIO, ci_nob was increased,
1749 * but that doesn't mean io have been finished, it just
1750 * means io have been submited, we will always return
1751 * EIOCBQUEUED to the caller, So we could only return
1752 * number of bytes in non-AIO case.
1754 if (io->ci_nob > 0) {
1757 result += io->ci_nob;
1758 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1763 count -= io->ci_nob;
1765 /* prepare IO restart */
1767 args->u.normal.via_iter = vio->vui_iter;
1771 * Reexpand iov count because it was zero
1774 iov_iter_reexpand(vio->vui_iter, count);
1775 if (per_bytes == io->ci_nob)
1776 io->ci_need_restart = 1;
1780 cl_io_fini(env, io);
1783 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1784 file->f_path.dentry->d_name.name,
1785 iot, rc, result, io->ci_need_restart);
1787 if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
1788 count > 0 && io->ci_need_restart) {
1790 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1791 file_dentry(file)->d_name.name,
1792 iot == CIT_READ ? "read" : "write",
1793 *ppos, count, result, rc);
1794 /* preserve the tried count for FLR */
1795 retried = io->ci_ndelay_tried;
1796 dio_lock = io->ci_dio_lock;
1802 * VFS will call aio_complete() if no -EIOCBQUEUED
1803 * is returned for AIO, so we can not call aio_complete()
1806 if (rc != -EIOCBQUEUED)
1807 io->ci_aio->cda_no_aio_complete = 1;
1809 * Drop one extra reference so that end_io() could be
1810 * called for this IO context, we could call it after
1811 * we make sure all AIO requests have been proceed.
1813 cl_sync_io_note(env, &io->ci_aio->cda_sync,
1814 rc == -EIOCBQUEUED ? 0 : rc);
1816 cl_aio_free(env, io->ci_aio);
1821 if (iot == CIT_READ) {
1823 ll_stats_ops_tally(ll_i2sbi(inode),
1824 LPROC_LL_READ_BYTES, result);
1825 } else if (iot == CIT_WRITE) {
1827 ll_stats_ops_tally(ll_i2sbi(inode),
1828 LPROC_LL_WRITE_BYTES, result);
1829 fd->fd_write_failed = false;
1830 } else if (result == 0 && rc == 0) {
1833 fd->fd_write_failed = true;
1835 fd->fd_write_failed = false;
1836 } else if (rc != -ERESTARTSYS) {
1837 fd->fd_write_failed = true;
1841 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1843 ll_heat_add(inode, iot, result);
1845 RETURN(result > 0 ? result : rc);
1849 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1850 * especially for small I/O.
1852 * To serve a read request, CLIO has to create and initialize a cl_io and
1853 * then request DLM lock. This has turned out to have siginificant overhead
1854 * and affects the performance of small I/O dramatically.
1856 * It's not necessary to create a cl_io for each I/O. Under the help of read
1857 * ahead, most of the pages being read are already in memory cache and we can
1858 * read those pages directly because if the pages exist, the corresponding DLM
1859 * lock must exist so that page content must be valid.
1861 * In fast read implementation, the llite speculatively finds and reads pages
1862 * in memory cache. There are three scenarios for fast read:
1863 * - If the page exists and is uptodate, kernel VM will provide the data and
1864 * CLIO won't be intervened;
1865 * - If the page was brought into memory by read ahead, it will be exported
1866 * and read ahead parameters will be updated;
1867 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1868 * it will go back and invoke normal read, i.e., a cl_io will be created
1869 * and DLM lock will be requested.
1871 * POSIX compliance: posix standard states that read is intended to be atomic.
1872 * Lustre read implementation is in line with Linux kernel read implementation
1873 * and neither of them complies with POSIX standard in this matter. Fast read
1874 * doesn't make the situation worse on single node but it may interleave write
1875 * results from multiple nodes due to short read handling in ll_file_aio_read().
1877 * \param env - lu_env
1878 * \param iocb - kiocb from kernel
1879 * \param iter - user space buffers where the data will be copied
1881 * \retval - number of bytes have been read, or error code if error occurred.
1884 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1888 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1891 /* NB: we can't do direct IO for fast read because it will need a lock
1892 * to make IO engine happy. */
1893 if (iocb->ki_filp->f_flags & O_DIRECT)
1896 result = generic_file_read_iter(iocb, iter);
1898 /* If the first page is not in cache, generic_file_aio_read() will be
1899 * returned with -ENODATA.
1900 * See corresponding code in ll_readpage(). */
1901 if (result == -ENODATA)
1905 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1906 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1907 LPROC_LL_READ_BYTES, result);
1914 * Read from a file (through the page cache).
1916 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1919 struct vvp_io_args *args;
1920 struct file *file = iocb->ki_filp;
1924 ktime_t kstart = ktime_get();
1927 if (!iov_iter_count(to))
1931 * Currently when PCC read failed, we do not fall back to the
1932 * normal read path, just return the error.
1933 * The resaon is that: for RW-PCC, the file data may be modified
1934 * in the PCC and inconsistent with the data on OSTs (or file
1935 * data has been removed from the Lustre file system), at this
1936 * time, fallback to the normal read path may read the wrong
1938 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1939 * path: read data from data copy on OSTs.
1941 result = pcc_file_read_iter(iocb, to, &cached);
1945 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
1947 result = ll_do_fast_read(iocb, to);
1948 if (result < 0 || iov_iter_count(to) == 0)
1951 env = cl_env_get(&refcheck);
1953 return PTR_ERR(env);
1955 args = ll_env_args(env);
1956 args->u.normal.via_iter = to;
1957 args->u.normal.via_iocb = iocb;
1959 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
1960 &iocb->ki_pos, iov_iter_count(to));
1963 else if (result == 0)
1966 cl_env_put(env, &refcheck);
1969 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1970 file->private_data, iocb->ki_pos, result,
1972 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
1973 ktime_us_delta(ktime_get(), kstart));
1980 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
1981 * If a page is already in the page cache and dirty (and some other things -
1982 * See ll_tiny_write_begin for the instantiation of these rules), then we can
1983 * write to it without doing a full I/O, because Lustre already knows about it
1984 * and will write it out. This saves a lot of processing time.
1986 * All writes here are within one page, so exclusion is handled by the page
1987 * lock on the vm page. We do not do tiny writes for writes which touch
1988 * multiple pages because it's very unlikely multiple sequential pages are
1989 * are already dirty.
1991 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
1992 * and are unlikely to be to already dirty pages.
1994 * Attribute updates are important here, we do them in ll_tiny_write_end.
1996 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
1998 ssize_t count = iov_iter_count(iter);
1999 struct file *file = iocb->ki_filp;
2000 struct inode *inode = file_inode(file);
2001 bool lock_inode = !IS_NOSEC(inode);
2006 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2007 * of function for why.
2009 if (count >= PAGE_SIZE ||
2010 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2013 if (unlikely(lock_inode))
2015 result = __generic_file_write_iter(iocb, iter);
2017 if (unlikely(lock_inode))
2018 inode_unlock(inode);
2020 /* If the page is not already dirty, ll_tiny_write_begin returns
2021 * -ENODATA. We continue on to normal write.
2023 if (result == -ENODATA)
2027 ll_heat_add(inode, CIT_WRITE, result);
2028 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
2030 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2033 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2039 * Write to a file (through the page cache).
2041 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2043 struct vvp_io_args *args;
2045 ssize_t rc_tiny = 0, rc_normal;
2046 struct file *file = iocb->ki_filp;
2049 ktime_t kstart = ktime_get();
2054 if (!iov_iter_count(from))
2055 GOTO(out, rc_normal = 0);
2058 * When PCC write failed, we usually do not fall back to the normal
2059 * write path, just return the error. But there is a special case when
2060 * returned error code is -ENOSPC due to running out of space on PCC HSM
2061 * bakcend. At this time, it will fall back to normal I/O path and
2062 * retry the I/O. As the file is in HSM released state, it will restore
2063 * the file data to OSTs first and redo the write again. And the
2064 * restore process will revoke the layout lock and detach the file
2065 * from PCC cache automatically.
2067 result = pcc_file_write_iter(iocb, from, &cached);
2068 if (cached && result != -ENOSPC && result != -EDQUOT)
2069 GOTO(out, rc_normal = result);
2071 /* NB: we can't do direct IO for tiny writes because they use the page
2072 * cache, we can't do sync writes because tiny writes can't flush
2073 * pages, and we can't do append writes because we can't guarantee the
2074 * required DLM locks are held to protect file size.
2076 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2077 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2078 rc_tiny = ll_do_tiny_write(iocb, from);
2080 /* In case of error, go on and try normal write - Only stop if tiny
2081 * write completed I/O.
2083 if (iov_iter_count(from) == 0)
2084 GOTO(out, rc_normal = rc_tiny);
2086 env = cl_env_get(&refcheck);
2088 return PTR_ERR(env);
2090 args = ll_env_args(env);
2091 args->u.normal.via_iter = from;
2092 args->u.normal.via_iocb = iocb;
2094 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2095 &iocb->ki_pos, iov_iter_count(from));
2097 /* On success, combine bytes written. */
2098 if (rc_tiny >= 0 && rc_normal > 0)
2099 rc_normal += rc_tiny;
2100 /* On error, only return error from normal write if tiny write did not
2101 * write any bytes. Otherwise return bytes written by tiny write.
2103 else if (rc_tiny > 0)
2104 rc_normal = rc_tiny;
2106 cl_env_put(env, &refcheck);
2108 if (rc_normal > 0) {
2109 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2110 file->private_data, iocb->ki_pos,
2112 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2113 ktime_us_delta(ktime_get(), kstart));
2119 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2121 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2123 static int ll_file_get_iov_count(const struct iovec *iov,
2124 unsigned long *nr_segs, size_t *count,
2130 for (seg = 0; seg < *nr_segs; seg++) {
2131 const struct iovec *iv = &iov[seg];
2134 * If any segment has a negative length, or the cumulative
2135 * length ever wraps negative then return -EINVAL.
2138 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2140 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2145 cnt -= iv->iov_len; /* This segment is no good */
2152 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2153 unsigned long nr_segs, loff_t pos)
2160 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2167 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2168 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2169 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2170 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2171 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2173 result = ll_file_read_iter(iocb, &to);
2178 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2181 struct iovec iov = { .iov_base = buf, .iov_len = count };
2190 init_sync_kiocb(&kiocb, file);
2191 kiocb.ki_pos = *ppos;
2192 #ifdef HAVE_KIOCB_KI_LEFT
2193 kiocb.ki_left = count;
2194 #elif defined(HAVE_KI_NBYTES)
2195 kiocb.i_nbytes = count;
2198 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2199 *ppos = kiocb.ki_pos;
2205 * Write to a file (through the page cache).
2208 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2209 unsigned long nr_segs, loff_t pos)
2211 struct iov_iter from;
2216 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2223 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2224 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2225 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2226 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2227 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2229 result = ll_file_write_iter(iocb, &from);
2234 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2235 size_t count, loff_t *ppos)
2237 struct iovec iov = { .iov_base = (void __user *)buf,
2247 init_sync_kiocb(&kiocb, file);
2248 kiocb.ki_pos = *ppos;
2249 #ifdef HAVE_KIOCB_KI_LEFT
2250 kiocb.ki_left = count;
2251 #elif defined(HAVE_KI_NBYTES)
2252 kiocb.ki_nbytes = count;
2255 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2256 *ppos = kiocb.ki_pos;
2260 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2262 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2263 __u64 flags, struct lov_user_md *lum, int lum_size)
2265 struct lookup_intent oit = {
2267 .it_flags = flags | MDS_OPEN_BY_FID,
2272 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2273 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2274 /* this code will only exist for big-endian systems */
2275 lustre_swab_lov_user_md(lum, 0);
2278 ll_inode_size_lock(inode);
2279 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2281 GOTO(out_unlock, rc);
2283 ll_release_openhandle(dentry, &oit);
2286 ll_inode_size_unlock(inode);
2287 ll_intent_release(&oit);
2292 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2293 struct lov_mds_md **lmmp, int *lmm_size,
2294 struct ptlrpc_request **request)
2296 struct ll_sb_info *sbi = ll_i2sbi(inode);
2297 struct mdt_body *body;
2298 struct lov_mds_md *lmm = NULL;
2299 struct ptlrpc_request *req = NULL;
2300 struct md_op_data *op_data;
2305 rc = ll_get_default_mdsize(sbi, &lmmsize);
2309 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2310 strlen(filename), lmmsize,
2311 LUSTRE_OPC_ANY, NULL);
2312 if (IS_ERR(op_data))
2313 RETURN(PTR_ERR(op_data));
2315 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2316 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2317 ll_finish_md_op_data(op_data);
2319 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2324 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2325 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2327 lmmsize = body->mbo_eadatasize;
2329 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2331 GOTO(out, rc = -ENODATA);
2333 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2334 LASSERT(lmm != NULL);
2336 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2337 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2338 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2339 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2340 GOTO(out, rc = -EPROTO);
2343 * This is coming from the MDS, so is probably in
2344 * little endian. We convert it to host endian before
2345 * passing it to userspace.
