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/
32 #ifndef LLITE_INTERNAL_H
33 #define LLITE_INTERNAL_H
35 #include <lustre_disk.h> /* for s2sbi */
36 #include <lustre_linkea.h>
38 /* for struct cl_lock_descr and struct cl_io */
39 #include <cl_object.h>
40 #include <lustre_lmv.h>
41 #include <lustre_mdc.h>
42 #include <lustre_intent.h>
43 #include <linux/compat.h>
44 #include <linux/aio.h>
45 #include <linux/parser.h>
46 #include <linux/seqlock.h>
47 #include <lustre_compat.h>
48 #include <lustre_crypto.h>
49 #include <range_lock.h>
51 #include "vvp_internal.h"
53 #include "foreign_symlink.h"
59 #ifndef HAVE_VM_FAULT_RETRY
60 #define VM_FAULT_RETRY 0
63 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
64 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
65 #ifndef LOOKUP_CONTINUE
66 #define LOOKUP_CONTINUE LOOKUP_PARENT
69 #ifdef HAVE_DENTRY_D_U_D_ALIAS
70 #define d_alias d_u.d_alias
73 /** Only used on client-side for indicating the tail of dir hash/offset. */
74 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
75 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
77 /* 4UL * 1024 * 1024 */
78 #define LL_MAX_BLKSIZE_BITS 22
80 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
82 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
84 struct ll_dentry_data {
85 unsigned int lld_sa_generation;
86 unsigned int lld_invalid:1;
87 struct rcu_head lld_rcu_head;
90 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
92 #define LLI_INODE_MAGIC 0x111d0de5
93 #define LLI_INODE_DEAD 0xdeadd00d
95 struct ll_getname_data {
96 #ifdef HAVE_DIR_CONTEXT
97 struct dir_context ctx;
99 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
100 struct lu_fid lgd_fid; /* target fid we are looking for */
101 int lgd_found; /* inode matched? */
104 struct ll_grouplock {
105 struct lu_env *lg_env;
107 struct cl_lock *lg_lock;
108 unsigned long lg_gid;
111 /* See comment on trunc_sem_down_read_nowait */
112 struct ll_trunc_sem {
113 /* when positive, this is a count of readers, when -1, it indicates
114 * the semaphore is held for write, and 0 is unlocked
116 atomic_t ll_trunc_readers;
117 /* this tracks a count of waiting writers */
118 atomic_t ll_trunc_waiters;
122 LSA_PATTERN_NONE = 0x0000,
123 LSA_PATTERN_LIST = 0x0001,
124 LSA_PATTERN_FNAME = 0X0002,
128 struct ll_inode_info {
129 __u32 lli_inode_magic;
132 volatile unsigned long lli_flags;
133 struct posix_acl *lli_posix_acl;
135 /* identifying fields for both metadata and data stacks. */
136 struct lu_fid lli_fid;
137 /* master inode fid for stripe directory */
138 struct lu_fid lli_pfid;
140 /* We need all three because every inode may be opened in different
142 struct obd_client_handle *lli_mds_read_och;
143 struct obd_client_handle *lli_mds_write_och;
144 struct obd_client_handle *lli_mds_exec_och;
145 __u64 lli_open_fd_read_count;
146 __u64 lli_open_fd_write_count;
147 __u64 lli_open_fd_exec_count;
149 /* Protects access to och pointers and their usage counters */
150 struct mutex lli_och_mutex;
152 struct inode lli_vfs_inode;
154 /* the most recent timestamps obtained from mds */
159 spinlock_t lli_agl_lock;
160 bool lli_synced_to_mds;
162 /* inode specific open lock caching threshold */
163 u32 lli_open_thrsh_count;
164 /* Number of times this inode was opened */
165 u64 lli_open_fd_count;
166 /* When last close was performed on this inode */
167 ktime_t lli_close_fd_time;
169 /* Try to make the d::member and f::member are aligned. Before using
170 * these members, make clear whether it is directory or not. */
174 /* metadata statahead */
175 /* since parent-child threads can share the same @file
176 * struct, "opendir_key" is the token when dir close for
177 * case of parent exit before child -- it is me should
178 * cleanup the dir readahead. */
179 void *lli_opendir_key;
180 struct ll_statahead_info *lli_sai;
181 struct ll_statahead_context *lli_sax;
182 /* protect statahead stuff. */
183 spinlock_t lli_sa_lock;
184 /* "opendir_pid" is the token when lookup/revalid
185 * -- I am the owner of dir statahead. */
186 pid_t lli_opendir_pid;
187 /* directory depth to ROOT */
188 unsigned short lli_dir_depth;
189 /* directory depth to ancestor whose default LMV is
192 unsigned short lli_inherit_depth;
193 /* stat will try to access statahead entries or start
194 * statahead if this flag is set, and this flag will be
195 * set upon dir open, and cleared when dir is closed,
196 * statahead hit ratio is too low, or start statahead
198 unsigned short lli_sa_enabled:1,
199 /* default LMV is explicitly set in inode on MDT, this
200 * is for old server, or default LMV is set by
201 * "lfs setdirstripe -D".
