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) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #ifndef LLITE_INTERNAL_H
34 #define LLITE_INTERNAL_H
35 #include <lustre_debug.h>
36 #include <lustre_disk.h> /* for s2sbi */
37 #include <lustre_eacl.h>
38 #include <lustre_linkea.h>
40 /* for struct cl_lock_descr and struct cl_io */
41 #include <cl_object.h>
42 #include <lustre_lmv.h>
43 #include <lustre_mdc.h>
44 #include <lustre_intent.h>
45 #include <linux/compat.h>
46 #include <linux/aio.h>
47 #include <lustre_compat.h>
48 #include <lustre_crypto.h>
50 #include "vvp_internal.h"
51 #include "range_lock.h"
58 #ifndef HAVE_VM_FAULT_RETRY
59 #define VM_FAULT_RETRY 0
62 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
63 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
64 #ifndef LOOKUP_CONTINUE
65 #define LOOKUP_CONTINUE LOOKUP_PARENT
68 /** Only used on client-side for indicating the tail of dir hash/offset. */
69 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
70 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
72 /* 4UL * 1024 * 1024 */
73 #define LL_MAX_BLKSIZE_BITS 22
75 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
77 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
79 struct ll_dentry_data {
80 unsigned int lld_sa_generation;
81 unsigned int lld_invalid:1;
82 unsigned int lld_nfs_dentry:1;
83 struct rcu_head lld_rcu_head;
86 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
88 #define LLI_INODE_MAGIC 0x111d0de5
89 #define LLI_INODE_DEAD 0xdeadd00d
91 struct ll_getname_data {
92 #ifdef HAVE_DIR_CONTEXT
93 struct dir_context ctx;
95 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
96 struct lu_fid lgd_fid; /* target fid we are looking for */
97 int lgd_found; /* inode matched? */
100 struct ll_grouplock {
101 struct lu_env *lg_env;
103 struct cl_lock *lg_lock;
104 unsigned long lg_gid;
107 /* See comment on trunc_sem_down_read_nowait */
108 struct ll_trunc_sem {
109 /* when positive, this is a count of readers, when -1, it indicates
110 * the semaphore is held for write, and 0 is unlocked
112 atomic_t ll_trunc_readers;
113 /* this tracks a count of waiting writers */
114 atomic_t ll_trunc_waiters;
117 struct ll_inode_info {
118 __u32 lli_inode_magic;
121 volatile unsigned long lli_flags;
122 struct posix_acl *lli_posix_acl;
124 /* identifying fields for both metadata and data stacks. */
125 struct lu_fid lli_fid;
126 /* master inode fid for stripe directory */
127 struct lu_fid lli_pfid;
129 /* We need all three because every inode may be opened in different
131 struct obd_client_handle *lli_mds_read_och;
132 struct obd_client_handle *lli_mds_write_och;
133 struct obd_client_handle *lli_mds_exec_och;
134 __u64 lli_open_fd_read_count;
135 __u64 lli_open_fd_write_count;
136 __u64 lli_open_fd_exec_count;
137 /* Protects access to och pointers and their usage counters */
138 struct mutex lli_och_mutex;
140 struct inode lli_vfs_inode;
142 /* the most recent timestamps obtained from mds */
147 spinlock_t lli_agl_lock;
149 /* Try to make the d::member and f::member are aligned. Before using
150 * these members, make clear whether it is directory or not. */
154 /* metadata statahead */
155 /* since parent-child threads can share the same @file
156 * struct, "opendir_key" is the token when dir close for
157 * case of parent exit before child -- it is me should
158 * cleanup the dir readahead. */
159 void *lli_opendir_key;
160 struct ll_statahead_info *lli_sai;
161 /* protect statahead stuff. */
162 spinlock_t lli_sa_lock;
163 /* "opendir_pid" is the token when lookup/revalid
164 * -- I am the owner of dir statahead. */
165 pid_t lli_opendir_pid;
166 /* stat will try to access statahead entries or start
167 * statahead if this flag is set, and this flag will be
168 * set upon dir open, and cleared when dir is closed,
169 * statahead hit ratio is too low, or start statahead
171 unsigned int lli_sa_enabled:1;
172 /* generation for statahead */
173 unsigned int lli_sa_generation;
174 /* rw lock protects lli_lsm_md */
175 struct rw_semaphore lli_lsm_sem;
176 /* directory stripe information */
177 struct lmv_stripe_md *lli_lsm_md;
178 /* directory default LMV */
179 struct lmv_stripe_md *lli_default_lsm_md;
182 /* for non-directory */
184 struct mutex lli_size_mutex;
185 char *lli_symlink_name;
186 struct ll_trunc_sem lli_trunc_sem;
187 struct range_lock_tree lli_write_tree;
188 struct mutex lli_setattr_mutex;
190 struct rw_semaphore lli_glimpse_sem;
191 ktime_t lli_glimpse_time;
192 struct list_head lli_agl_list;
195 /* for writepage() only to communicate to fsync */
198 /* protect the file heat fields */
199 spinlock_t lli_heat_lock;
200 __u32 lli_heat_flags;
201 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
204 * Whenever a process try to read/write the file, the
205 * jobid of the process will be saved here, and it'll
206 * be packed into the write PRC when flush later.
208 * So the read/write statistics for jobid will not be
209 * accurate if the file is shared by different jobs.
211 char lli_jobid[LUSTRE_JOBID_SIZE];
213 struct mutex lli_pcc_lock;
214 enum lu_pcc_state_flags lli_pcc_state;
216 * @lli_pcc_generation saves the gobal PCC generation
217 * when the file was successfully attached into PCC.
218 * The flags of the PCC dataset are saved in
220 * The gobal PCC generation will be increased when add
221 * or delete a PCC backend, or change the configuration
222 * parameters for PCC.