2347 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2348 int stripe_count = 0;
2350 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2351 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2352 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2353 if (le32_to_cpu(lmm->lmm_pattern) &
2354 LOV_PATTERN_F_RELEASED)
2356 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2358 /* if function called for directory - we should
2359 * avoid swab not existent lsm objects
2361 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2362 S_ISREG(body->mbo_mode))
2363 lustre_swab_lov_user_md_objects(
2364 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2366 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2367 S_ISREG(body->mbo_mode))
2368 lustre_swab_lov_user_md_objects(
2369 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2371 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2372 lustre_swab_lov_comp_md_v1(
2373 (struct lov_comp_md_v1 *)lmm);
2377 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2378 struct lov_comp_md_v1 *comp_v1 = NULL;
2379 struct lov_comp_md_entry_v1 *ent;
2380 struct lov_user_md_v1 *v1;
2384 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2385 /* Dump the striping information */
2386 for (; i < comp_v1->lcm_entry_count; i++) {
2387 ent = &comp_v1->lcm_entries[i];
2388 off = ent->lcme_offset;
2389 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2391 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2392 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2396 * Return valid stripe_count and stripe_size instead of 0 for
2397 * DoM files to avoid divide-by-zero for older userspace that
2398 * calls this ioctl, e.g. lustre ADIO driver.
2400 if (lmm->lmm_stripe_count == 0)
2401 lmm->lmm_stripe_count = 1;
2402 if (lmm->lmm_stripe_size == 0) {
2403 /* Since the first component of the file data is placed
2404 * on the MDT for faster access, the stripe_size of the
2405 * second one is always that applications which are
2408 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2409 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2411 i = comp_v1->lcm_entry_count > 1 ?
2412 comp_v1->lcm_entry_count - 1 : 0;
2413 ent = &comp_v1->lcm_entries[i];
2414 off = ent->lcme_offset;
2415 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2416 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2421 *lmm_size = lmmsize;
2426 static int ll_lov_setea(struct inode *inode, struct file *file,
2429 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2430 struct lov_user_md *lump;
2431 int lum_size = sizeof(struct lov_user_md) +
2432 sizeof(struct lov_user_ost_data);
2436 if (!capable(CAP_SYS_ADMIN))
2439 OBD_ALLOC_LARGE(lump, lum_size);
2443 if (copy_from_user(lump, arg, lum_size))
2444 GOTO(out_lump, rc = -EFAULT);
2446 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2448 cl_lov_delay_create_clear(&file->f_flags);
2451 OBD_FREE_LARGE(lump, lum_size);
2455 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2462 env = cl_env_get(&refcheck);
2464 RETURN(PTR_ERR(env));
2466 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2467 cl_env_put(env, &refcheck);
2471 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2474 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2475 struct lov_user_md *klum;
2477 __u64 flags = FMODE_WRITE;
2480 rc = ll_copy_user_md(lum, &klum);
2485 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2490 rc = put_user(0, &lum->lmm_stripe_count);
2494 rc = ll_layout_refresh(inode, &gen);
2498 rc = ll_file_getstripe(inode, arg, lum_size);
2499 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2500 ll_i2info(inode)->lli_clob) {
2501 struct iattr attr = { 0 };
2503 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2504 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2507 cl_lov_delay_create_clear(&file->f_flags);
2510 OBD_FREE_LARGE(klum, lum_size);
2516 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2518 struct ll_inode_info *lli = ll_i2info(inode);
2519 struct cl_object *obj = lli->lli_clob;
2520 struct ll_file_data *fd = file->private_data;
2521 struct ll_grouplock grouplock;
2526 CWARN("group id for group lock must not be 0\n");
2530 if (ll_file_nolock(file))
2531 RETURN(-EOPNOTSUPP);
2533 if (file->f_flags & O_NONBLOCK) {
2534 if (!mutex_trylock(&lli->lli_group_mutex))
2537 mutex_lock(&lli->lli_group_mutex);
2539 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2540 CWARN("group lock already existed with gid %lu\n",
2541 fd->fd_grouplock.lg_gid);
2542 GOTO(out, rc = -EINVAL);
2544 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2545 if (file->f_flags & O_NONBLOCK)
2546 GOTO(out, rc = -EAGAIN);
2547 mutex_unlock(&lli->lli_group_mutex);
2548 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2549 GOTO(retry, rc = 0);
2551 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2554 * XXX: group lock needs to protect all OST objects while PFL
2555 * can add new OST objects during the IO, so we'd instantiate
2556 * all OST objects before getting its group lock.
2561 struct cl_layout cl = {
2562 .cl_is_composite = false,
2564 struct lu_extent ext = {
2566 .e_end = OBD_OBJECT_EOF,
2569 env = cl_env_get(&refcheck);
2571 GOTO(out, rc = PTR_ERR(env));
2573 rc = cl_object_layout_get(env, obj, &cl);
2574 if (rc >= 0 && cl.cl_is_composite)
2575 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2578 cl_env_put(env, &refcheck);
2583 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2584 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2589 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2590 fd->fd_grouplock = grouplock;
2591 if (lli->lli_group_users == 0)
2592 lli->lli_group_gid = grouplock.lg_gid;
2593 lli->lli_group_users++;
2595 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2597 mutex_unlock(&lli->lli_group_mutex);
2602 static int ll_put_grouplock(struct inode *inode, struct file *file,
2605 struct ll_inode_info *lli = ll_i2info(inode);
2606 struct ll_file_data *fd = file->private_data;
2607 struct ll_grouplock grouplock;
2611 mutex_lock(&lli->lli_group_mutex);
2612 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2613 CWARN("no group lock held\n");
2614 GOTO(out, rc = -EINVAL);
2617 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2619 if (fd->fd_grouplock.lg_gid != arg) {
2620 CWARN("group lock %lu doesn't match current id %lu\n",
2621 arg, fd->fd_grouplock.lg_gid);
2622 GOTO(out, rc = -EINVAL);
2625 grouplock = fd->fd_grouplock;
2626 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2627 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2629 cl_put_grouplock(&grouplock);
2631 lli->lli_group_users--;
2632 if (lli->lli_group_users == 0) {
2633 lli->lli_group_gid = 0;
2634 wake_up_var(&lli->lli_group_users);
2636 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2639 mutex_unlock(&lli->lli_group_mutex);
2645 * Close inode open handle
2647 * \param dentry [in] dentry which contains the inode
2648 * \param it [in,out] intent which contains open info and result
2651 * \retval <0 failure
2653 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2655 struct inode *inode = dentry->d_inode;
2656 struct obd_client_handle *och;
2662 /* Root ? Do nothing. */
2663 if (is_root_inode(inode))
2666 /* No open handle to close? Move away */
2667 if (!it_disposition(it, DISP_OPEN_OPEN))
2670 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2672 OBD_ALLOC(och, sizeof(*och));
2674 GOTO(out, rc = -ENOMEM);
2676 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2680 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2682 /* this one is in place of ll_file_open */
2683 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2684 ptlrpc_req_finished(it->it_request);
2685 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2691 * Get size for inode for which FIEMAP mapping is requested.
2692 * Make the FIEMAP get_info call and returns the result.
2693 * \param fiemap kernel buffer to hold extens
2694 * \param num_bytes kernel buffer size
2696 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2702 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2705 /* Checks for fiemap flags */
2706 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2707 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2711 /* Check for FIEMAP_FLAG_SYNC */
2712 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2713 rc = filemap_fdatawrite(inode->i_mapping);
2718 env = cl_env_get(&refcheck);
2720 RETURN(PTR_ERR(env));
2722 if (i_size_read(inode) == 0) {
2723 rc = ll_glimpse_size(inode);
2728 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2729 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2730 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2732 /* If filesize is 0, then there would be no objects for mapping */
2733 if (fmkey.lfik_oa.o_size == 0) {
2734 fiemap->fm_mapped_extents = 0;
2738 fmkey.lfik_fiemap = *fiemap;
2740 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2741 &fmkey, fiemap, &num_bytes);
2743 cl_env_put(env, &refcheck);
2747 int ll_fid2path(struct inode *inode, void __user *arg)
2749 struct obd_export *exp = ll_i2mdexp(inode);
2750 const struct getinfo_fid2path __user *gfin = arg;
2752 struct getinfo_fid2path *gfout;
2758 if (!capable(CAP_DAC_READ_SEARCH) &&
2759 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2762 /* Only need to get the buflen */
2763 if (get_user(pathlen, &gfin->gf_pathlen))
2766 if (pathlen > PATH_MAX)
2769 outsize = sizeof(*gfout) + pathlen;
2770 OBD_ALLOC(gfout, outsize);
2774 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2775 GOTO(gf_free, rc = -EFAULT);
2776 /* append root FID after gfout to let MDT know the root FID so that it
2777 * can lookup the correct path, this is mainly for fileset.
2778 * old server without fileset mount support will ignore this. */
2779 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2781 /* Call mdc_iocontrol */
2782 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2786 if (copy_to_user(arg, gfout, outsize))
2790 OBD_FREE(gfout, outsize);
2795 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2797 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2805 ioc->idv_version = 0;
2806 ioc->idv_layout_version = UINT_MAX;
2808 /* If no file object initialized, we consider its version is 0. */
2812 env = cl_env_get(&refcheck);
2814 RETURN(PTR_ERR(env));
2816 io = vvp_env_thread_io(env);
2818 io->u.ci_data_version.dv_data_version = 0;
2819 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2820 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2823 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2824 result = cl_io_loop(env, io);
2826 result = io->ci_result;
2828 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2829 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2831 cl_io_fini(env, io);
2833 if (unlikely(io->ci_need_restart))
2836 cl_env_put(env, &refcheck);
2842 * Read the data_version for inode.
2844 * This value is computed using stripe object version on OST.
2845 * Version is computed using server side locking.
2847 * @param flags if do sync on the OST side;
2849 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2850 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2852 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2854 struct ioc_data_version ioc = { .idv_flags = flags };
2857 rc = ll_ioc_data_version(inode, &ioc);
2859 *data_version = ioc.idv_version;
2865 * Trigger a HSM release request for the provided inode.
2867 int ll_hsm_release(struct inode *inode)
2870 struct obd_client_handle *och = NULL;
2871 __u64 data_version = 0;
2876 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2877 ll_i2sbi(inode)->ll_fsname,
2878 PFID(&ll_i2info(inode)->lli_fid));
2880 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2882 GOTO(out, rc = PTR_ERR(och));
2884 /* Grab latest data_version and [am]time values */
2885 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2889 env = cl_env_get(&refcheck);
2891 GOTO(out, rc = PTR_ERR(env));
2893 rc = ll_merge_attr(env, inode);
2894 cl_env_put(env, &refcheck);
2896 /* If error happen, we have the wrong size for a file.
2902 /* Release the file.