203 lli_def_lsm_obj_set:1;
204 /* generation for statahead */
205 unsigned int lli_sa_generation;
206 /* access pattern for statahead */
207 enum ll_sa_pattern lli_sa_pattern;
208 /* rw lock protects lli_lsm_md */
209 struct rw_semaphore lli_lsm_sem;
210 /* directory stripe information */
211 struct lmv_stripe_object *lli_lsm_obj;
212 /* directory default LMV */
213 struct lmv_stripe_object *lli_def_lsm_obj;
216 /* for non-directory */
218 struct mutex lli_size_mutex;
219 struct task_struct *lli_size_lock_owner;
220 char *lli_symlink_name;
221 struct ll_trunc_sem lli_trunc_sem;
222 struct range_lock_tree lli_write_tree;
223 struct mutex lli_setattr_mutex;
225 struct rw_semaphore lli_glimpse_sem;
226 ktime_t lli_glimpse_time;
227 struct list_head lli_agl_list;
230 /* for writepage() only to communicate to fsync */
233 /* protect the file heat fields */
234 spinlock_t lli_heat_lock;
235 __u32 lli_heat_flags;
236 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
239 * Whenever a process try to read/write the file, the
240 * jobid, uid and gid of the process will be saved here,
241 * and it'll be packed into write RPC when flush later.
243 * So the read/write statistics or TBF rules for jobid,
244 * uid or gid will not be accurate if the file is shared
247 char lli_jobid[LUSTRE_JOBID_SIZE];
251 struct mutex lli_pcc_lock;
252 enum lu_pcc_state_flags lli_pcc_state;
254 * @lli_pcc_generation saves the gobal PCC generation
255 * when the file was successfully attached into PCC.
256 * The flags of the PCC dataset are saved in
258 * The gobal PCC generation will be increased when add
259 * or delete a PCC backend, or change the configuration
260 * parameters for PCC.
261 * If @lli_pcc_generation is same as the gobal PCC
262 * generation, we can use the saved flags of the PCC
263 * dataset to determine whether need to try auto attach
266 __u64 lli_pcc_generation;
267 enum pcc_dataset_flags lli_pcc_dsflags;
268 struct pcc_inode *lli_pcc_inode;
270 struct mutex lli_group_mutex;
271 __u64 lli_group_users;
272 unsigned long lli_group_gid;
274 __u64 lli_attr_valid;
276 __u64 lli_lazyblocks;
280 /* XXX: For following frequent used members, although they maybe special
281 * used for non-directory object, it is some time-wasting to check
282 * whether the object is directory or not before using them. On the
283 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
284 * the "ll_inode_info" size even if moving those members into u.f.
285 * So keep them out side.
287 * In the future, if more members are added only for directory,
288 * some of the following members can be moved into u.f.