223 * If @lli_pcc_generation is same as the gobal PCC
224 * generation, we can use the saved flags of the PCC
225 * dataset to determine whether need to try auto attach
228 __u64 lli_pcc_generation;
229 enum pcc_dataset_flags lli_pcc_dsflags;
230 struct pcc_inode *lli_pcc_inode;
232 struct mutex lli_group_mutex;
233 __u64 lli_group_users;
234 unsigned long lli_group_gid;
236 __u64 lli_attr_valid;
238 __u64 lli_lazyblocks;
242 /* XXX: For following frequent used members, although they maybe special
243 * used for non-directory object, it is some time-wasting to check
244 * whether the object is directory or not before using them. On the
245 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
246 * the "ll_inode_info" size even if moving those members into u.f.
247 * So keep them out side.
249 * In the future, if more members are added only for directory,
250 * some of the following members can be moved into u.f.
252 struct cl_object *lli_clob;
254 /* mutex to request for layout lock exclusively. */
255 struct mutex lli_layout_mutex;
256 /* Layout version, protected by lli_layout_lock */
257 __u32 lli_layout_gen;
258 spinlock_t lli_layout_lock;
260 __u32 lli_projid; /* project id */
262 struct rw_semaphore lli_xattrs_list_rwsem;
263 struct mutex lli_xattrs_enq_lock;
264 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
267 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
269 atomic_set(&sem->ll_trunc_readers, 0);
270 atomic_set(&sem->ll_trunc_waiters, 0);
273 /* This version of down read ignores waiting writers, meaning if the semaphore
274 * is already held for read, this down_read will 'join' that reader and also
275 * take the semaphore.
277 * This lets us avoid an unusual deadlock.
279 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
280 * in Lustre, in order to exclude truncates. Some of these paths then need to
281 * take the mmap_sem, while still holding the trunc_sem. The problem is that
282 * page faults hold the mmap_sem when calling in to Lustre, and then must also
283 * take the trunc_sem to exclude truncate.
285 * This means the locking order for trunc_sem and mmap_sem is sometimes AB,
286 * sometimes BA. This is almost OK because in both cases, we take the trunc
287 * sem for read, so it doesn't block.
289 * However, if a write mode user (truncate, a setattr op) arrives in the
290 * middle of this, the second reader on the truncate_sem will wait behind that
293 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
294 * to the mode in which they take the semaphore):
295 * reader (holding mmap_sem, needs truncate_sem)
297 * reader (holding truncate sem, waiting for mmap_sem)
299 * And so the readers deadlock.
301 * The solution is this modified semaphore, where this down_read ignores
302 * waiting write operations, and all waiters are woken up at once, so readers
303 * using down_read_nowait cannot get stuck behind waiting writers, regardless
304 * of the order they arrived in.
306 * down_read_nowait is only used in the page fault case, where we already hold
307 * the mmap_sem. This is because otherwise repeated read and write operations
308 * (which take the truncate sem) could prevent a truncate from ever starting.
309 * This could still happen with page faults, but without an even more complex
310 * mechanism, this is unavoidable.
314 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
316 wait_var_event(&sem->ll_trunc_readers,
317 atomic_inc_unless_negative(&sem->ll_trunc_readers));
320 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
322 wait_var_event(&sem->ll_trunc_readers,
323 atomic_read(&sem->ll_trunc_waiters) == 0 &&
324 atomic_inc_unless_negative(&sem->ll_trunc_readers));
327 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
329 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
330 atomic_read(&sem->ll_trunc_waiters))
331 wake_up_var(&sem->ll_trunc_readers);
334 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
336 atomic_inc(&sem->ll_trunc_waiters);
337 wait_var_event(&sem->ll_trunc_readers,
338 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
339 atomic_dec(&sem->ll_trunc_waiters);
342 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
344 atomic_set(&sem->ll_trunc_readers, 0);
345 wake_up_var(&sem->ll_trunc_readers);
348 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
349 static inline void lli_clear_acl(struct ll_inode_info *lli)
351 if (lli->lli_posix_acl) {
352 posix_acl_release(lli->lli_posix_acl);
353 lli->lli_posix_acl = NULL;
357 static inline void lli_replace_acl(struct ll_inode_info *lli,
358 struct lustre_md *md)
360 spin_lock(&lli->lli_lock);
361 if (lli->lli_posix_acl)
362 posix_acl_release(lli->lli_posix_acl);
363 lli->lli_posix_acl = md->posix_acl;
364 spin_unlock(&lli->lli_lock);
367 static inline void lli_clear_acl(struct ll_inode_info *lli)
371 static inline void lli_replace_acl(struct ll_inode_info *lli,
372 struct lustre_md *md)
377 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
381 spin_lock(&lli->lli_layout_lock);
382 gen = lli->lli_layout_gen;
383 spin_unlock(&lli->lli_layout_lock);
388 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
390 spin_lock(&lli->lli_layout_lock);
391 lli->lli_layout_gen = gen;
392 spin_unlock(&lli->lli_layout_lock);
396 /* File data is modified. */
397 LLIF_DATA_MODIFIED = 0,
398 /* File is being restored */
399 LLIF_FILE_RESTORING = 1,
400 /* Xattr cache is attached to the file */
401 LLIF_XATTR_CACHE = 2,
402 /* Project inherit */
403 LLIF_PROJECT_INHERIT = 3,
404 /* update atime from MDS even if it's older than local inode atime. */
405 LLIF_UPDATE_ATIME = 4,
409 static inline void ll_file_set_flag(struct ll_inode_info *lli,
410 enum ll_file_flags flag)
412 set_bit(flag, &lli->lli_flags);
415 static inline void ll_file_clear_flag(struct ll_inode_info *lli,
416 enum ll_file_flags flag)
418 clear_bit(flag, &lli->lli_flags);
421 static inline bool ll_file_test_flag(struct ll_inode_info *lli,
422 enum ll_file_flags flag)
424 return test_bit(flag, &lli->lli_flags);
427 static inline bool ll_file_test_and_clear_flag(struct ll_inode_info *lli,
428 enum ll_file_flags flag)
430 return test_and_clear_bit(flag, &lli->lli_flags);
433 int ll_xattr_cache_destroy(struct inode *inode);
435 int ll_xattr_cache_get(struct inode *inode,
441 int ll_xattr_cache_insert(struct inode *inode,
446 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
448 #ifdef CONFIG_SECURITY
449 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
450 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
456 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
458 #ifdef CONFIG_SECURITY
459 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
463 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
464 const char **secctx_name, void **secctx,
466 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
469 int ll_listsecurity(struct inode *inode, char *secctx_name,
470 size_t secctx_name_size);
472 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
474 #ifdef HAVE_LUSTRE_CRYPTO
475 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
476 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
482 static inline void obd_connect_set_enc(struct obd_connect_data *data)
484 #ifdef HAVE_LUSTRE_CRYPTO
485 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
490 * Locking to guarantee consistency of non-atomic updates to long long i_size,
491 * consistency between file size and KMS.