2903 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2904 * we still need it to pack l_remote_handle to MDT. */
2905 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2911 if (och != NULL && !IS_ERR(och)) /* close the file */
2912 ll_lease_close(och, inode, NULL);
2917 struct ll_swap_stack {
2920 struct inode *inode1;
2921 struct inode *inode2;
2926 static int ll_swap_layouts(struct file *file1, struct file *file2,
2927 struct lustre_swap_layouts *lsl)
2929 struct mdc_swap_layouts msl;
2930 struct md_op_data *op_data;
2933 struct ll_swap_stack *llss = NULL;
2936 OBD_ALLOC_PTR(llss);
2940 llss->inode1 = file_inode(file1);
2941 llss->inode2 = file_inode(file2);
2943 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2947 /* we use 2 bool because it is easier to swap than 2 bits */
2948 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2949 llss->check_dv1 = true;
2951 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2952 llss->check_dv2 = true;
2954 /* we cannot use lsl->sl_dvX directly because we may swap them */
2955 llss->dv1 = lsl->sl_dv1;
2956 llss->dv2 = lsl->sl_dv2;
2958 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2959 if (rc == 0) /* same file, done! */
2962 if (rc < 0) { /* sequentialize it */
2963 swap(llss->inode1, llss->inode2);
2965 swap(llss->dv1, llss->dv2);
2966 swap(llss->check_dv1, llss->check_dv2);
2970 if (gid != 0) { /* application asks to flush dirty cache */
2971 rc = ll_get_grouplock(llss->inode1, file1, gid);
2975 rc = ll_get_grouplock(llss->inode2, file2, gid);
2977 ll_put_grouplock(llss->inode1, file1, gid);
2982 /* ultimate check, before swaping the layouts we check if
2983 * dataversion has changed (if requested) */
2984 if (llss->check_dv1) {
2985 rc = ll_data_version(llss->inode1, &dv, 0);
2988 if (dv != llss->dv1)
2989 GOTO(putgl, rc = -EAGAIN);
2992 if (llss->check_dv2) {
2993 rc = ll_data_version(llss->inode2, &dv, 0);
2996 if (dv != llss->dv2)
2997 GOTO(putgl, rc = -EAGAIN);
3000 /* struct md_op_data is used to send the swap args to the mdt
3001 * only flags is missing, so we use struct mdc_swap_layouts
3002 * through the md_op_data->op_data */
3003 /* flags from user space have to be converted before they are send to
3004 * server, no flag is sent today, they are only used on the client */
3007 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3008 0, LUSTRE_OPC_ANY, &msl);
3009 if (IS_ERR(op_data))
3010 GOTO(free, rc = PTR_ERR(op_data));
3012 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3013 sizeof(*op_data), op_data, NULL);
3014 ll_finish_md_op_data(op_data);
3021 ll_put_grouplock(llss->inode2, file2, gid);
3022 ll_put_grouplock(llss->inode1, file1, gid);
3032 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3034 struct obd_export *exp = ll_i2mdexp(inode);
3035 struct md_op_data *op_data;
3039 /* Detect out-of range masks */
3040 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3043 /* Non-root users are forbidden to set or clear flags which are
3044 * NOT defined in HSM_USER_MASK. */
3045 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3046 !capable(CAP_SYS_ADMIN))
3049 if (!exp_connect_archive_id_array(exp)) {
3050 /* Detect out-of range archive id */
3051 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3052 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3056 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3057 LUSTRE_OPC_ANY, hss);
3058 if (IS_ERR(op_data))
3059 RETURN(PTR_ERR(op_data));
3061 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3064 ll_finish_md_op_data(op_data);
3069 static int ll_hsm_import(struct inode *inode, struct file *file,
3070 struct hsm_user_import *hui)
3072 struct hsm_state_set *hss = NULL;
3073 struct iattr *attr = NULL;
3077 if (!S_ISREG(inode->i_mode))
3083 GOTO(out, rc = -ENOMEM);
3085 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3086 hss->hss_archive_id = hui->hui_archive_id;
3087 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3088 rc = ll_hsm_state_set(inode, hss);
3092 OBD_ALLOC_PTR(attr);
3094 GOTO(out, rc = -ENOMEM);
3096 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3097 attr->ia_mode |= S_IFREG;
3098 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3099 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3100 attr->ia_size = hui->hui_size;
3101 attr->ia_mtime.tv_sec = hui->hui_mtime;
3102 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3103 attr->ia_atime.tv_sec = hui->hui_atime;
3104 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3106 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3107 ATTR_UID | ATTR_GID |
3108 ATTR_MTIME | ATTR_MTIME_SET |
3109 ATTR_ATIME | ATTR_ATIME_SET;
3113 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3117 inode_unlock(inode);
3129 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3131 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3132 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3135 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3137 struct inode *inode = file_inode(file);
3139 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3140 ATTR_MTIME | ATTR_MTIME_SET |
3143 .tv_sec = lfu->lfu_atime_sec,
3144 .tv_nsec = lfu->lfu_atime_nsec,
3147 .tv_sec = lfu->lfu_mtime_sec,
3148 .tv_nsec = lfu->lfu_mtime_nsec,
3151 .tv_sec = lfu->lfu_ctime_sec,
3152 .tv_nsec = lfu->lfu_ctime_nsec,
3158 if (!capable(CAP_SYS_ADMIN))
3161 if (!S_ISREG(inode->i_mode))
3165 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3167 inode_unlock(inode);
3172 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3175 case MODE_READ_USER:
3177 case MODE_WRITE_USER:
3184 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3186 /* Used to allow the upper layers of the client to request an LDLM lock
3187 * without doing an actual read or write.
3189 * Used for ladvise lockahead to manually request specific locks.
3191 * \param[in] file file this ladvise lock request is on
3192 * \param[in] ladvise ladvise struct describing this lock request
3194 * \retval 0 success, no detailed result available (sync requests
3195 * and requests sent to the server [not handled locally]
3196 * cannot return detailed results)
3197 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3198 * see definitions for details.
3199 * \retval negative negative errno on error
3201 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3203 struct lu_env *env = NULL;
3204 struct cl_io *io = NULL;
3205 struct cl_lock *lock = NULL;
3206 struct cl_lock_descr *descr = NULL;
3207 struct dentry *dentry = file->f_path.dentry;
3208 struct inode *inode = dentry->d_inode;
3209 enum cl_lock_mode cl_mode;
3210 off_t start = ladvise->lla_start;
3211 off_t end = ladvise->lla_end;
3218 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3219 dentry, dentry->d_inode,
3220 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3223 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3225 GOTO(out, result = cl_mode);
3227 /* Get IO environment */
3228 result = cl_io_get(inode, &env, &io, &refcheck);
3232 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3235 * nothing to do for this io. This currently happens when
3236 * stripe sub-object's are not yet created.
3238 result = io->ci_result;
3239 } else if (result == 0) {
3240 lock = vvp_env_lock(env);
3241 descr = &lock->cll_descr;
3243 descr->cld_obj = io->ci_obj;
3244 /* Convert byte offsets to pages */
3245 descr->cld_start = cl_index(io->ci_obj, start);
3246 descr->cld_end = cl_index(io->ci_obj, end);
3247 descr->cld_mode = cl_mode;
3248 /* CEF_MUST is used because we do not want to convert a
3249 * lockahead request to a lockless lock */
3250 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3252 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3253 descr->cld_enq_flags |= CEF_SPECULATIVE;
3255 result = cl_lock_request(env, io, lock);
3257 /* On success, we need to release the lock */
3259 cl_lock_release(env, lock);
3261 cl_io_fini(env, io);
3262 cl_env_put(env, &refcheck);
3264 /* -ECANCELED indicates a matching lock with a different extent
3265 * was already present, and -EEXIST indicates a matching lock
3266 * on exactly the same extent was already present.
3267 * We convert them to positive values for userspace to make
3268 * recognizing true errors easier.
3269 * Note we can only return these detailed results on async requests,
3270 * as sync requests look the same as i/o requests for locking. */
3271 if (result == -ECANCELED)
3272 result = LLA_RESULT_DIFFERENT;
3273 else if (result == -EEXIST)
3274 result = LLA_RESULT_SAME;
3279 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3281 static int ll_ladvise_sanity(struct inode *inode,
3282 struct llapi_lu_ladvise *ladvise)
3284 struct ll_sb_info *sbi = ll_i2sbi(inode);
3285 enum lu_ladvise_type advice = ladvise->lla_advice;
3286 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3287 * be in the first 32 bits of enum ladvise_flags */
3288 __u32 flags = ladvise->lla_peradvice_flags;
3289 /* 3 lines at 80 characters per line, should be plenty */
3292 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3295 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3296 sbi->ll_fsname, advice,
3297 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3301 /* Per-advice checks */
3303 case LU_LADVISE_LOCKNOEXPAND:
3304 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3306 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3307 "rc = %d\n", sbi->ll_fsname, flags,
3308 ladvise_names[advice], rc);
3312 case LU_LADVISE_LOCKAHEAD:
3313 /* Currently only READ and WRITE modes can be requested */
3314 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3315 ladvise->lla_lockahead_mode == 0) {
3317 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3318 "rc = %d\n", sbi->ll_fsname,
3319 ladvise->lla_lockahead_mode,
3320 ladvise_names[advice], rc);
3324 case LU_LADVISE_WILLREAD:
3325 case LU_LADVISE_DONTNEED:
3327 /* Note fall through above - These checks apply to all advices
3328 * except LOCKNOEXPAND */
3329 if (flags & ~LF_DEFAULT_MASK) {
3331 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3332 "rc = %d\n", sbi->ll_fsname, flags,
3333 ladvise_names[advice], rc);
3336 if (ladvise->lla_start >= ladvise->lla_end) {
3338 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3339 "for %s: rc = %d\n", sbi->ll_fsname,
3340 ladvise->lla_start, ladvise->lla_end,
3341 ladvise_names[advice], rc);
3353 * Give file access advices
3355 * The ladvise interface is similar to Linux fadvise() system call, except it
3356 * forwards the advices directly from Lustre client to server. The server side
3357 * codes will apply appropriate read-ahead and caching techniques for the
3358 * corresponding files.
3360 * A typical workload for ladvise is e.g. a bunch of different clients are
3361 * doing small random reads of a file, so prefetching pages into OSS cache
3362 * with big linear reads before the random IO is a net benefit. Fetching
3363 * all that data into each client cache with fadvise() may not be, due to
3364 * much more data being sent to the client.
3366 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3367 struct llapi_lu_ladvise *ladvise)
3371 struct cl_ladvise_io *lio;
3376 env = cl_env_get(&refcheck);
3378 RETURN(PTR_ERR(env));
3380 io = vvp_env_thread_io(env);
3381 io->ci_obj = ll_i2info(inode)->lli_clob;
3383 /* initialize parameters for ladvise */
3384 lio = &io->u.ci_ladvise;
3385 lio->li_start = ladvise->lla_start;
3386 lio->li_end = ladvise->lla_end;
3387 lio->li_fid = ll_inode2fid(inode);
3388 lio->li_advice = ladvise->lla_advice;
3389 lio->li_flags = flags;
3391 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3392 rc = cl_io_loop(env, io);
3396 cl_io_fini(env, io);
3397 cl_env_put(env, &refcheck);
3401 static int ll_lock_noexpand(struct file *file, int flags)
3403 struct ll_file_data *fd = file->private_data;
3405 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3410 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3413 struct fsxattr fsxattr;
3415 if (copy_from_user(&fsxattr,
3416 (const struct fsxattr __user *)arg,
3420 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3421 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3422 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3423 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3424 if (copy_to_user((struct fsxattr __user *)arg,
3425 &fsxattr, sizeof(fsxattr)))
3431 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3435 * Project Quota ID state is only allowed to change from within the init
3436 * namespace. Enforce that restriction only if we are trying to change
3437 * the quota ID state. Everything else is allowed in user namespaces.
3439 if (current_user_ns() == &init_user_ns) {
3441 * Caller is allowed to change the project ID. if it is being
3442 * changed, make sure that the new value is valid.