290 struct cl_object *lli_clob;
292 /* mutex to request for layout lock exclusively. */
293 struct mutex lli_layout_mutex;
294 struct task_struct *lli_layout_lock_owner;
295 /* Layout version, protected by lli_layout_lock */
296 __u32 lli_layout_gen;
297 spinlock_t lli_layout_lock;
299 __u32 lli_projid; /* project id */
301 struct rw_semaphore lli_xattrs_list_rwsem;
302 struct mutex lli_xattrs_enq_lock;
303 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
304 struct list_head lli_lccs; /* list of ll_cl_context */
305 seqlock_t lli_page_inv_lock;
307 struct task_struct *lli_inode_lock_owner;
310 #ifndef HAVE_USER_NAMESPACE_ARG
311 #define inode_permission(ns, inode, mask) inode_permission(inode, mask)
312 #define generic_permission(ns, inode, mask) generic_permission(inode, mask)
313 #define simple_setattr(ns, de, iattr) simple_setattr(de, iattr)
314 #define ll_inode_permission(ns, inode, mask) ll_inode_permission(inode, mask)
315 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
316 #define ll_getattr(ns, path, stat, mask, fl) ll_getattr(path, stat, mask, fl)
317 #endif /* HAVE_INODEOPS_ENHANCED_GETATTR */
318 #define ll_setattr(ns, de, attr) ll_setattr(de, attr)
321 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
323 atomic_set(&sem->ll_trunc_readers, 0);
324 atomic_set(&sem->ll_trunc_waiters, 0);
327 /* This version of down read ignores waiting writers, meaning if the semaphore
328 * is already held for read, this down_read will 'join' that reader and also
329 * take the semaphore.
331 * This lets us avoid an unusual deadlock.
333 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
334 * in Lustre, in order to exclude truncates. Some of these paths then need to
335 * take the mmap_lock, while still holding the trunc_sem. The problem is that
336 * page faults hold the mmap_lock when calling in to Lustre, and then must also
337 * take the trunc_sem to exclude truncate.
339 * This means the locking order for trunc_sem and mmap_lock is sometimes AB,
340 * sometimes BA. This is almost OK because in both cases, we take the trunc
341 * sem for read, so it doesn't block.
343 * However, if a write mode user (truncate, a setattr op) arrives in the
344 * middle of this, the second reader on the truncate_sem will wait behind that
347 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
348 * to the mode in which they take the semaphore):
349 * reader (holding mmap_lock, needs truncate_sem)
351 * reader (holding truncate sem, waiting for mmap_lock)
353 * And so the readers deadlock.
355 * The solution is this modified semaphore, where this down_read ignores
356 * waiting write operations, and all waiters are woken up at once, so readers
357 * using down_read_nowait cannot get stuck behind waiting writers, regardless
358 * of the order they arrived in.
360 * down_read_nowait is only used in the page fault case, where we already hold
361 * the mmap_lock. This is because otherwise repeated read and write operations
362 * (which take the truncate sem) could prevent a truncate from ever starting.
363 * This could still happen with page faults, but without an even more complex
364 * mechanism, this is unavoidable.
368 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
370 wait_var_event(&sem->ll_trunc_readers,
371 atomic_inc_unless_negative(&sem->ll_trunc_readers));
374 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
376 wait_var_event(&sem->ll_trunc_readers,
377 atomic_read(&sem->ll_trunc_waiters) == 0 &&
378 atomic_inc_unless_negative(&sem->ll_trunc_readers));
381 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
383 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
384 atomic_read(&sem->ll_trunc_waiters))
385 wake_up_var(&sem->ll_trunc_readers);
388 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
390 atomic_inc(&sem->ll_trunc_waiters);
391 wait_var_event(&sem->ll_trunc_readers,
392 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
393 atomic_dec(&sem->ll_trunc_waiters);
396 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
398 atomic_set(&sem->ll_trunc_readers, 0);
399 /* match the smp_mb() in wait_var_event()->prepare_to_wait() */
401 wake_up_var(&sem->ll_trunc_readers);
404 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