493 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
496 void ll_inode_size_lock(struct inode *inode);
497 void ll_inode_size_unlock(struct inode *inode);
499 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
501 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
504 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
506 return ll_i2info(inode)->lli_pcc_inode;
509 /* default to use at least 16M for fast read if possible */
510 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
512 /* default read-ahead on a given client mountpoint. */
513 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(1024UL)
515 /* default read-ahead for a single file descriptor */
516 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX MiB_TO_PAGES(256UL)
518 /* default read-ahead full files smaller than limit on the second read */
519 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
521 /* default range pages */
522 #define SBI_DEFAULT_RA_RANGE_PAGES MiB_TO_PAGES(1ULL)
524 /* Min range pages */
525 #define RA_MIN_MMAP_RANGE_PAGES 16UL
530 RA_STAT_DISTANT_READPAGE,
531 RA_STAT_MISS_IN_WINDOW,
532 RA_STAT_FAILED_GRAB_PAGE,
533 RA_STAT_FAILED_MATCH,
538 RA_STAT_MAX_IN_FLIGHT,
539 RA_STAT_WRONG_GRAB_PAGE,
540 RA_STAT_FAILED_REACH_END,
542 RA_STAT_FAILED_FAST_READ,
543 RA_STAT_MMAP_RANGE_READ,
548 atomic_t ra_cur_pages;
549 unsigned long ra_max_pages;
550 unsigned long ra_max_pages_per_file;
551 unsigned long ra_range_pages;
552 unsigned long ra_max_read_ahead_whole_pages;
553 struct workqueue_struct *ll_readahead_wq;
555 * Max number of active works could be triggered
556 * for async readahead.
558 unsigned int ra_async_max_active;
559 /* how many async readahead triggered in flight */
560 atomic_t ra_async_inflight;
561 /* Threshold to control when to trigger async readahead */
562 unsigned long ra_async_pages_per_file_threshold;
565 /* ra_io_arg will be filled in the beginning of ll_readahead with
566 * ras_lock, then the following ll_read_ahead_pages will read RA
567 * pages according to this arg, all the items in this structure are
568 * counted by page index.
571 pgoff_t ria_start_idx; /* start offset of read-ahead*/
572 pgoff_t ria_end_idx; /* end offset of read-ahead*/
573 unsigned long ria_reserved; /* reserved pages for read-ahead */
574 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
575 bool ria_eof; /* reach end of file */
576 /* If stride read pattern is detected, ria_stoff is the byte offset
577 * where stride read is started. Note: for normal read-ahead, the
578 * value here is meaningless, and also it will not be accessed*/
580 /* ria_length and ria_bytes are the length and pages length in the
581 * stride I/O mode. And they will also be used to check whether
582 * it is stride I/O read-ahead in the read-ahead pages*/
587 /* LL_HIST_MAX=32 causes an overflow */
588 #define LL_HIST_MAX 28
589 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
590 #define LL_PROCESS_HIST_MAX 10
591 struct per_process_info {
593 struct obd_histogram pp_r_hist;
594 struct obd_histogram pp_w_hist;
597 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
598 struct ll_rw_extents_info {
599 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
602 #define LL_OFFSET_HIST_MAX 100
603 struct ll_rw_process_info {
606 loff_t rw_range_start;
608 loff_t rw_last_file_pos;
610 size_t rw_smallest_extent;
611 size_t rw_largest_extent;
612 struct ll_file_data *rw_last_file;
615 enum stats_track_type {
616 STATS_TRACK_ALL = 0, /* track all processes */
617 STATS_TRACK_PID, /* track process with this pid */
618 STATS_TRACK_PPID, /* track processes with this ppid */
619 STATS_TRACK_GID, /* track processes with this gid */
623 /* flags for sbi->ll_flags */
624 #define LL_SBI_NOLCK 0x01 /* DLM locking disabled (directio-only) */
625 #define LL_SBI_CHECKSUM 0x02 /* checksum each page as it's written */
626 #define LL_SBI_FLOCK 0x04
627 #define LL_SBI_USER_XATTR 0x08 /* support user xattr */
628 #define LL_SBI_ACL 0x10 /* support ACL */
629 /* LL_SBI_RMT_CLIENT 0x40 remote client */
630 #define LL_SBI_MDS_CAPA 0x80 /* support mds capa, obsolete */
631 #define LL_SBI_OSS_CAPA 0x100 /* support oss capa, obsolete */
632 #define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
633 #define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
634 #define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
635 /* LL_SBI_SOM_PREVIEW 0x1000 SOM preview mount option, obsolete */
636 #define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
637 #define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
638 #define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
639 #define LL_SBI_VERBOSE 0x10000 /* verbose mount/umount */
640 #define LL_SBI_LAYOUT_LOCK 0x20000 /* layout lock support */
641 #define LL_SBI_USER_FID2PATH 0x40000 /* allow fid2path by unprivileged users */
642 #define LL_SBI_XATTR_CACHE 0x80000 /* support for xattr cache */
643 #define LL_SBI_NOROOTSQUASH 0x100000 /* do not apply root squash */
644 #define LL_SBI_ALWAYS_PING 0x200000 /* always ping even if server
646 #define LL_SBI_FAST_READ 0x400000 /* fast read support */
647 #define LL_SBI_FILE_SECCTX 0x800000 /* set file security context at create */