3444 if (ll_i2info(inode)->lli_projid != projid &&
3445 !projid_valid(make_kprojid(&init_user_ns, projid)))
3451 if (ll_i2info(inode)->lli_projid != projid)
3454 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3455 if (!(xflags & FS_XFLAG_PROJINHERIT))
3458 if (xflags & FS_XFLAG_PROJINHERIT)
3465 static int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3467 struct md_op_data *op_data;
3468 struct ptlrpc_request *req = NULL;
3469 struct cl_object *obj;
3470 unsigned int inode_flags;
3473 rc = ll_ioctl_check_project(inode, xflags, projid);
3477 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3478 LUSTRE_OPC_ANY, NULL);
3479 if (IS_ERR(op_data))
3480 RETURN(PTR_ERR(op_data));
3482 inode_flags = ll_xflags_to_inode_flags(xflags);
3483 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3484 if (xflags & FS_XFLAG_PROJINHERIT)
3485 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3486 op_data->op_projid = projid;
3487 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3488 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3489 ptlrpc_req_finished(req);
3491 GOTO(out_fsxattr, rc);
3492 ll_update_inode_flags(inode, op_data->op_attr_flags);
3494 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3495 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3496 GOTO(out_fsxattr, rc);
3498 obj = ll_i2info(inode)->lli_clob;
3500 struct iattr attr = { 0 };
3502 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3506 ll_finish_md_op_data(op_data);
3510 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3513 struct fsxattr fsxattr;
3517 if (copy_from_user(&fsxattr,
3518 (const struct fsxattr __user *)arg,
3522 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3523 fsxattr.fsx_projid));
3526 int ll_ioctl_project(struct file *file, unsigned int cmd,
3529 struct lu_project lu_project;
3530 struct dentry *dentry = file_dentry(file);
3531 struct inode *inode = file_inode(file);
3532 struct dentry *child_dentry = NULL;
3533 int rc = 0, name_len;
3535 if (copy_from_user(&lu_project,
3536 (const struct lu_project __user *)arg,
3537 sizeof(lu_project)))
3540 /* apply child dentry if name is valid */
3541 name_len = strnlen(lu_project.project_name, NAME_MAX);
3542 if (name_len > 0 && name_len <= NAME_MAX) {
3544 child_dentry = lookup_one_len(lu_project.project_name,
3546 inode_unlock(inode);
3547 if (IS_ERR(child_dentry)) {
3548 rc = PTR_ERR(child_dentry);
3551 inode = child_dentry->d_inode;
3556 } else if (name_len > NAME_MAX) {
3561 switch (lu_project.project_type) {
3562 case LU_PROJECT_SET:
3563 rc = ll_set_project(inode, lu_project.project_xflags,
3564 lu_project.project_id);
3566 case LU_PROJECT_GET:
3567 lu_project.project_xflags =
3568 ll_inode_flags_to_xflags(inode->i_flags);
3569 if (test_bit(LLIF_PROJECT_INHERIT,
3570 &ll_i2info(inode)->lli_flags))
3571 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3572 lu_project.project_id = ll_i2info(inode)->lli_projid;
3573 if (copy_to_user((struct lu_project __user *)arg,
3574 &lu_project, sizeof(lu_project))) {
3584 if (!IS_ERR_OR_NULL(child_dentry))
3589 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3592 struct inode *inode = file_inode(file);
3593 struct ll_file_data *fd = file->private_data;
3594 struct ll_inode_info *lli = ll_i2info(inode);
3595 struct obd_client_handle *och = NULL;
3596 struct split_param sp;
3597 struct pcc_param param;
3598 bool lease_broken = false;
3600 enum mds_op_bias bias = 0;
3601 struct file *layout_file = NULL;
3603 size_t data_size = 0;
3604 bool attached = false;
3609 mutex_lock(&lli->lli_och_mutex);
3610 if (fd->fd_lease_och != NULL) {
3611 och = fd->fd_lease_och;
3612 fd->fd_lease_och = NULL;
3614 mutex_unlock(&lli->lli_och_mutex);
3619 fmode = och->och_flags;
3621 switch (ioc->lil_flags) {
3622 case LL_LEASE_RESYNC_DONE:
3623 if (ioc->lil_count > IOC_IDS_MAX)
3624 GOTO(out_lease_close, rc = -EINVAL);
3626 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3627 OBD_ALLOC(data, data_size);
3629 GOTO(out_lease_close, rc = -ENOMEM);
3631 if (copy_from_user(data, (void __user *)arg, data_size))
3632 GOTO(out_lease_close, rc = -EFAULT);
3634 bias = MDS_CLOSE_RESYNC_DONE;
3636 case LL_LEASE_LAYOUT_MERGE: {
3639 if (ioc->lil_count != 1)
3640 GOTO(out_lease_close, rc = -EINVAL);
3642 arg += sizeof(*ioc);
3643 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3644 GOTO(out_lease_close, rc = -EFAULT);
3646 layout_file = fget(fd);
3648 GOTO(out_lease_close, rc = -EBADF);
3650 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3651 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3652 GOTO(out_lease_close, rc = -EPERM);
3654 data = file_inode(layout_file);
3655 bias = MDS_CLOSE_LAYOUT_MERGE;
3658 case LL_LEASE_LAYOUT_SPLIT: {
3662 if (ioc->lil_count != 2)
3663 GOTO(out_lease_close, rc = -EINVAL);
3665 arg += sizeof(*ioc);
3666 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3667 GOTO(out_lease_close, rc = -EFAULT);
3669 arg += sizeof(__u32);
3670 if (copy_from_user(&mirror_id, (void __user *)arg,
3672 GOTO(out_lease_close, rc = -EFAULT);
3674 layout_file = fget(fdv);
3676 GOTO(out_lease_close, rc = -EBADF);
3678 /* if layout_file == file, it means to destroy the mirror */
3679 sp.sp_inode = file_inode(layout_file);
3680 sp.sp_mirror_id = (__u16)mirror_id;
3682 bias = MDS_CLOSE_LAYOUT_SPLIT;
3685 case LL_LEASE_PCC_ATTACH:
3686 if (ioc->lil_count != 1)
3689 arg += sizeof(*ioc);
3690 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3692 GOTO(out_lease_close, rc2 = -EFAULT);
3694 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3696 GOTO(out_lease_close, rc2);
3699 /* Grab latest data version */
3700 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3703 GOTO(out_lease_close, rc2);
3706 bias = MDS_PCC_ATTACH;
3709 /* without close intent */
3714 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3718 rc = ll_lease_och_release(inode, file);
3727 switch (ioc->lil_flags) {
3728 case LL_LEASE_RESYNC_DONE:
3730 OBD_FREE(data, data_size);
3732 case LL_LEASE_LAYOUT_MERGE:
3733 case LL_LEASE_LAYOUT_SPLIT:
3737 ll_layout_refresh(inode, &fd->fd_layout_version);
3739 case LL_LEASE_PCC_ATTACH:
3742 rc = pcc_readwrite_attach_fini(file, inode,
3743 param.pa_layout_gen,
3750 rc = ll_lease_type_from_fmode(fmode);
3754 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3757 struct inode *inode = file_inode(file);
3758 struct ll_inode_info *lli = ll_i2info(inode);
3759 struct ll_file_data *fd = file->private_data;
3760 struct obd_client_handle *och = NULL;
3761 __u64 open_flags = 0;
3767 switch (ioc->lil_mode) {
3768 case LL_LEASE_WRLCK:
3769 if (!(file->f_mode & FMODE_WRITE))
3771 fmode = FMODE_WRITE;
3773 case LL_LEASE_RDLCK:
3774 if (!(file->f_mode & FMODE_READ))
3778 case LL_LEASE_UNLCK:
3779 RETURN(ll_file_unlock_lease(file, ioc, arg));
3784 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3786 /* apply for lease */
3787 if (ioc->lil_flags & LL_LEASE_RESYNC)
3788 open_flags = MDS_OPEN_RESYNC;
3789 och = ll_lease_open(inode, file, fmode, open_flags);
3791 RETURN(PTR_ERR(och));
3793 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3794 rc = ll_lease_file_resync(och, inode, arg);
3796 ll_lease_close(och, inode, NULL);
3799 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3801 ll_lease_close(och, inode, NULL);
3807 mutex_lock(&lli->lli_och_mutex);
3808 if (fd->fd_lease_och == NULL) {
3809 fd->fd_lease_och = och;
3812 mutex_unlock(&lli->lli_och_mutex);
3814 /* impossible now that only excl is supported for now */
3815 ll_lease_close(och, inode, &lease_broken);
3821 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3823 struct ll_inode_info *lli = ll_i2info(inode);
3824 struct ll_sb_info *sbi = ll_i2sbi(inode);
3825 __u64 now = ktime_get_real_seconds();
3828 spin_lock(&lli->lli_heat_lock);
3829 heat->lh_flags = lli->lli_heat_flags;
3830 for (i = 0; i < heat->lh_count; i++)
3831 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3832 now, sbi->ll_heat_decay_weight,
3833 sbi->ll_heat_period_second);
3834 spin_unlock(&lli->lli_heat_lock);
3837 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3839 struct ll_inode_info *lli = ll_i2info(inode);
3842 spin_lock(&lli->lli_heat_lock);
3843 if (flags & LU_HEAT_FLAG_CLEAR)
3844 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3846 if (flags & LU_HEAT_FLAG_OFF)
3847 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3849 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3851 spin_unlock(&lli->lli_heat_lock);
3857 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3859 struct inode *inode = file_inode(file);
3860 struct ll_file_data *fd = file->private_data;
3864 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3865 PFID(ll_inode2fid(inode)), inode, cmd);
3866 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3868 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3869 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3873 case LL_IOC_GETFLAGS:
3874 /* Get the current value of the file flags */
3875 return put_user(fd->fd_flags, (int __user *)arg);
3876 case LL_IOC_SETFLAGS:
3877 case LL_IOC_CLRFLAGS:
3878 /* Set or clear specific file flags */
3879 /* XXX This probably needs checks to ensure the flags are
3880 * not abused, and to handle any flag side effects.
3882 if (get_user(flags, (int __user *) arg))
3885 if (cmd == LL_IOC_SETFLAGS) {
3886 if ((flags & LL_FILE_IGNORE_LOCK) &&
3887 !(file->f_flags & O_DIRECT)) {
3888 CERROR("%s: unable to disable locking on "
3889 "non-O_DIRECT file\n", current->comm);
3893 fd->fd_flags |= flags;
3895 fd->fd_flags &= ~flags;
3898 case LL_IOC_LOV_SETSTRIPE:
3899 case LL_IOC_LOV_SETSTRIPE_NEW:
3900 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3901 case LL_IOC_LOV_SETEA:
3902 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3903 case LL_IOC_LOV_SWAP_LAYOUTS: {
3905 struct lustre_swap_layouts lsl;
3907 if (copy_from_user(&lsl, (char __user *)arg,
3908 sizeof(struct lustre_swap_layouts)))
3911 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3914 file2 = fget(lsl.sl_fd);
3918 /* O_WRONLY or O_RDWR */
3919 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3920 GOTO(out, rc = -EPERM);
3922 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3923 struct inode *inode2;
3924 struct ll_inode_info *lli;
3925 struct obd_client_handle *och = NULL;
3927 lli = ll_i2info(inode);
3928 mutex_lock(&lli->lli_och_mutex);
3929 if (fd->fd_lease_och != NULL) {
3930 och = fd->fd_lease_och;
3931 fd->fd_lease_och = NULL;
3933 mutex_unlock(&lli->lli_och_mutex);
3935 GOTO(out, rc = -ENOLCK);
3936 inode2 = file_inode(file2);
3937 rc = ll_swap_layouts_close(och, inode, inode2);
3939 rc = ll_swap_layouts(file, file2, &lsl);
3945 case LL_IOC_LOV_GETSTRIPE:
3946 case LL_IOC_LOV_GETSTRIPE_NEW:
3947 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3948 case FS_IOC_GETFLAGS:
3949 case FS_IOC_SETFLAGS:
3950 RETURN(ll_iocontrol(inode, file, cmd, arg));
3951 case FSFILT_IOC_GETVERSION:
3952 case FS_IOC_GETVERSION:
3953 RETURN(put_user(inode->i_generation, (int __user *)arg));
3954 /* We need to special case any other ioctls we want to handle,
3955 * to send them to the MDS/OST as appropriate and to properly
3956 * network encode the arg field. */
3957 case FS_IOC_SETVERSION:
3960 case LL_IOC_GROUP_LOCK:
3961 RETURN(ll_get_grouplock(inode, file, arg));
3962 case LL_IOC_GROUP_UNLOCK:
3963 RETURN(ll_put_grouplock(inode, file, arg));
3964 case IOC_OBD_STATFS:
3965 RETURN(ll_obd_statfs(inode, (void __user *)arg));
3967 case LL_IOC_FLUSHCTX:
3968 RETURN(ll_flush_ctx(inode));
3969 case LL_IOC_PATH2FID: {
3970 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
3971 sizeof(struct lu_fid)))
3976 case LL_IOC_GETPARENT:
3977 RETURN(ll_getparent(file, (struct getparent __user *)arg));
3979 case OBD_IOC_FID2PATH:
3980 RETURN(ll_fid2path(inode, (void __user *)arg));
3981 case LL_IOC_DATA_VERSION: {
3982 struct ioc_data_version idv;
3985 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
3988 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
3989 rc = ll_ioc_data_version(inode, &idv);
3992 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
3998 case LL_IOC_GET_MDTIDX: {
4001 mdtidx = ll_get_mdt_idx(inode);
4005 if (put_user((int)mdtidx, (int __user *)arg))
4010 case OBD_IOC_GETNAME_OLD:
4012 case OBD_IOC_GETDTNAME:
4014 case OBD_IOC_GETMDNAME:
4015 RETURN(ll_get_obd_name(inode, cmd, arg));
4016 case LL_IOC_HSM_STATE_GET: {
4017 struct md_op_data *op_data;
4018 struct hsm_user_state *hus;
4025 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4026 LUSTRE_OPC_ANY, hus);
4027 if (IS_ERR(op_data)) {
4029 RETURN(PTR_ERR(op_data));
4032 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4035 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
4038 ll_finish_md_op_data(op_data);
4042 case LL_IOC_HSM_STATE_SET: {
4043 struct hsm_state_set *hss;
4050 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
4055 rc = ll_hsm_state_set(inode, hss);
4060 case LL_IOC_HSM_ACTION: {
4061 struct md_op_data *op_data;
4062 struct hsm_current_action *hca;
4070 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4071 LUSTRE_OPC_ANY, hca);
4072 if (IS_ERR(op_data)) {
4074 RETURN(PTR_ERR(op_data));
4077 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4080 GOTO(skip_copy, rc);
4082 /* The hsm_current_action retreived from the server could
4083 * contain corrupt information. If it is incorrect data collect
4084 * debug information. We still send the data even if incorrect
4085 * to user land to handle.