405 static inline void lli_clear_acl(struct ll_inode_info *lli)
407 if (lli->lli_posix_acl) {
408 posix_acl_release(lli->lli_posix_acl);
409 lli->lli_posix_acl = NULL;
413 static inline void lli_replace_acl(struct ll_inode_info *lli,
414 struct lustre_md *md)
416 write_lock(&lli->lli_lock);
417 if (lli->lli_posix_acl)
418 posix_acl_release(lli->lli_posix_acl);
419 lli->lli_posix_acl = md->posix_acl;
420 write_unlock(&lli->lli_lock);
423 static inline void lli_clear_acl(struct ll_inode_info *lli)
427 static inline void lli_replace_acl(struct ll_inode_info *lli,
428 struct lustre_md *md)
433 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
437 spin_lock(&lli->lli_layout_lock);
438 gen = lli->lli_layout_gen;
439 spin_unlock(&lli->lli_layout_lock);
444 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
446 spin_lock(&lli->lli_layout_lock);
447 lli->lli_layout_gen = gen;
448 spin_unlock(&lli->lli_layout_lock);
451 enum ll_inode_flags {
452 /* File data is modified. */
453 LLIF_DATA_MODIFIED = 0,
454 /* File is being restored */
455 LLIF_FILE_RESTORING = 1,
456 /* Xattr cache is attached to the file */
457 LLIF_XATTR_CACHE = 2,
458 /* Project inherit */
459 LLIF_PROJECT_INHERIT = 3,
460 /* update atime from MDS even if it's older than local inode atime. */
461 LLIF_UPDATE_ATIME = 4,
462 /* foreign file/dir can be unlinked unconditionnaly */
463 LLIF_FOREIGN_REMOVABLE = 5,
464 /* 6 is not used for now */
465 /* Xattr cache is filled */
466 LLIF_XATTR_CACHE_FILLED = 7,
470 int ll_xattr_cache_destroy(struct inode *inode);
471 int ll_xattr_cache_empty(struct inode *inode);
473 int ll_xattr_cache_get(struct inode *inode,
479 int ll_xattr_cache_insert(struct inode *inode,
484 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
486 #ifdef CONFIG_SECURITY
487 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
488 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
494 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
496 #ifdef CONFIG_SECURITY
497 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
501 /* Only smack and selinux is known to use security contexts */
502 static inline bool ll_xattr_is_seclabel(const char *name)
504 return !strcmp(name, XATTR_NAME_SELINUX) ||
505 !strcmp(name, XATTR_NAME_SMACK);
508 static inline bool ll_xattr_suffix_is_seclabel(const char *suffix)
510 return !strcmp(suffix, XATTR_SELINUX_SUFFIX) ||
511 !strcmp(suffix, XATTR_SMACK_SUFFIX);
514 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
515 const char **secctx_name, __u32 *secctx_name_size,
516 void **secctx, __u32 *secctx_size,
519 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
522 int ll_inode_notifysecctx(struct inode *inode,
523 void *secctx, __u32 secctxlen);
525 void ll_secctx_name_free(struct ll_sb_info *sbi);
527 int ll_secctx_name_store(struct inode *in);
529 __u32 ll_secctx_name_get(struct ll_sb_info *sbi, const char **secctx_name);
531 int ll_security_secctx_name_filter(struct ll_sb_info *sbi, int xattr_type,
534 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
536 #ifdef HAVE_LUSTRE_CRYPTO
537 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
538 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
544 static inline void obd_connect_set_enc(struct obd_connect_data *data)
546 #ifdef HAVE_LUSTRE_CRYPTO
547 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
551 static inline bool obd_connect_has_name_enc(struct obd_connect_data *data)
553 #ifdef HAVE_LUSTRE_CRYPTO
554 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
555 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT_NAME;
561 static inline void obd_connect_set_name_enc(struct obd_connect_data *data)
563 #ifdef HAVE_LUSTRE_CRYPTO
564 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT_NAME;
568 static inline bool obd_connect_has_enc_fid2path(struct obd_connect_data *data)
570 #ifdef HAVE_LUSTRE_CRYPTO
571 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
572 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT_FID2PATH;
578 static inline void obd_connect_set_enc_fid2path(struct obd_connect_data *data)
580 #ifdef HAVE_LUSTRE_CRYPTO
581 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT_FID2PATH;
586 * Locking to guarantee consistency of non-atomic updates to long long i_size,
587 * consistency between file size and KMS.