648 /* LL_SBI_PIO 0x1000000 parallel IO support, introduced in
650 #define LL_SBI_TINY_WRITE 0x2000000 /* tiny write support */
651 #define LL_SBI_FILE_HEAT 0x4000000 /* file heat support */
652 #define LL_SBI_TEST_DUMMY_ENCRYPTION 0x8000000 /* test dummy encryption */
653 #define LL_SBI_ENCRYPT 0x10000000 /* client side encryption */
654 #define LL_SBI_FLAGS { \
682 "test_dummy_encryption", \
686 /* This is embedded into llite super-blocks to keep track of connect
687 * flags (capabilities) supported by all imports given mount is
689 struct lustre_client_ocd {
690 /* This is conjunction of connect_flags across all imports
691 * (LOVs) this mount is connected to. This field is updated by
692 * cl_ocd_update() under ->lco_lock. */
694 struct mutex lco_lock;
695 struct obd_export *lco_md_exp;
696 struct obd_export *lco_dt_exp;
700 /* this protects pglist and ra_info. It isn't safe to
701 * grab from interrupt contexts */
703 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
704 spinlock_t ll_process_lock; /* ll_rw_process_info */
705 struct obd_uuid ll_sb_uuid;
706 struct obd_export *ll_md_exp;
707 struct obd_export *ll_dt_exp;
708 struct obd_device *ll_md_obd;
709 struct obd_device *ll_dt_obd;
710 struct dentry *ll_debugfs_entry;
711 struct lu_fid ll_root_fid; /* root object fid */
714 unsigned int ll_xattr_cache_enabled:1,
715 ll_xattr_cache_set:1, /* already set to 0/1 */
716 ll_client_common_fill_super_succeeded:1,
719 struct lustre_client_ocd ll_lco;
721 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
723 /* Used to track "unstable" pages on a client, and maintain a
724 * LRU list of clean pages. An "unstable" page is defined as
725 * any page which is sent to a server as part of a bulk request,
726 * but is uncommitted to stable storage. */
727 struct cl_client_cache *ll_cache;
729 struct lprocfs_stats *ll_ra_stats;
731 struct ll_ra_info ll_ra_info;
732 unsigned int ll_namelen;
733 struct file_operations *ll_fop;
735 struct lu_site *ll_site;
736 struct cl_device *ll_cl;
738 struct ll_rw_extents_info ll_rw_extents_info;
739 int ll_extent_process_count;
740 struct ll_rw_process_info ll_rw_process_info[LL_PROCESS_HIST_MAX];
741 unsigned int ll_offset_process_count;
742 struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
743 unsigned int ll_rw_offset_entry_count;
744 int ll_stats_track_id;
745 enum stats_track_type ll_stats_track_type;
748 /* metadata stat-ahead */
749 unsigned int ll_sa_running_max;/* max concurrent
750 * statahead instances */
751 unsigned int ll_sa_max; /* max statahead RPCs */
752 atomic_t ll_sa_total; /* statahead thread started
754 atomic_t ll_sa_wrong; /* statahead thread stopped for
756 atomic_t ll_sa_running; /* running statahead thread
758 atomic_t ll_agl_total; /* AGL thread started count */
760 dev_t ll_sdev_orig; /* save s_dev before assign for
763 struct root_squash_info ll_squash;
766 /* st_blksize returned by stat(2), when non-zero */
767 unsigned int ll_stat_blksize;
769 /* maximum relative age of cached statfs results */
770 unsigned int ll_statfs_max_age;
772 struct kset ll_kset; /* sysfs object */
773 struct completion ll_kobj_unregister;
776 unsigned int ll_heat_decay_weight;
777 unsigned int ll_heat_period_second;
779 /* filesystem fsname */
780 char ll_fsname[LUSTRE_MAXFSNAME + 1];
782 /* Persistent Client Cache */
783 struct pcc_super ll_pcc_super;
786 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
787 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
789 * per file-descriptor read-ahead data.
791 struct ll_readahead_state {
793 /* End byte that read(2) try to read. */
794 loff_t ras_last_read_end_bytes;
796 * number of bytes read after last read-ahead window reset. As window
797 * is reset on each seek, this is effectively a number of consecutive
798 * accesses. Maybe ->ras_accessed_in_window is better name.
800 * XXX nikita: window is also reset (by ras_update()) when Lustre
801 * believes that memory pressure evicts read-ahead pages. In that
802 * case, it probably doesn't make sense to expand window to
803 * PTLRPC_MAX_BRW_PAGES on the third access.
805 loff_t ras_consecutive_bytes;
807 * number of read requests after the last read-ahead window reset
808 * As window is reset on each seek, this is effectively the number
809 * on consecutive read request and is used to trigger read-ahead.
811 unsigned long ras_consecutive_requests;
813 * Parameters of current read-ahead window. Handled by
814 * ras_update(). On the initial access to the file or after a seek,
815 * window is reset to 0. After 3 consecutive accesses, window is
816 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
817 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
819 pgoff_t ras_window_start_idx;
820 pgoff_t ras_window_pages;
822 /* Page index where min range read starts */
823 pgoff_t ras_range_min_start_idx;
824 /* Page index where mmap range read ends */
825 pgoff_t ras_range_max_end_idx;
826 /* number of mmap pages where last time detected */
827 pgoff_t ras_last_range_pages;
828 /* number of mmap range requests */
829 pgoff_t ras_range_requests;
832 * Optimal RPC size in pages.