4087 action = hsm_user_action2name(hca->hca_action);
4088 if (strcmp(action, "UNKNOWN") == 0 ||
4089 hca->hca_state > HPS_DONE) {
4091 "HSM current state %s action %s, offset = %llu, length %llu\n",
4092 hsm_progress_state2name(hca->hca_state), action,
4093 hca->hca_location.offset, hca->hca_location.length);
4096 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
4099 ll_finish_md_op_data(op_data);
4103 case LL_IOC_SET_LEASE_OLD: {
4104 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
4106 RETURN(ll_file_set_lease(file, &ioc, 0));
4108 case LL_IOC_SET_LEASE: {
4109 struct ll_ioc_lease ioc;
4111 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
4114 RETURN(ll_file_set_lease(file, &ioc, arg));
4116 case LL_IOC_GET_LEASE: {
4117 struct ll_inode_info *lli = ll_i2info(inode);
4118 struct ldlm_lock *lock = NULL;
4121 mutex_lock(&lli->lli_och_mutex);
4122 if (fd->fd_lease_och != NULL) {
4123 struct obd_client_handle *och = fd->fd_lease_och;
4125 lock = ldlm_handle2lock(&och->och_lease_handle);
4127 lock_res_and_lock(lock);
4128 if (!ldlm_is_cancel(lock))
4129 fmode = och->och_flags;
4131 unlock_res_and_lock(lock);
4132 LDLM_LOCK_PUT(lock);
4135 mutex_unlock(&lli->lli_och_mutex);
4137 RETURN(ll_lease_type_from_fmode(fmode));
4139 case LL_IOC_HSM_IMPORT: {
4140 struct hsm_user_import *hui;
4146 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
4151 rc = ll_hsm_import(inode, file, hui);
4156 case LL_IOC_FUTIMES_3: {
4157 struct ll_futimes_3 lfu;
4159 if (copy_from_user(&lfu,
4160 (const struct ll_futimes_3 __user *)arg,
4164 RETURN(ll_file_futimes_3(file, &lfu));
4166 case LL_IOC_LADVISE: {
4167 struct llapi_ladvise_hdr *k_ladvise_hdr;
4168 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4171 int alloc_size = sizeof(*k_ladvise_hdr);
4174 u_ladvise_hdr = (void __user *)arg;
4175 OBD_ALLOC_PTR(k_ladvise_hdr);
4176 if (k_ladvise_hdr == NULL)
4179 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4180 GOTO(out_ladvise, rc = -EFAULT);
4182 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4183 k_ladvise_hdr->lah_count < 1)
4184 GOTO(out_ladvise, rc = -EINVAL);
4186 num_advise = k_ladvise_hdr->lah_count;
4187 if (num_advise >= LAH_COUNT_MAX)
4188 GOTO(out_ladvise, rc = -EFBIG);
4190 OBD_FREE_PTR(k_ladvise_hdr);
4191 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4192 lah_advise[num_advise]);
4193 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4194 if (k_ladvise_hdr == NULL)
4198 * TODO: submit multiple advices to one server in a single RPC
4200 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4201 GOTO(out_ladvise, rc = -EFAULT);
4203 for (i = 0; i < num_advise; i++) {
4204 struct llapi_lu_ladvise *k_ladvise =
4205 &k_ladvise_hdr->lah_advise[i];
4206 struct llapi_lu_ladvise __user *u_ladvise =
4207 &u_ladvise_hdr->lah_advise[i];
4209 rc = ll_ladvise_sanity(inode, k_ladvise);
4211 GOTO(out_ladvise, rc);
4213 switch (k_ladvise->lla_advice) {
4214 case LU_LADVISE_LOCKNOEXPAND:
4215 rc = ll_lock_noexpand(file,
4216 k_ladvise->lla_peradvice_flags);
4217 GOTO(out_ladvise, rc);
4218 case LU_LADVISE_LOCKAHEAD:
4220 rc = ll_file_lock_ahead(file, k_ladvise);
4223 GOTO(out_ladvise, rc);
4226 &u_ladvise->lla_lockahead_result))
4227 GOTO(out_ladvise, rc = -EFAULT);
4230 rc = ll_ladvise(inode, file,
4231 k_ladvise_hdr->lah_flags,
4234 GOTO(out_ladvise, rc);
4241 OBD_FREE(k_ladvise_hdr, alloc_size);
4244 case LL_IOC_FLR_SET_MIRROR: {
4245 /* mirror I/O must be direct to avoid polluting page cache
4247 if (!(file->f_flags & O_DIRECT))
4250 fd->fd_designated_mirror = (__u32)arg;
4253 case FS_IOC_FSGETXATTR:
4254 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
4255 case FS_IOC_FSSETXATTR:
4256 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
4257 case LL_IOC_PROJECT:
4258 RETURN(ll_ioctl_project(file, cmd, arg));
4260 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
4261 case LL_IOC_HEAT_GET: {
4262 struct lu_heat uheat;
4263 struct lu_heat *heat;
4266 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
4269 if (uheat.lh_count > OBD_HEAT_COUNT)
4270 uheat.lh_count = OBD_HEAT_COUNT;
4272 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4273 OBD_ALLOC(heat, size);
4277 heat->lh_count = uheat.lh_count;
4278 ll_heat_get(inode, heat);
4279 rc = copy_to_user((char __user *)arg, heat, size);
4280 OBD_FREE(heat, size);
4281 RETURN(rc ? -EFAULT : 0);
4283 case LL_IOC_HEAT_SET: {
4286 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
4289 rc = ll_heat_set(inode, flags);
4292 case LL_IOC_PCC_DETACH: {
4293 struct lu_pcc_detach *detach;
4295 OBD_ALLOC_PTR(detach);
4299 if (copy_from_user(detach,
4300 (const struct lu_pcc_detach __user *)arg,
4302 GOTO(out_detach_free, rc = -EFAULT);
4304 if (!S_ISREG(inode->i_mode))
4305 GOTO(out_detach_free, rc = -EINVAL);
4307 if (!inode_owner_or_capable(inode))
4308 GOTO(out_detach_free, rc = -EPERM);
4310 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4312 OBD_FREE_PTR(detach);
4315 case LL_IOC_PCC_STATE: {
4316 struct lu_pcc_state __user *ustate =
4317 (struct lu_pcc_state __user *)arg;
4318 struct lu_pcc_state *state;
4320 OBD_ALLOC_PTR(state);
4324 if (copy_from_user(state, ustate, sizeof(*state)))
4325 GOTO(out_state, rc = -EFAULT);
4327 rc = pcc_ioctl_state(file, inode, state);
4329 GOTO(out_state, rc);
4331 if (copy_to_user(ustate, state, sizeof(*state)))
4332 GOTO(out_state, rc = -EFAULT);
4335 OBD_FREE_PTR(state);
4338 #ifdef HAVE_LUSTRE_CRYPTO
4339 case LL_IOC_SET_ENCRYPTION_POLICY:
4340 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4342 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
4343 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
4344 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4346 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
4347 case LL_IOC_ADD_ENCRYPTION_KEY:
4348 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4350 return llcrypt_ioctl_add_key(file, (void __user *)arg);
4351 case LL_IOC_REMOVE_ENCRYPTION_KEY:
4352 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4354 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
4355 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
4356 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4358 return llcrypt_ioctl_remove_key_all_users(file,
4359 (void __user *)arg);
4360 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
4361 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4363 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
4366 case LL_IOC_UNLOCK_FOREIGN: {
4367 struct dentry *dentry = file_dentry(file);
4369 /* if not a foreign symlink do nothing */
4370 if (ll_foreign_is_removable(dentry, true)) {
4372 "prevent unlink of non-foreign file ("DFID")\n",
4373 PFID(ll_inode2fid(inode)));
4374 RETURN(-EOPNOTSUPP);
4380 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
4381 (void __user *)arg));
4385 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4387 struct inode *inode = file_inode(file);
4390 struct cl_lseek_io *lsio;
4397 env = cl_env_get(&refcheck);
4399 RETURN(PTR_ERR(env));
4401 io = vvp_env_thread_io(env);
4402 io->ci_obj = ll_i2info(inode)->lli_clob;
4403 ll_io_set_mirror(io, file);
4405 lsio = &io->u.ci_lseek;
4406 lsio->ls_start = offset;
4407 lsio->ls_whence = whence;
4408 lsio->ls_result = -ENXIO;
4411 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4413 struct vvp_io *vio = vvp_env_io(env);
4415 vio->vui_fd = file->private_data;
4416 rc = cl_io_loop(env, io);
4420 retval = rc ? : lsio->ls_result;
4421 cl_io_fini(env, io);
4422 } while (unlikely(io->ci_need_restart));
4424 cl_env_put(env, &refcheck);
4429 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4431 struct inode *inode = file_inode(file);
4432 loff_t retval = offset, eof = 0;
4433 ktime_t kstart = ktime_get();
4437 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4438 PFID(ll_inode2fid(inode)), inode, retval, retval,
4441 if (origin == SEEK_END) {
4442 retval = ll_glimpse_size(inode);
4445 eof = i_size_read(inode);
4448 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4452 /* flush local cache first if any */
4453 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4456 retval = ll_lseek(file, offset, origin);
4459 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4461 retval = generic_file_llseek_size(file, offset, origin,
4462 ll_file_maxbytes(inode), eof);
4465 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4466 ktime_us_delta(ktime_get(), kstart));
4470 static int ll_flush(struct file *file, fl_owner_t id)
4472 struct inode *inode = file_inode(file);
4473 struct ll_inode_info *lli = ll_i2info(inode);
4474 struct ll_file_data *fd = file->private_data;
4477 LASSERT(!S_ISDIR(inode->i_mode));
4479 /* catch async errors that were recorded back when async writeback
4480 * failed for pages in this mapping. */
4481 rc = lli->lli_async_rc;
4482 lli->lli_async_rc = 0;
4483 if (lli->lli_clob != NULL) {
4484 err = lov_read_and_clear_async_rc(lli->lli_clob);
4489 /* The application has been told write failure already.
4490 * Do not report failure again. */
4491 if (fd->fd_write_failed)
4493 return rc ? -EIO : 0;
4497 * Called to make sure a portion of file has been written out.
4498 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4500 * Return how many pages have been written.