589 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
592 void ll_inode_size_lock(struct inode *inode);
593 void ll_inode_size_unlock(struct inode *inode);
594 int ll_inode_size_trylock(struct inode *inode);
596 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
598 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
601 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
603 return ll_i2info(inode)->lli_pcc_inode;
606 static inline void ll_set_inode_lock_owner(struct inode *inode)
608 ll_i2info(inode)->lli_inode_lock_owner = current;
611 static inline void ll_clear_inode_lock_owner(struct inode *inode)
613 ll_i2info(inode)->lli_inode_lock_owner = NULL;
616 static inline struct task_struct *ll_get_inode_lock_owner(struct inode *inode)
618 return ll_i2info(inode)->lli_inode_lock_owner;
621 /* lock inode and set inode lock owener */
622 static inline void ll_inode_lock(struct inode *inode)
625 ll_set_inode_lock_owner(inode);
628 /* clear inode lock owner and unlock it */
629 static inline void ll_inode_unlock(struct inode *inode)
631 ll_clear_inode_lock_owner(inode);
635 /* default to use at least 16M for fast read if possible */
636 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
638 /* default read-ahead on a given client mountpoint. */
639 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(1024UL)
641 /* default read-ahead for a single file descriptor */
642 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX MiB_TO_PAGES(256UL)
644 /* default read-ahead full files smaller than limit on the second read */
645 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
647 /* default range pages */
648 #define SBI_DEFAULT_RA_RANGE_PAGES MiB_TO_PAGES(1ULL)
650 /* Min range pages */
651 #define RA_MIN_MMAP_RANGE_PAGES 16UL
656 RA_STAT_DISTANT_READPAGE,
657 RA_STAT_MISS_IN_WINDOW,
658 RA_STAT_FAILED_GRAB_PAGE,
659 RA_STAT_FAILED_MATCH,
664 RA_STAT_MAX_IN_FLIGHT,
665 RA_STAT_WRONG_GRAB_PAGE,
666 RA_STAT_FAILED_REACH_END,
668 RA_STAT_FAILED_FAST_READ,
669 RA_STAT_MMAP_RANGE_READ,
670 RA_STAT_READAHEAD_PAGES,
675 atomic_t ra_cur_pages;
676 unsigned long ra_max_pages;
677 unsigned long ra_max_pages_per_file;
678 unsigned long ra_range_pages;
679 unsigned long ra_max_read_ahead_whole_pages;
680 struct workqueue_struct *ll_readahead_wq;
682 * Max number of active works could be triggered
683 * for async readahead.
685 unsigned int ra_async_max_active;
686 /* how many async readahead triggered in flight */
687 atomic_t ra_async_inflight;
688 /* Threshold to control when to trigger async readahead */
689 unsigned long ra_async_pages_per_file_threshold;
692 /* ra_io_arg will be filled in the beginning of ll_readahead with
693 * ras_lock, then the following ll_read_ahead_pages will read RA
694 * pages according to this arg, all the items in this structure are
695 * counted by page index.
698 pgoff_t ria_start_idx; /* start offset of read-ahead*/
699 pgoff_t ria_end_idx; /* end offset of read-ahead*/
700 unsigned long ria_reserved; /* reserved pages for read-ahead */
701 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
702 bool ria_eof; /* reach end of file */
703 /* If stride read pattern is detected, ria_stoff is the byte offset
704 * where stride read is started. Note: for normal read-ahead, the
705 * value here is meaningless, and also it will not be accessed*/
707 /* ria_length and ria_bytes are the length and pages length in the
708 * stride I/O mode. And they will also be used to check whether
709 * it is stride I/O read-ahead in the read-ahead pages*/
714 /* LL_HIST_MAX=32 causes an overflow */
715 #define LL_HIST_MAX 28
716 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
717 #define LL_PROCESS_HIST_MAX 10
718 struct per_process_info {
720 struct obd_histogram pp_r_hist;
721 struct obd_histogram pp_w_hist;
724 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