833 * It decides how many pages will be sent for each read-ahead.
835 unsigned long ras_rpc_pages;
837 * Where next read-ahead should start at. This lies within read-ahead
838 * window. Read-ahead window is read in pieces rather than at once
839 * because: 1. lustre limits total number of pages under read-ahead by
840 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
841 * not covered by DLM lock.
843 pgoff_t ras_next_readahead_idx;
845 * Total number of ll_file_read requests issued, reads originating
846 * due to mmap are not counted in this total. This value is used to
847 * trigger full file read-ahead after multiple reads to a small file.
849 unsigned long ras_requests;
851 * The following 3 items are used for detecting the stride I/O
853 * In stride I/O mode,
854 * ...............|-----data-----|****gap*****|--------|******|....
855 * offset |-stride_bytes-|-stride_gap-|
856 * ras_stride_offset = offset;
857 * ras_stride_length = stride_bytes + stride_gap;
858 * ras_stride_bytes = stride_bytes;
859 * Note: all these three items are counted by bytes.
861 loff_t ras_stride_offset;
862 loff_t ras_stride_length;
863 loff_t ras_stride_bytes;
865 * number of consecutive stride request count, and it is similar as
866 * ras_consecutive_requests, but used for stride I/O mode.
867 * Note: only more than 2 consecutive stride request are detected,
868 * stride read-ahead will be enable
870 unsigned long ras_consecutive_stride_requests;
871 /* index of the last page that async readahead starts */
872 pgoff_t ras_async_last_readpage_idx;
873 /* whether we should increase readahead window */
874 bool ras_need_increase_window;
875 /* whether ra miss check should be skipped */
876 bool ras_no_miss_check;
879 struct ll_readahead_work {
880 /** File to readahead */
881 struct file *lrw_file;
882 pgoff_t lrw_start_idx;
885 /* async worker to handler read */
886 struct work_struct lrw_readahead_work;
887 char lrw_jobid[LUSTRE_JOBID_SIZE];
890 extern struct kmem_cache *ll_file_data_slab;
891 struct lustre_handle;
892 struct ll_file_data {
893 struct ll_readahead_state fd_ras;
894 struct ll_grouplock fd_grouplock;
898 /* openhandle if lease exists for this file.
899 * Borrow lli->lli_och_mutex to protect assignment */
900 struct obd_client_handle *fd_lease_och;
901 struct obd_client_handle *fd_och;
902 struct file *fd_file;
903 /* Indicate whether need to report failure when close.
904 * true: failure is known, not report again.
905 * false: unknown failure, should report. */
906 bool fd_write_failed;
907 bool ll_lock_no_expand;
908 rwlock_t fd_lock; /* protect lcc list */
909 struct list_head fd_lccs; /* list of ll_cl_context */
910 /* Used by mirrored file to lead IOs to a specific mirror, usually
911 * for mirror resync. 0 means default. */
912 __u32 fd_designated_mirror;
913 /* The layout version when resync starts. Resync I/O should carry this
914 * layout version for verification to OST objects */
915 __u32 fd_layout_version;
916 struct pcc_file fd_pcc_file;
919 void llite_tunables_unregister(void);
920 int llite_tunables_register(void);
922 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
924 return &lli->lli_vfs_inode;
927 __u32 ll_i2suppgid(struct inode *i);
928 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
930 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
932 #if BITS_PER_LONG == 32
934 #elif defined(CONFIG_COMPAT)
935 if (unlikely(sbi->ll_flags & LL_SBI_32BIT_API))
938 # ifdef CONFIG_X86_X32
939 /* in_compat_syscall() returns true when called from a kthread
940 * and CONFIG_X86_X32 is enabled, which is wrong. So check
941 * whether the caller comes from a syscall (ie. not a kthread)
942 * before calling in_compat_syscall(). */
943 if (current->flags & PF_KTHREAD)
947 return unlikely(in_compat_syscall());
949 return unlikely(sbi->ll_flags & LL_SBI_32BIT_API);
953 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
955 return !!(sbi->ll_flags & LL_SBI_FAST_READ);
958 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
960 return !!(sbi->ll_flags & LL_SBI_TINY_WRITE);
963 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
965 return !!(sbi->ll_flags & LL_SBI_FILE_HEAT);
968 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
970 /* llite/lcommon_misc.c */
971 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
972 enum obd_notify_event ev, void *owner);
973 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
974 struct ll_grouplock *lg);
975 void cl_put_grouplock(struct ll_grouplock *lg);
977 /* llite/lproc_llite.c */
978 int ll_debugfs_register_super(struct super_block *sb, const char *name);
979 void ll_debugfs_unregister_super(struct super_block *sb);
980 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
984 LPROC_LL_WRITE_BYTES,
1011 LPROC_LL_GETXATTR_HITS,
1013 LPROC_LL_REMOVEXATTR,
1014 LPROC_LL_INODE_PERM,
1016 LPROC_LL_FILE_OPCODES
1020 enum get_default_layout_type {
1021 GET_DEFAULT_LAYOUT_ROOT = 1,
1024 struct ll_dir_chain {
1027 static inline void ll_dir_chain_init(struct ll_dir_chain *chain)
1031 static inline void ll_dir_chain_fini(struct ll_dir_chain *chain)
1035 extern const struct file_operations ll_dir_operations;
1036 extern const struct inode_operations ll_dir_inode_operations;
1037 #ifdef HAVE_DIR_CONTEXT
1038 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1039 struct dir_context *ctx);
1041 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1042 void *cookie, filldir_t filldir);
1044 int ll_get_mdt_idx(struct inode *inode);
1045 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1046 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1047 __u64 offset, struct ll_dir_chain *chain);
1048 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1049 int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl);
1052 extern const struct inode_operations ll_special_inode_operations;
1054 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1055 struct lustre_md *lic);
1056 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1057 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1058 void *data, int flag);
1059 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1060 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1061 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1064 int ll_writepage(struct page *page, struct writeback_control *wbc);
1065 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1066 int ll_readpage(struct file *file, struct page *page);
1067 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1068 struct cl_page *page, struct file *file);
1069 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1070 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1073 void ll_cl_add(struct file *file, const struct lu_env *env, struct cl_io *io,
1074 enum lcc_type type);
1075 void ll_cl_remove(struct file *file, const struct lu_env *env);
1076 struct ll_cl_context *ll_cl_find(struct file *file);
1078 extern const struct address_space_operations ll_aops;
1081 extern struct file_operations ll_file_operations;
1082 extern struct file_operations ll_file_operations_flock;
1083 extern struct file_operations ll_file_operations_noflock;
1084 extern struct inode_operations ll_file_inode_operations;