4502 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4503 enum cl_fsync_mode mode, int ignore_layout)
4507 struct cl_fsync_io *fio;
4512 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4513 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4516 env = cl_env_get(&refcheck);
4518 RETURN(PTR_ERR(env));
4520 io = vvp_env_thread_io(env);
4521 io->ci_obj = ll_i2info(inode)->lli_clob;
4522 io->ci_ignore_layout = ignore_layout;
4524 /* initialize parameters for sync */
4525 fio = &io->u.ci_fsync;
4526 fio->fi_start = start;
4528 fio->fi_fid = ll_inode2fid(inode);
4529 fio->fi_mode = mode;
4530 fio->fi_nr_written = 0;
4532 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4533 result = cl_io_loop(env, io);
4535 result = io->ci_result;
4537 result = fio->fi_nr_written;
4538 cl_io_fini(env, io);
4539 cl_env_put(env, &refcheck);
4545 * When dentry is provided (the 'else' case), file_dentry() may be
4546 * null and dentry must be used directly rather than pulled from
4547 * file_dentry() as is done otherwise.
4550 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4552 struct dentry *dentry = file_dentry(file);
4553 struct inode *inode = dentry->d_inode;
4554 struct ll_inode_info *lli = ll_i2info(inode);
4555 struct ptlrpc_request *req;
4556 ktime_t kstart = ktime_get();
4562 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4563 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4565 /* fsync's caller has already called _fdata{sync,write}, we want
4566 * that IO to finish before calling the osc and mdc sync methods */
4567 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4569 /* catch async errors that were recorded back when async writeback
4570 * failed for pages in this mapping. */
4571 if (!S_ISDIR(inode->i_mode)) {
4572 err = lli->lli_async_rc;
4573 lli->lli_async_rc = 0;
4576 if (lli->lli_clob != NULL) {
4577 err = lov_read_and_clear_async_rc(lli->lli_clob);
4583 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4587 ptlrpc_req_finished(req);
4589 if (S_ISREG(inode->i_mode)) {
4590 struct ll_file_data *fd = file->private_data;
4593 /* Sync metadata on MDT first, and then sync the cached data
4596 err = pcc_fsync(file, start, end, datasync, &cached);
4598 err = cl_sync_file_range(inode, start, end,
4600 if (rc == 0 && err < 0)
4603 fd->fd_write_failed = true;
4605 fd->fd_write_failed = false;
4609 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4610 ktime_us_delta(ktime_get(), kstart));
4615 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4617 struct inode *inode = file_inode(file);
4618 struct ll_sb_info *sbi = ll_i2sbi(inode);
4619 struct ldlm_enqueue_info einfo = {
4620 .ei_type = LDLM_FLOCK,
4621 .ei_cb_cp = ldlm_flock_completion_ast,
4622 .ei_cbdata = file_lock,
4624 struct md_op_data *op_data;
4625 struct lustre_handle lockh = { 0 };
4626 union ldlm_policy_data flock = { { 0 } };
4627 int fl_type = file_lock->fl_type;
4628 ktime_t kstart = ktime_get();
4634 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4635 PFID(ll_inode2fid(inode)), file_lock);
4637 if (file_lock->fl_flags & FL_FLOCK) {
4638 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4639 /* flocks are whole-file locks */
4640 flock.l_flock.end = OFFSET_MAX;
4641 /* For flocks owner is determined by the local file desctiptor*/
4642 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4643 } else if (file_lock->fl_flags & FL_POSIX) {
4644 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4645 flock.l_flock.start = file_lock->fl_start;
4646 flock.l_flock.end = file_lock->fl_end;
4650 flock.l_flock.pid = file_lock->fl_pid;
4652 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4653 /* Somewhat ugly workaround for svc lockd.
4654 * lockd installs custom fl_lmops->lm_compare_owner that checks
4655 * for the fl_owner to be the same (which it always is on local node
4656 * I guess between lockd processes) and then compares pid.
4657 * As such we assign pid to the owner field to make it all work,
4658 * conflict with normal locks is unlikely since pid space and
4659 * pointer space for current->files are not intersecting */
4660 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4661 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4666 einfo.ei_mode = LCK_PR;
4669 /* An unlock request may or may not have any relation to
4670 * existing locks so we may not be able to pass a lock handle
4671 * via a normal ldlm_lock_cancel() request. The request may even
4672 * unlock a byte range in the middle of an existing lock. In
4673 * order to process an unlock request we need all of the same
4674 * information that is given with a normal read or write record
4675 * lock request. To avoid creating another ldlm unlock (cancel)
4676 * message we'll treat a LCK_NL flock request as an unlock. */
4677 einfo.ei_mode = LCK_NL;
4680 einfo.ei_mode = LCK_PW;
4683 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4698 flags = LDLM_FL_BLOCK_NOWAIT;
4704 flags = LDLM_FL_TEST_LOCK;
4707 CERROR("unknown fcntl lock command: %d\n", cmd);
4711 /* Save the old mode so that if the mode in the lock changes we
4712 * can decrement the appropriate reader or writer refcount. */
4713 file_lock->fl_type = einfo.ei_mode;
4715 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4716 LUSTRE_OPC_ANY, NULL);
4717 if (IS_ERR(op_data))
4718 RETURN(PTR_ERR(op_data));
4720 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4721 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4722 flock.l_flock.pid, flags, einfo.ei_mode,
4723 flock.l_flock.start, flock.l_flock.end);
4725 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4728 /* Restore the file lock type if not TEST lock. */
4729 if (!(flags & LDLM_FL_TEST_LOCK))
4730 file_lock->fl_type = fl_type;
4732 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4733 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4734 !(flags & LDLM_FL_TEST_LOCK))
4735 rc2 = locks_lock_file_wait(file, file_lock);
4737 if ((file_lock->fl_flags & FL_FLOCK) &&
4738 (rc == 0 || file_lock->fl_type == F_UNLCK))
4739 rc2 = flock_lock_file_wait(file, file_lock);
4740 if ((file_lock->fl_flags & FL_POSIX) &&
4741 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4742 !(flags & LDLM_FL_TEST_LOCK))
4743 rc2 = posix_lock_file_wait(file, file_lock);
4744 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4746 if (rc2 && file_lock->fl_type != F_UNLCK) {
4747 einfo.ei_mode = LCK_NL;
4748 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4753 ll_finish_md_op_data(op_data);
4756 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4757 ktime_us_delta(ktime_get(), kstart));
4761 int ll_get_fid_by_name(struct inode *parent, const char *name,
4762 int namelen, struct lu_fid *fid,
4763 struct inode **inode)
4765 struct md_op_data *op_data = NULL;
4766 struct mdt_body *body;
4767 struct ptlrpc_request *req;
4771 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4772 LUSTRE_OPC_ANY, NULL);
4773 if (IS_ERR(op_data))
4774 RETURN(PTR_ERR(op_data));
4776 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4777 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4778 ll_finish_md_op_data(op_data);
4782 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4784 GOTO(out_req, rc = -EFAULT);
4786 *fid = body->mbo_fid1;
4789 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
4791 ptlrpc_req_finished(req);
4795 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4798 struct dentry *dchild = NULL;
4799 struct inode *child_inode = NULL;
4800 struct md_op_data *op_data;
4801 struct ptlrpc_request *request = NULL;
4802 struct obd_client_handle *och = NULL;
4804 struct mdt_body *body;
4805 __u64 data_version = 0;
4806 size_t namelen = strlen(name);
4807 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4811 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4812 PFID(ll_inode2fid(parent)), name,
4813 lum->lum_stripe_offset, lum->lum_stripe_count);
4815 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4816 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4817 lustre_swab_lmv_user_md(lum);
4819 /* Get child FID first */
4820 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4823 dchild = d_lookup(file_dentry(file), &qstr);
4825 if (dchild->d_inode)
4826 child_inode = igrab(dchild->d_inode);
4831 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4840 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4841 OBD_CONNECT2_DIR_MIGRATE)) {
4842 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4843 ll_dir_striped(child_inode)) {
4844 CERROR("%s: MDT doesn't support stripe directory "
4845 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4846 GOTO(out_iput, rc = -EOPNOTSUPP);
4851 * lfs migrate command needs to be blocked on the client
4852 * by checking the migrate FID against the FID of the
4855 if (is_root_inode(child_inode))
4856 GOTO(out_iput, rc = -EINVAL);
4858 if (IS_ENCRYPTED(child_inode)) {
4859 rc = llcrypt_get_encryption_info(child_inode);
4862 if (!llcrypt_has_encryption_key(child_inode)) {
4863 CDEBUG(D_SEC, "no enc key for "DFID"\n",
4864 PFID(ll_inode2fid(child_inode)));
4865 GOTO(out_iput, rc = -ENOKEY);
4867 if (unlikely(!llcrypt_policy_has_filename_enc(child_inode))) {
4869 "cannot migrate old format encrypted "DFID", please move to new enc dir first\n",
4870 PFID(ll_inode2fid(child_inode)));
4871 GOTO(out_iput, rc = -EUCLEAN);
4875 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4876 child_inode->i_mode, LUSTRE_OPC_MIGR,
4878 if (IS_ERR(op_data))
4879 GOTO(out_iput, rc = PTR_ERR(op_data));
4881 inode_lock(child_inode);
4882 op_data->op_fid3 = *ll_inode2fid(child_inode);
4883 if (!fid_is_sane(&op_data->op_fid3)) {
4884 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4885 ll_i2sbi(parent)->ll_fsname, name,
4886 PFID(&op_data->op_fid3));
4887 GOTO(out_unlock, rc = -EINVAL);
4890 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4891 op_data->op_data = lum;
4892 op_data->op_data_size = lumlen;
4895 if (S_ISREG(child_inode->i_mode)) {
4896 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4900 GOTO(out_unlock, rc);
4903 rc = ll_data_version(child_inode, &data_version,
4906 GOTO(out_close, rc);
4908 op_data->op_open_handle = och->och_open_handle;
4909 op_data->op_data_version = data_version;
4910 op_data->op_lease_handle = och->och_lease_handle;
4911 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4913 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4914 och->och_mod->mod_open_req->rq_replay = 0;
4915 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4918 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
4919 op_data->op_name, op_data->op_namelen,
4920 op_data->op_name, op_data->op_namelen, &request);
4922 LASSERT(request != NULL);
4923 ll_update_times(request, parent);
4926 if (rc == 0 || rc == -EAGAIN) {
4927 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4928 LASSERT(body != NULL);
4930 /* If the server does release layout lock, then we cleanup
4931 * the client och here, otherwise release it in out_close: */
4932 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4933 obd_mod_put(och->och_mod);
4934 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4936 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4942 if (request != NULL) {
4943 ptlrpc_req_finished(request);
4947 /* Try again if the lease has cancelled. */
4948 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4953 ll_lease_close(och, child_inode, NULL);
4955 clear_nlink(child_inode);
4957 inode_unlock(child_inode);
4958 ll_finish_md_op_data(op_data);
4965 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4967 struct ll_file_data *fd = file->private_data;
4971 * In order to avoid flood of warning messages, only print one message
4972 * for one file. And the entire message rate on the client is limited
4973 * by CDEBUG_LIMIT too.