725 struct ll_rw_extents_info {
727 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
730 #define LL_OFFSET_HIST_MAX 100
731 struct ll_rw_process_info {
734 loff_t rw_range_start;
736 loff_t rw_last_file_pos;
738 size_t rw_smallest_extent;
739 size_t rw_largest_extent;
740 struct ll_file_data *rw_last_file;
743 enum stats_track_type {
744 STATS_TRACK_ALL = 0, /* track all processes */
745 STATS_TRACK_PID, /* track process with this pid */
746 STATS_TRACK_PPID, /* track processes with this ppid */
747 STATS_TRACK_GID, /* track processes with this gid */
751 /* flags for sbi->ll_flags */
753 LL_SBI_NOLCK, /* DLM locking disabled directio-only */
754 LL_SBI_CHECKSUM, /* checksum each page as it's written */
755 LL_SBI_LOCALFLOCK, /* local flocks instead of fs-wide */
756 LL_SBI_FLOCK, /* flock enabled */
757 LL_SBI_USER_XATTR, /* support user xattr */
758 LL_SBI_LRU_RESIZE, /* lru resize support */
759 LL_SBI_LAZYSTATFS, /* lazystatfs mount option */
760 LL_SBI_32BIT_API, /* generate 32 bit inodes. */
761 LL_SBI_USER_FID2PATH, /* fid2path by unprivileged users */
762 LL_SBI_VERBOSE, /* verbose mount/umount */
763 LL_SBI_ALWAYS_PING, /* ping even if server suppress_pings */
764 LL_SBI_TEST_DUMMY_ENCRYPTION, /* test dummy encryption */
765 LL_SBI_ENCRYPT, /* client side encryption */
766 LL_SBI_FOREIGN_SYMLINK, /* foreign fake-symlink support */
767 LL_SBI_FOREIGN_SYMLINK_UPCALL, /* foreign fake-symlink upcall set */
768 LL_SBI_NUM_MOUNT_OPT,
770 LL_SBI_ACL, /* support ACL */
771 LL_SBI_AGL_ENABLED, /* enable agl */
772 LL_SBI_64BIT_HASH, /* support 64-bits dir hash/offset */
773 LL_SBI_LAYOUT_LOCK, /* layout lock support */
774 LL_SBI_XATTR_CACHE, /* support for xattr cache */
775 LL_SBI_NOROOTSQUASH, /* do not apply root squash */
776 LL_SBI_FAST_READ, /* fast read support */
777 LL_SBI_FILE_SECCTX, /* file security context at create */
778 LL_SBI_TINY_WRITE, /* tiny write support */
779 LL_SBI_FILE_HEAT, /* file heat support */
780 LL_SBI_PARALLEL_DIO, /* parallel (async) O_DIRECT RPCs */
781 LL_SBI_ENCRYPT_NAME, /* name encryption */
785 int ll_sbi_flags_seq_show(struct seq_file *m, void *v);
787 /* This is embedded into llite super-blocks to keep track of connect
788 * flags (capabilities) supported by all imports given mount is
790 struct lustre_client_ocd {
791 /* This is conjunction of connect_flags across all imports
792 * (LOVs) this mount is connected to. This field is updated by
793 * cl_ocd_update() under ->lco_lock. */
795 struct mutex lco_lock;
796 struct obd_export *lco_md_exp;
797 struct obd_export *lco_dt_exp;
801 /* this protects pglist and ra_info. It isn't safe to
802 * grab from interrupt contexts */
804 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
805 spinlock_t ll_process_lock; /* ll_rw_process_info */
806 struct lustre_sb_info *lsi;
807 struct obd_uuid ll_sb_uuid;
808 struct obd_export *ll_md_exp;
809 struct obd_export *ll_dt_exp;
810 struct obd_device *ll_md_obd;
811 struct obd_device *ll_dt_obd;
812 struct dentry *ll_debugfs_entry;
813 struct lu_fid ll_root_fid; /* root object fid */
814 struct mnt_namespace *ll_mnt_ns;
816 DECLARE_BITMAP(ll_flags, LL_SBI_NUM_FLAGS); /* enum ll_sbi_flags */
817 unsigned int ll_xattr_cache_enabled:1,
818 ll_xattr_cache_set:1, /* already set to 0/1 */
819 ll_client_common_fill_super_succeeded:1,
821 ll_inode_cache_enabled:1;
823 struct lustre_client_ocd ll_lco;
825 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
827 /* Used to track "unstable" pages on a client, and maintain a
828 * LRU list of clean pages. An "unstable" page is defined as
829 * any page which is sent to a server as part of a bulk request,
830 * but is uncommitted to stable storage. */
831 struct cl_client_cache *ll_cache;
833 struct lprocfs_stats *ll_ra_stats;
835 struct ll_ra_info ll_ra_info;
836 unsigned int ll_namelen;
837 const struct file_operations *ll_fop;