1085 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1086 enum ldlm_mode l_req_mode);
1087 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1088 struct lustre_handle *lockh, __u64 flags,
1089 enum ldlm_mode mode);
1091 int ll_file_open(struct inode *inode, struct file *file);
1092 int ll_file_release(struct inode *inode, struct file *file);
1093 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1094 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1095 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1096 struct ll_file_data *file, loff_t pos,
1097 size_t count, int rw);
1098 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1099 int ll_getattr(const struct path *path, struct kstat *stat,
1100 u32 request_mask, unsigned int flags);
1102 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1104 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1105 unsigned int flags);
1106 struct posix_acl *ll_get_acl(struct inode *inode, int type);
1107 #ifdef HAVE_IOP_SET_ACL
1108 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1109 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type);
1110 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1111 #define ll_set_acl NULL
1112 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1116 static inline int ll_xflags_to_inode_flags(int xflags)
1118 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1119 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1120 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1121 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1124 static inline int ll_inode_flags_to_xflags(int flags)
1126 return ((flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1127 ((flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1128 ((flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1129 ((flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1132 int ll_migrate(struct inode *parent, struct file *file,
1133 struct lmv_user_md *lum, const char *name);
1134 int ll_get_fid_by_name(struct inode *parent, const char *name,
1135 int namelen, struct lu_fid *fid, struct inode **inode);
1136 int ll_inode_permission(struct inode *inode, int mask);
1137 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa);
1138 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1140 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1143 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1144 __u64 flags, struct lov_user_md *lum,
1146 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1147 struct lov_mds_md **lmm, int *lmm_size,
1148 struct ptlrpc_request **request);
1149 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1151 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1152 int *lmm_size, struct ptlrpc_request **request,
1153 struct ptlrpc_request **root_request, u64 valid);
1154 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1155 struct ptlrpc_request **request, u64 valid);
1156 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1157 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1158 int ll_fid2path(struct inode *inode, void __user *arg);
1159 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1160 int ll_hsm_release(struct inode *inode);
1161 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1162 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1164 /* llite/dcache.c */
1166 int ll_d_init(struct dentry *de);
1167 extern const struct dentry_operations ll_d_ops;
1168 void ll_intent_drop_lock(struct lookup_intent *);
1169 void ll_intent_release(struct lookup_intent *);
1170 void ll_prune_aliases(struct inode *inode);
1171 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1172 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1173 struct lookup_intent *it, struct dentry *de);
1175 /* llite/llite_lib.c */
1176 extern struct super_operations lustre_super_operations;
1178 void ll_lli_init(struct ll_inode_info *lli);
1179 int ll_fill_super(struct super_block *sb);
1180 void ll_put_super(struct super_block *sb);
1181 void ll_kill_super(struct super_block *sb);
1182 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1183 void ll_dir_clear_lsm_md(struct inode *inode);
1184 void ll_clear_inode(struct inode *inode);
1185 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1186 enum op_xvalid xvalid, bool hsm_import);
1187 int ll_setattr(struct dentry *de, struct iattr *attr);
1188 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1189 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1191 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1192 void ll_update_inode_flags(struct inode *inode, int ext_flags);
1193 int ll_read_inode2(struct inode *inode, void *opaque);
1194 void ll_delete_inode(struct inode *inode);
1195 int ll_iocontrol(struct inode *inode, struct file *file,
1196 unsigned int cmd, unsigned long arg);
1197 int ll_flush_ctx(struct inode *inode);
1198 void ll_umount_begin(struct super_block *sb);
1199 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1200 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1201 void ll_dirty_page_discard_warn(struct page *page, int ioret);
1202 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1203 struct super_block *, struct lookup_intent *);
1204 int ll_obd_statfs(struct inode *inode, void __user *arg);
1205 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1206 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1207 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1209 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1210 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1211 struct inode *i1, struct inode *i2,
1212 const char *name, size_t namelen,
1213 __u32 mode, enum md_op_code opc,
1215 void ll_finish_md_op_data(struct md_op_data *op_data);
1216 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1217 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1218 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1219 struct lov_user_md **kbuf);
1220 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req);
1222 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1224 /* Compute expected user md size when passing in a md from user space */
1225 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1227 switch (lum->lmm_magic) {
1228 case LOV_USER_MAGIC_V1:
1229 return sizeof(struct lov_user_md_v1);
1230 case LOV_USER_MAGIC_V3:
1231 return sizeof(struct lov_user_md_v3);
1232 case LOV_USER_MAGIC_SPECIFIC:
1233 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1236 return lov_user_md_size(lum->lmm_stripe_count,
1237 LOV_USER_MAGIC_SPECIFIC);
1238 case LOV_USER_MAGIC_COMP_V1:
1239 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1240 case LOV_USER_MAGIC_FOREIGN:
1241 return foreign_size(lum);
1247 /* llite/llite_nfs.c */
1248 extern struct export_operations lustre_export_operations;
1249 __u32 get_uuid2int(const char *name, int len);
1250 struct inode *search_inode_for_lustre(struct super_block *sb,
1251 const struct lu_fid *fid);
1252 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1254 /* llite/symlink.c */
1255 extern struct inode_operations ll_fast_symlink_inode_operations;
1258 * IO arguments for various VFS I/O interfaces.