4975 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
4976 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
4977 CDEBUG_LIMIT(D_TTY | D_CONSOLE,
4978 "flock disabled, mount with '-o [local]flock' to enable\r\n");
4984 * test if some locks matching bits and l_req_mode are acquired
4985 * - bits can be in different locks
4986 * - if found clear the common lock bits in *bits
4987 * - the bits not found, are kept in *bits
4989 * \param bits [IN] searched lock bits [IN]
4990 * \param l_req_mode [IN] searched lock mode
4991 * \retval boolean, true iff all bits are found
4993 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
4995 struct lustre_handle lockh;
4996 union ldlm_policy_data policy;
4997 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
4998 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5007 fid = &ll_i2info(inode)->lli_fid;
5008 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5009 ldlm_lockname[mode]);
5011 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5012 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5013 policy.l_inodebits.bits = *bits & BIT(i);
5014 if (policy.l_inodebits.bits == 0)
5017 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
5018 &policy, mode, &lockh)) {
5019 struct ldlm_lock *lock;
5021 lock = ldlm_handle2lock(&lockh);
5024 ~(lock->l_policy_data.l_inodebits.bits);
5025 LDLM_LOCK_PUT(lock);
5027 *bits &= ~policy.l_inodebits.bits;
5034 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5035 struct lustre_handle *lockh, __u64 flags,
5036 enum ldlm_mode mode)
5038 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5043 fid = &ll_i2info(inode)->lli_fid;
5044 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5046 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5047 fid, LDLM_IBITS, &policy, mode, lockh);
5052 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5054 /* Already unlinked. Just update nlink and return success */
5055 if (rc == -ENOENT) {
5057 /* If it is striped directory, and there is bad stripe
5058 * Let's revalidate the dentry again, instead of returning
5060 if (ll_dir_striped(inode))
5063 /* This path cannot be hit for regular files unless in
5064 * case of obscure races, so no need to to validate
5066 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5068 } else if (rc != 0) {
5069 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5070 "%s: revalidate FID "DFID" error: rc = %d\n",
5071 ll_i2sbi(inode)->ll_fsname,
5072 PFID(ll_inode2fid(inode)), rc);
5078 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5080 struct inode *parent;
5081 struct inode *inode = dentry->d_inode;
5082 struct obd_export *exp = ll_i2mdexp(inode);
5083 struct lookup_intent oit = {
5086 struct ptlrpc_request *req = NULL;
5087 struct md_op_data *op_data;
5088 const char *name = NULL;
5093 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5094 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5096 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5097 parent = dentry->d_parent->d_inode;
5098 name = dentry->d_name.name;
5099 namelen = dentry->d_name.len;
5104 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5105 LUSTRE_OPC_ANY, NULL);
5106 if (IS_ERR(op_data))
5107 RETURN(PTR_ERR(op_data));
5109 /* Call getattr by fid */
5110 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5111 op_data->op_flags = MF_GETATTR_BY_FID;
5112 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5113 ll_finish_md_op_data(op_data);
5115 rc = ll_inode_revalidate_fini(inode, rc);
5119 rc = ll_revalidate_it_finish(req, &oit, dentry);
5121 ll_intent_release(&oit);
5125 /* Unlinked? Unhash dentry, so it is not picked up later by
5126 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5127 * here to preserve get_cwd functionality on 2.6.
5129 if (!dentry->d_inode->i_nlink)
5130 d_lustre_invalidate(dentry);
5132 ll_lookup_finish_locks(&oit, dentry);
5134 ptlrpc_req_finished(req);
5139 static int ll_merge_md_attr(struct inode *inode)
5141 struct ll_inode_info *lli = ll_i2info(inode);
5142 struct cl_attr attr = { 0 };
5145 LASSERT(lli->lli_lsm_md != NULL);
5147 if (!lmv_dir_striped(lli->lli_lsm_md))
5150 down_read(&lli->lli_lsm_sem);
5151 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
5152 &attr, ll_md_blocking_ast);
5153 up_read(&lli->lli_lsm_sem);
5157 set_nlink(inode, attr.cat_nlink);
5158 inode->i_blocks = attr.cat_blocks;
5159 i_size_write(inode, attr.cat_size);
5161 ll_i2info(inode)->lli_atime = attr.cat_atime;
5162 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5163 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5168 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5169 unsigned int flags, bool foreign)
5171 struct inode *inode = de->d_inode;
5172 struct ll_sb_info *sbi = ll_i2sbi(inode);
5173 struct ll_inode_info *lli = ll_i2info(inode);
5174 struct inode *dir = de->d_parent->d_inode;
5175 bool need_glimpse = true;
5176 ktime_t kstart = ktime_get();
5179 /* The OST object(s) determine the file size, blocks and mtime. */
5180 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5181 request_mask & STATX_MTIME))
5182 need_glimpse = false;
5184 if (dentry_may_statahead(dir, de))
5185 ll_start_statahead(dir, de, need_glimpse &&
5186 !(flags & AT_STATX_DONT_SYNC));
5188 if (flags & AT_STATX_DONT_SYNC)
5189 GOTO(fill_attr, rc = 0);
5191 rc = ll_inode_revalidate(de, IT_GETATTR);
5195 /* foreign file/dir are always of zero length, so don't
5196 * need to validate size.
5198 if (S_ISREG(inode->i_mode) && !foreign) {
5202 GOTO(fill_attr, rc);
5204 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5205 if (cached && rc < 0)
5209 GOTO(fill_attr, rc);
5212 * If the returned attr is masked with OBD_MD_FLSIZE &
5213 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5214 * or blocks obtained from MDT is strictly correct, and the file
5215 * is usually not being modified by clients, and the [a|m|c]time
5216 * got from MDT is also strictly correct.
5217 * Under this circumstance, it does not need to send glimpse
5218 * RPCs to OSTs for file attributes such as the size and blocks.
5220 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5221 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5222 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5223 inode->i_mtime.tv_sec = lli->lli_mtime;
5224 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5225 inode->i_atime.tv_sec = lli->lli_atime;
5226 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5227 inode->i_ctime.tv_sec = lli->lli_ctime;
5228 GOTO(fill_attr, rc);
5231 /* In case of restore, the MDT has the right size and has
5232 * already send it back without granting the layout lock,
5233 * inode is up-to-date so glimpse is useless.
5234 * Also to glimpse we need the layout, in case of a running
5235 * restore the MDT holds the layout lock so the glimpse will
5236 * block up to the end of restore (getattr will block)
5238 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5239 rc = ll_glimpse_size(inode);
5244 /* If object isn't regular a file then don't validate size. */
5245 /* foreign dir is not striped dir */
5246 if (ll_dir_striped(inode) && !foreign) {
5247 rc = ll_merge_md_attr(inode);
5252 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5253 inode->i_atime.tv_sec = lli->lli_atime;
5254 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5255 inode->i_mtime.tv_sec = lli->lli_mtime;
5256 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5257 inode->i_ctime.tv_sec = lli->lli_ctime;
5261 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5263 if (ll_need_32bit_api(sbi)) {
5264 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5265 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5266 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5268 stat->ino = inode->i_ino;
5269 stat->dev = inode->i_sb->s_dev;
5270 stat->rdev = inode->i_rdev;
5273 /* foreign symlink to be exposed as a real symlink */
5275 stat->mode = inode->i_mode;
5277 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5279 stat->uid = inode->i_uid;
5280 stat->gid = inode->i_gid;
5281 stat->atime = inode->i_atime;
5282 stat->mtime = inode->i_mtime;
5283 stat->ctime = inode->i_ctime;
5284 /* stat->blksize is used to tell about preferred IO size */
5285 if (sbi->ll_stat_blksize)
5286 stat->blksize = sbi->ll_stat_blksize;
5287 else if (S_ISREG(inode->i_mode))
5288 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5289 LL_MAX_BLKSIZE_BITS);
5291 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5293 stat->nlink = inode->i_nlink;
5294 stat->size = i_size_read(inode);
5295 stat->blocks = inode->i_blocks;
5297 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
5298 if (flags & AT_STATX_DONT_SYNC) {
5299 if (stat->size == 0 &&
5300 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5301 stat->size = lli->lli_lazysize;
5302 if (stat->blocks == 0 &&
5303 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5304 stat->blocks = lli->lli_lazyblocks;
5307 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5308 stat->result_mask |= STATX_BTIME;
5309 stat->btime.tv_sec = lli->lli_btime;
5312 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5313 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5314 stat->result_mask &= request_mask;
5317 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5318 ktime_us_delta(ktime_get(), kstart));
5323 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
5324 int ll_getattr(const struct path *path, struct kstat *stat,
5325 u32 request_mask, unsigned int flags)
5327 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5331 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5333 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5334 AT_STATX_SYNC_AS_STAT, false);
5338 int cl_falloc(struct inode *inode, int mode, loff_t offset, loff_t len)
5344 loff_t size = i_size_read(inode);
5348 env = cl_env_get(&refcheck);
5350 RETURN(PTR_ERR(env));
5352 io = vvp_env_thread_io(env);
5353 io->ci_obj = ll_i2info(inode)->lli_clob;
5354 io->ci_verify_layout = 1;
5355 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5356 io->u.ci_setattr.sa_falloc_mode = mode;
5357 io->u.ci_setattr.sa_falloc_offset = offset;
5358 io->u.ci_setattr.sa_falloc_end = offset + len;
5359 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5360 if (io->u.ci_setattr.sa_falloc_end > size) {
5361 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5363 /* Check new size against VFS/VM file size limit and rlimit */
5364 rc = inode_newsize_ok(inode, newsize);
5367 if (newsize > ll_file_maxbytes(inode)) {
5368 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5369 (unsigned long long)newsize,
5370 ll_file_maxbytes(inode));
5377 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5379 rc = cl_io_loop(env, io);
5382 cl_io_fini(env, io);
5383 } while (unlikely(io->ci_need_restart));
5386 cl_env_put(env, &refcheck);
5390 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5392 struct inode *inode = filp->f_path.dentry->d_inode;
5395 if (offset < 0 || len <= 0)
5398 * Encrypted inodes can't handle collapse range or zero range or insert
5399 * range since we would need to re-encrypt blocks with a different IV or
5400 * XTS tweak (which are based on the logical block number).
5401 * Similar to what ext4 does.
5403 if (IS_ENCRYPTED(inode) &&
5404 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5405 FALLOC_FL_ZERO_RANGE)))
5406 RETURN(-EOPNOTSUPP);
5409 * mode == 0 (which is standard prealloc) and PUNCH is supported
5410 * Rest of mode options are not supported yet.
5412 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5413 RETURN(-EOPNOTSUPP);
5415 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5417 rc = cl_falloc(inode, mode, offset, len);
5419 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5420 * used by Lustre in several places. Retuning it here would
5421 * confuse applications that explicity test for EOPNOTSUPP
5422 * (95) and fall back to ftruncate().
5424 if (rc == -ENOTSUPP)
5430 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5431 __u64 start, __u64 len)
5435 struct fiemap *fiemap;
5436 unsigned int extent_count = fieinfo->fi_extents_max;
5438 num_bytes = sizeof(*fiemap) + (extent_count *
5439 sizeof(struct fiemap_extent));
5440 OBD_ALLOC_LARGE(fiemap, num_bytes);
5445 fiemap->fm_flags = fieinfo->fi_flags;
5446 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5447 fiemap->fm_start = start;
5448 fiemap->fm_length = len;
5449 if (extent_count > 0 &&
5450 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5451 sizeof(struct fiemap_extent)) != 0)
5452 GOTO(out, rc = -EFAULT);
5454 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5456 if (IS_ENCRYPTED(inode)) {
5459 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5460 fiemap->fm_extents[i].fe_flags |=
5461 FIEMAP_EXTENT_DATA_ENCRYPTED |
5462 FIEMAP_EXTENT_ENCODED;
5465 fieinfo->fi_flags = fiemap->fm_flags;
5466 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5467 if (extent_count > 0 &&
5468 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5469 fiemap->fm_mapped_extents *
5470 sizeof(struct fiemap_extent)) != 0)
5471 GOTO(out, rc = -EFAULT);
5473 OBD_FREE_LARGE(fiemap, num_bytes);
5477 int ll_inode_permission(struct inode *inode, int mask)
5480 struct ll_sb_info *sbi;
5481 struct root_squash_info *squash;
5482 struct cred *cred = NULL;
5483 const struct cred *old_cred = NULL;
5484 bool squash_id = false;
5485 ktime_t kstart = ktime_get();
5489 if (mask & MAY_NOT_BLOCK)
5493 * as root inode are NOT getting validated in lookup operation,
5494 * need to do it before permission check.
5497 if (is_root_inode(inode)) {
5498 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
5503 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5504 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5506 /* squash fsuid/fsgid if needed */
5507 sbi = ll_i2sbi(inode);
5508 squash = &sbi->ll_squash;
5509 if (unlikely(squash->rsi_uid != 0 &&
5510 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5511 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
5515 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5516 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5517 squash->rsi_uid, squash->rsi_gid);
5519 /* update current process's credentials
5520 * and FS capability */
5521 cred = prepare_creds();
5525 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5526 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5527 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5528 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5530 old_cred = override_creds(cred);
5533 rc = generic_permission(inode, mask);
5534 /* restore current process's credentials and FS capability */
5536 revert_creds(old_cred);
5541 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5542 ktime_us_delta(ktime_get(), kstart));
5547 /* -o localflock - only provides locally consistent flock locks */
5548 static const struct file_operations ll_file_operations = {
5549 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5550 # ifdef HAVE_SYNC_READ_WRITE
5551 .read = new_sync_read,
5552 .write = new_sync_write,
5554 .read_iter = ll_file_read_iter,
5555 .write_iter = ll_file_write_iter,
5556 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5557 .read = ll_file_read,
5558 .aio_read = ll_file_aio_read,
5559 .write = ll_file_write,
5560 .aio_write = ll_file_aio_write,
5561 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5562 .unlocked_ioctl = ll_file_ioctl,
5563 .open = ll_file_open,
5564 .release = ll_file_release,
5565 .mmap = ll_file_mmap,
5566 .llseek = ll_file_seek,
5567 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5568 .splice_read = generic_file_splice_read,
5570 .splice_read = pcc_file_splice_read,
5574 .fallocate = ll_fallocate,
5577 static const struct file_operations ll_file_operations_flock = {
5578 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5579 # ifdef HAVE_SYNC_READ_WRITE
5580 .read = new_sync_read,
5581 .write = new_sync_write,
5582 # endif /* HAVE_SYNC_READ_WRITE */
5583 .read_iter = ll_file_read_iter,
5584 .write_iter = ll_file_write_iter,
5585 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5586 .read = ll_file_read,
5587 .aio_read = ll_file_aio_read,
5588 .write = ll_file_write,
5589 .aio_write = ll_file_aio_write,
5590 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5591 .unlocked_ioctl = ll_file_ioctl,
5592 .open = ll_file_open,
5593 .release = ll_file_release,
5594 .mmap = ll_file_mmap,
5595 .llseek = ll_file_seek,
5596 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5597 .splice_read = generic_file_splice_read,
5599 .splice_read = pcc_file_splice_read,
5603 .flock = ll_file_flock,
5604 .lock = ll_file_flock,
5605 .fallocate = ll_fallocate,
5608 /* These are for -o noflock - to return ENOSYS on flock calls */
5609 static const struct file_operations ll_file_operations_noflock = {
5610 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5611 # ifdef HAVE_SYNC_READ_WRITE
5612 .read = new_sync_read,
5613 .write = new_sync_write,
5614 # endif /* HAVE_SYNC_READ_WRITE */
5615 .read_iter = ll_file_read_iter,
5616 .write_iter = ll_file_write_iter,
5617 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5618 .read = ll_file_read,
5619 .aio_read = ll_file_aio_read,
5620 .write = ll_file_write,
5621 .aio_write = ll_file_aio_write,
5622 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5623 .unlocked_ioctl = ll_file_ioctl,
5624 .open = ll_file_open,
5625 .release = ll_file_release,
5626 .mmap = ll_file_mmap,
5627 .llseek = ll_file_seek,
5628 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5629 .splice_read = generic_file_splice_read,
5631 .splice_read = pcc_file_splice_read,
5635 .flock = ll_file_noflock,
5636 .lock = ll_file_noflock,
5637 .fallocate = ll_fallocate,
5640 const struct inode_operations ll_file_inode_operations = {
5641 .setattr = ll_setattr,
5642 .getattr = ll_getattr,
5643 .permission = ll_inode_permission,
5644 #ifdef HAVE_IOP_XATTR
5645 .setxattr = ll_setxattr,
5646 .getxattr = ll_getxattr,
5647 .removexattr = ll_removexattr,
5649 .listxattr = ll_listxattr,
5650 .fiemap = ll_fiemap,
5651 .get_acl = ll_get_acl,
5652 #ifdef HAVE_IOP_SET_ACL
5653 .set_acl = ll_set_acl,
5657 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5659 const struct file_operations *fops = &ll_file_operations_noflock;
5661 if (sbi->ll_flags & LL_SBI_FLOCK)
5662 fops = &ll_file_operations_flock;
5663 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
5664 fops = &ll_file_operations;
5669 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5671 struct ll_inode_info *lli = ll_i2info(inode);
5672 struct cl_object *obj = lli->lli_clob;
5681 env = cl_env_get(&refcheck);
5683 RETURN(PTR_ERR(env));
5685 rc = cl_conf_set(env, lli->lli_clob, conf);
5689 if (conf->coc_opc == OBJECT_CONF_SET) {
5690 struct ldlm_lock *lock = conf->coc_lock;
5691 struct cl_layout cl = {
5695 LASSERT(lock != NULL);
5696 LASSERT(ldlm_has_layout(lock));
5698 /* it can only be allowed to match after layout is
5699 * applied to inode otherwise false layout would be
5700 * seen. Applying layout shoud happen before dropping
5701 * the intent lock. */
5702 ldlm_lock_allow_match(lock);
5704 rc = cl_object_layout_get(env, obj, &cl);
5709 DFID": layout version change: %u -> %u\n",
5710 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5712 ll_layout_version_set(lli, cl.cl_layout_gen);
5716 cl_env_put(env, &refcheck);
5718 RETURN(rc < 0 ? rc : 0);
5721 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5722 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5725 struct ll_sb_info *sbi = ll_i2sbi(inode);
5726 struct ptlrpc_request *req;
5733 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5734 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5735 lock->l_lvb_data, lock->l_lvb_len);
5737 if (lock->l_lvb_data != NULL)
5740 /* if layout lock was granted right away, the layout is returned
5741 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5742 * blocked and then granted via completion ast, we have to fetch
5743 * layout here. Please note that we can't use the LVB buffer in
5744 * completion AST because it doesn't have a large enough buffer */
5745 rc = ll_get_default_mdsize(sbi, &lmmsize);
5749 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5750 XATTR_NAME_LOV, lmmsize, &req);
5753 GOTO(out, rc = 0); /* empty layout */
5760 if (lmmsize == 0) /* empty layout */
5763 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5765 GOTO(out, rc = -EFAULT);
5767 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5768 if (lvbdata == NULL)
5769 GOTO(out, rc = -ENOMEM);
5771 memcpy(lvbdata, lmm, lmmsize);
5772 lock_res_and_lock(lock);
5773 if (unlikely(lock->l_lvb_data == NULL)) {
5774 lock->l_lvb_type = LVB_T_LAYOUT;
5775 lock->l_lvb_data = lvbdata;
5776 lock->l_lvb_len = lmmsize;
5779 unlock_res_and_lock(lock);
5782 OBD_FREE_LARGE(lvbdata, lmmsize);
5787 ptlrpc_req_finished(req);
5792 * Apply the layout to the inode. Layout lock is held and will be released
5795 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5796 struct inode *inode)
5798 struct ll_inode_info *lli = ll_i2info(inode);
5799 struct ll_sb_info *sbi = ll_i2sbi(inode);
5800 struct ldlm_lock *lock;
5801 struct cl_object_conf conf;
5804 bool wait_layout = false;
5807 LASSERT(lustre_handle_is_used(lockh));
5809 lock = ldlm_handle2lock(lockh);
5810 LASSERT(lock != NULL);
5812 if (!ldlm_has_layout(lock))
5813 GOTO(out, rc = -EAGAIN);
5815 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5816 PFID(&lli->lli_fid), inode);
5818 /* in case this is a caching lock and reinstate with new inode */
5819 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5821 lock_res_and_lock(lock);
5822 lvb_ready = ldlm_is_lvb_ready(lock);
5823 unlock_res_and_lock(lock);
5825 /* checking lvb_ready is racy but this is okay. The worst case is
5826 * that multi processes may configure the file on the same time. */
5830 rc = ll_layout_fetch(inode, lock);
5834 /* for layout lock, lmm is stored in lock's lvb.
5835 * lvb_data is immutable if the lock is held so it's safe to access it
5838 * set layout to file. Unlikely this will fail as old layout was
5839 * surely eliminated */
5840 memset(&conf, 0, sizeof conf);
5841 conf.coc_opc = OBJECT_CONF_SET;
5842 conf.coc_inode = inode;
5843 conf.coc_lock = lock;
5844 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5845 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5846 rc = ll_layout_conf(inode, &conf);
5848 /* refresh layout failed, need to wait */
5849 wait_layout = rc == -EBUSY;
5852 LDLM_LOCK_PUT(lock);
5853 ldlm_lock_decref(lockh, mode);
5855 /* wait for IO to complete if it's still being used. */
5857 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5858 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5860 memset(&conf, 0, sizeof conf);
5861 conf.coc_opc = OBJECT_CONF_WAIT;
5862 conf.coc_inode = inode;
5863 rc = ll_layout_conf(inode, &conf);
5867 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5868 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5874 * Issue layout intent RPC to MDS.
5875 * \param inode [in] file inode
5876 * \param intent [in] layout intent
5878 * \retval 0 on success
5879 * \retval < 0 error code
5881 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5883 struct ll_inode_info *lli = ll_i2info(inode);
5884 struct ll_sb_info *sbi = ll_i2sbi(inode);
5885 struct md_op_data *op_data;
5886 struct lookup_intent it;
5887 struct ptlrpc_request *req;
5891 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5892 0, 0, LUSTRE_OPC_ANY, NULL);
5893 if (IS_ERR(op_data))
5894 RETURN(PTR_ERR(op_data));
5896 op_data->op_data = intent;
5897 op_data->op_data_size = sizeof(*intent);
5899 memset(&it, 0, sizeof(it));
5900 it.it_op = IT_LAYOUT;
5901 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5902 intent->li_opc == LAYOUT_INTENT_TRUNC)
5903 it.it_flags = FMODE_WRITE;
5905 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5906 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5908 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5909 &ll_md_blocking_ast, 0);
5910 if (it.it_request != NULL)
5911 ptlrpc_req_finished(it.it_request);
5912 it.it_request = NULL;
5914 ll_finish_md_op_data(op_data);
5916 /* set lock data in case this is a new lock */
5918 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5920 ll_intent_drop_lock(&it);
5926 * This function checks if there exists a LAYOUT lock on the client side,
5927 * or enqueues it if it doesn't have one in cache.
5929 * This function will not hold layout lock so it may be revoked any time after
5930 * this function returns. Any operations depend on layout should be redone
5933 * This function should be called before lov_io_init() to get an uptodate
5934 * layout version, the caller should save the version number and after IO
5935 * is finished, this function should be called again to verify that layout
5936 * is not changed during IO time.
5938 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5940 struct ll_inode_info *lli = ll_i2info(inode);
5941 struct ll_sb_info *sbi = ll_i2sbi(inode);
5942 struct lustre_handle lockh;
5943 struct layout_intent intent = {
5944 .li_opc = LAYOUT_INTENT_ACCESS,
5946 enum ldlm_mode mode;
5950 *gen = ll_layout_version_get(lli);
5951 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
5955 LASSERT(fid_is_sane(ll_inode2fid(inode)));
5956 LASSERT(S_ISREG(inode->i_mode));
5958 /* take layout lock mutex to enqueue layout lock exclusively. */
5959 mutex_lock(&lli->lli_layout_mutex);
5962 /* mostly layout lock is caching on the local side, so try to
5963 * match it before grabbing layout lock mutex. */
5964 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
5965 LCK_CR | LCK_CW | LCK_PR |
5967 if (mode != 0) { /* hit cached lock */
5968 rc = ll_layout_lock_set(&lockh, mode, inode);
5974 rc = ll_layout_intent(inode, &intent);
5980 *gen = ll_layout_version_get(lli);
5981 mutex_unlock(&lli->lli_layout_mutex);
5987 * Issue layout intent RPC indicating where in a file an IO is about to write.
5989 * \param[in] inode file inode.
5990 * \param[in] ext write range with start offset of fille in bytes where
5991 * an IO is about to write, and exclusive end offset in
5994 * \retval 0 on success
5995 * \retval < 0 error code
5997 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
5998 struct lu_extent *ext)
6000 struct layout_intent intent = {
6002 .li_extent.e_start = ext->e_start,
6003 .li_extent.e_end = ext->e_end,
6008 rc = ll_layout_intent(inode, &intent);
6014 * This function send a restore request to the MDT
6016 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6018 struct hsm_user_request *hur;
6022 len = sizeof(struct hsm_user_request) +
6023 sizeof(struct hsm_user_item);
6024 OBD_ALLOC(hur, len);
6028 hur->hur_request.hr_action = HUA_RESTORE;
6029 hur->hur_request.hr_archive_id = 0;
6030 hur->hur_request.hr_flags = 0;
6031 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6032 sizeof(hur->hur_user_item[0].hui_fid));
6033 hur->hur_user_item[0].hui_extent.offset = offset;
6034 hur->hur_user_item[0].hui_extent.length = length;
6035 hur->hur_request.hr_itemcount = 1;
6036 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,