1260 struct vvp_io_args {
1261 /** normal/sendfile/splice */
1264 struct kiocb *via_iocb;
1265 struct iov_iter *via_iter;
1275 struct ll_cl_context {
1276 struct list_head lcc_list;
1278 const struct lu_env *lcc_env;
1279 struct cl_io *lcc_io;
1280 struct cl_page *lcc_page;
1281 enum lcc_type lcc_type;
1284 struct ll_thread_info {
1285 struct vvp_io_args lti_args;
1286 struct ra_io_arg lti_ria;
1287 struct ll_cl_context lti_io_ctx;
1290 extern struct lu_context_key ll_thread_key;
1292 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1294 struct ll_thread_info *lti;
1296 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1297 LASSERT(lti != NULL);
1302 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1304 return &ll_env_info(env)->lti_args;
1307 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1308 struct vvp_io_args *args);
1310 /* llite/llite_mmap.c */
1312 int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
1313 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1314 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1315 unsigned long addr, size_t count);
1316 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1319 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1321 /* don't need an addref as the sb_info should be holding one */
1322 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1324 return ll_s2sbi(sb)->ll_dt_exp;
1327 /* don't need an addref as the sb_info should be holding one */
1328 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1330 return ll_s2sbi(sb)->ll_md_exp;
1333 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1335 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1341 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1342 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1344 return ll_s2sbi(inode->i_sb);
1347 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1349 return ll_s2dtexp(inode->i_sb);
1352 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1354 return ll_s2mdexp(inode->i_sb);
1357 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1361 LASSERT(inode != NULL);
1362 fid = &ll_i2info(inode)->lli_fid;
1367 static inline bool ll_dir_striped(struct inode *inode)
1370 return S_ISDIR(inode->i_mode) &&
1371 lmv_dir_striped(ll_i2info(inode)->lli_lsm_md);
1374 static inline loff_t ll_file_maxbytes(struct inode *inode)
1376 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1379 return MAX_LFS_FILESIZE;
1381 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1385 extern const struct xattr_handler *ll_xattr_handlers[];
1387 #define XATTR_USER_T 1
1388 #define XATTR_TRUSTED_T 2
1389 #define XATTR_SECURITY_T 3
1390 #define XATTR_ACL_ACCESS_T 4
1391 #define XATTR_ACL_DEFAULT_T 5
1392 #define XATTR_LUSTRE_T 6
1393 #define XATTR_OTHER_T 7
1395 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1396 int ll_xattr_list(struct inode *inode, const char *name, int type,
1397 void *buffer, size_t size, u64 valid);
1398 const struct xattr_handler *get_xattr_type(const char *name);
1401 * Common IO arguments for various VFS I/O interfaces.
1403 int cl_sb_init(struct super_block *sb);
1404 int cl_sb_fini(struct super_block *sb);
1406 enum ras_update_flags {
1410 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1411 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1415 #define LL_SA_RPC_MIN 2
1416 #define LL_SA_RPC_DEF 32
1417 #define LL_SA_RPC_MAX 512
1419 /* XXX: If want to support more concurrent statahead instances,
1420 * please consider to decentralize the RPC lists attached
1421 * on related import, such as imp_{sending,delayed}_list.
1423 #define LL_SA_RUNNING_MAX 256
1424 #define LL_SA_RUNNING_DEF 16
1426 #define LL_SA_CACHE_BIT 5
1427 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1428 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1430 /* per inode struct, for dir only */
1431 struct ll_statahead_info {
1432 struct dentry *sai_dentry;
1433 atomic_t sai_refcount; /* when access this struct, hold
1435 unsigned int sai_max; /* max ahead of lookup */
1436 __u64 sai_sent; /* stat requests sent count */
1437 __u64 sai_replied; /* stat requests which received
1439 __u64 sai_index; /* index of statahead entry */
1440 __u64 sai_index_wait; /* index of entry which is the
1441 * caller is waiting for */
1442 __u64 sai_hit; /* hit count */
1443 __u64 sai_miss; /* miss count:
1444 * for "ls -al" case, includes
1445 * hidden dentry miss;
1446 * for "ls -l" case, it does not
1447 * include hidden dentry miss.
1448 * "sai_miss_hidden" is used for
1451 unsigned int sai_consecutive_miss; /* consecutive miss */
1452 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1453 * is not a hidden one */
1454 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1456 unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
1458 sai_in_readpage:1;/* statahead is in readdir()*/
1459 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1460 struct task_struct *sai_task; /* stat-ahead thread */
1461 struct task_struct *sai_agl_task; /* AGL thread */
1462 struct list_head sai_interim_entries; /* entries which got async
1463 * stat reply, but not
1465 struct list_head sai_entries; /* completed entries */
1466 struct list_head sai_agls; /* AGLs to be sent */
1467 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1468 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1469 atomic_t sai_cache_count; /* entry count in cache */
1472 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1474 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1475 void ll_authorize_statahead(struct inode *dir, void *key);
1476 void ll_deauthorize_statahead(struct inode *dir, void *key);
1479 blkcnt_t dirty_cnt(struct inode *inode);
1481 int cl_glimpse_size0(struct inode *inode, int agl);
1482 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1483 struct inode *inode, struct cl_object *clob, int agl);
1485 static inline int cl_glimpse_size(struct inode *inode)
1487 return cl_glimpse_size0(inode, 0);
1490 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1491 * part of statahead */
1492 static inline int cl_agl(struct inode *inode)
1494 return cl_glimpse_size0(inode, 1);
1497 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1499 int cl_io_get(struct inode *inode, struct lu_env **envout,
1500 struct cl_io **ioout, __u16 *refcheck);
1502 static inline int ll_glimpse_size(struct inode *inode)
1504 struct ll_inode_info *lli = ll_i2info(inode);
1507 down_read(&lli->lli_glimpse_sem);
1508 rc = cl_glimpse_size(inode);
1509 lli->lli_glimpse_time = ktime_get();
1510 up_read(&lli->lli_glimpse_sem);
1514 /* dentry may statahead when statahead is enabled and current process has opened
1515 * parent directory, and this dentry hasn't accessed statahead cache before */
1517 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1519 struct ll_inode_info *lli;
1520 struct ll_dentry_data *ldd;
1522 if (ll_i2sbi(dir)->ll_sa_max == 0)
1525 lli = ll_i2info(dir);
1527 /* statahead is not allowed for this dir, there may be three causes:
1528 * 1. dir is not opened.
1529 * 2. statahead hit ratio is too low.
1530 * 3. previous stat started statahead thread failed. */
1531 if (!lli->lli_sa_enabled)
1534 /* not the same process, don't statahead */
1535 if (lli->lli_opendir_pid != current->pid)
1539 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1540 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1541 * For patchless client, lookup intent is not accurate, which may
1542 * misguide statahead. For example:
1543 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1544 * have the same intent -- IT_GETATTR, while one dentry should access
1545 * statahead cache once, otherwise statahead windows is messed up.
1546 * The solution is as following:
1547 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1548 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1549 * bypass interacting with statahead cache by checking
1550 * 'lld_sa_generation == lli->lli_sa_generation'.
1552 ldd = ll_d2d(dentry);
1553 if (ldd != NULL && lli->lli_sa_generation &&
1554 ldd->lld_sa_generation == lli->lli_sa_generation)
1560 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1561 enum cl_fsync_mode mode, int ignore_layout);
1563 static inline int ll_file_nolock(const struct file *file)
1565 struct ll_file_data *fd = file->private_data;
1566 struct inode *inode = file_inode((struct file *)file);
1568 LASSERT(fd != NULL);
1569 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1570 (ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
1573 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1574 struct lookup_intent *it, __u64 *bits)
1576 if (!it->it_lock_set) {
1577 struct lustre_handle handle;
1579 /* If this inode is a remote object, it will get two
1580 * separate locks in different namespaces, Master MDT,
1581 * where the name entry is, will grant LOOKUP lock,
1582 * remote MDT, where the object is, will grant
1583 * UPDATE|PERM lock. The inode will be attched to both
1584 * LOOKUP and PERM locks, so revoking either locks will
1585 * case the dcache being cleared */
1586 if (it->it_remote_lock_mode) {
1587 handle.cookie = it->it_remote_lock_handle;
1588 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1589 "(%p) for remote lock %#llx\n",
1590 PFID(ll_inode2fid(inode)), inode,
1592 md_set_lock_data(exp, &handle, inode, NULL);
1595 handle.cookie = it->it_lock_handle;
1597 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1598 " for lock %#llx\n",
1599 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1601 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1602 it->it_lock_set = 1;
1606 *bits = it->it_lock_bits;
1609 static inline int d_lustre_invalid(const struct dentry *dentry)
1611 struct ll_dentry_data *lld = ll_d2d(dentry);
1613 return (lld == NULL) || lld->lld_invalid;
1616 static inline void __d_lustre_invalidate(struct dentry *dentry)
1618 struct ll_dentry_data *lld = ll_d2d(dentry);
1621 lld->lld_invalid = 1;
1625 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1626 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1627 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1628 * unhash this dentry and kill it.
1630 static inline void d_lustre_invalidate(struct dentry *dentry)
1632 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1634 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1636 spin_lock(&dentry->d_lock);
1637 __d_lustre_invalidate(dentry);
1638 spin_unlock(&dentry->d_lock);
1641 static inline void d_lustre_revalidate(struct dentry *dentry)
1643 spin_lock(&dentry->d_lock);
1644 LASSERT(ll_d2d(dentry) != NULL);
1645 ll_d2d(dentry)->lld_invalid = 0;
1646 spin_unlock(&dentry->d_lock);
1649 static inline dev_t ll_compat_encode_dev(dev_t dev)
1651 /* The compat_sys_*stat*() syscalls will fail unless the
1652 * device majors and minors are both less than 256. Note that
1653 * the value returned here will be passed through
1654 * old_encode_dev() in cp_compat_stat(). And so we are not
1655 * trying to return a valid compat (u16) device number, just
1656 * one that will pass the old_valid_dev() check. */
1658 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1661 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1662 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1663 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1664 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1665 struct lu_extent *ext);
1667 int ll_xattr_init(void);
1668 void ll_xattr_fini(void);
1670 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1671 struct cl_page *page, enum cl_req_type crt);
1673 int ll_getparent(struct file *file, struct getparent __user *arg);
1676 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1677 enum op_xvalid xvalid, unsigned int attr_flags);
1679 extern struct lu_env *cl_inode_fini_env;
1680 extern __u16 cl_inode_fini_refcheck;
1682 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1683 void cl_inode_fini(struct inode *inode);
1685 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1686 u32 cl_fid_build_gen(const struct lu_fid *fid);
1688 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1690 return &ll_i2sbi(inode)->ll_pcc_super;
1693 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1695 return ll_i2pccs(ll_info2i(lli));
1698 #ifdef HAVE_LUSTRE_CRYPTO
1700 extern const struct llcrypt_operations lustre_cryptops;
1703 #endif /* LLITE_INTERNAL_H */