839 struct lu_site *ll_site;
840 struct cl_device *ll_cl;
843 struct ll_rw_extents_info *ll_rw_extents_info;
844 int ll_extent_process_count;
845 unsigned int ll_offset_process_count;
846 struct ll_rw_process_info *ll_rw_process_info;
847 struct ll_rw_process_info *ll_rw_offset_info;
848 ktime_t ll_process_stats_init;
849 unsigned int ll_rw_offset_entry_count;
850 int ll_stats_track_id;
851 enum stats_track_type ll_stats_track_type;
854 /* metadata stat-ahead */
855 unsigned int ll_sa_running_max;/* max concurrent
856 * statahead instances */
857 unsigned int ll_sa_batch_max;/* max SUB request count in
858 * a batch PTLRPC request */
859 unsigned int ll_sa_max; /* max statahead RPCs */
860 atomic_t ll_sa_total; /* statahead thread started
862 atomic_t ll_sa_wrong; /* statahead thread stopped for
864 atomic_t ll_sa_running; /* running statahead thread
866 atomic_t ll_agl_total; /* AGL thread started count */
867 atomic_t ll_sa_hit_total; /* total hit count */
868 atomic_t ll_sa_miss_total; /* total miss count */
870 dev_t ll_sdev_orig; /* save s_dev before assign for
873 struct root_squash_info ll_squash;
876 /* st_blksize returned by stat(2), when non-zero */
877 unsigned int ll_stat_blksize;
879 /* maximum relative age of cached statfs results */
880 unsigned int ll_statfs_max_age;
882 struct kset ll_kset; /* sysfs object */
883 struct completion ll_kobj_unregister;
886 unsigned int ll_heat_decay_weight;
887 unsigned int ll_heat_period_second;
889 /* Opens of the same inode before we start requesting open lock */
890 u32 ll_oc_thrsh_count;
892 /* Time in ms between last inode close and next open to be considered
893 * instant back to back and would trigger an open lock request
897 /* Time in ms after last file close that we no longer count prior opens*/
900 /* filesystem fsname */
901 char ll_fsname[LUSTRE_MAXFSNAME + 1];
903 /* Persistent Client Cache */
904 struct pcc_super ll_pcc_super;
906 /* to protect vs updates in all following foreign symlink fields */
907 struct rw_semaphore ll_foreign_symlink_sem;
908 /* foreign symlink path prefix */
909 char *ll_foreign_symlink_prefix;
910 /* full prefix size including leading '\0' */
911 size_t ll_foreign_symlink_prefix_size;
912 /* foreign symlink path upcall */
913 char *ll_foreign_symlink_upcall;
914 /* foreign symlink path upcall infos */
915 struct ll_foreign_symlink_upcall_item *ll_foreign_symlink_upcall_items;
916 /* foreign symlink path upcall nb infos */
917 unsigned int ll_foreign_symlink_upcall_nb_items;
919 /* cached file security context xattr name. e.g: security.selinux */
920 char *ll_secctx_name;
921 __u32 ll_secctx_name_size;
924 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
925 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
927 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT (5)
928 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MS (100) /* 0.1 second */
929 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS (60000) /* 1 minute */
932 * per file-descriptor read-ahead data.
934 struct ll_readahead_state {
936 /* End byte that read(2) try to read. */
937 loff_t ras_last_read_end_bytes;
939 * number of bytes read after last read-ahead window reset. As window
940 * is reset on each seek, this is effectively a number of consecutive
941 * accesses. Maybe ->ras_accessed_in_window is better name.
943 * XXX nikita: window is also reset (by ras_update()) when Lustre
944 * believes that memory pressure evicts read-ahead pages. In that
945 * case, it probably doesn't make sense to expand window to
946 * PTLRPC_MAX_BRW_PAGES on the third access.
948 loff_t ras_consecutive_bytes;
950 * number of read requests after the last read-ahead window reset
951 * As window is reset on each seek, this is effectively the number
952 * on consecutive read request and is used to trigger read-ahead.
954 unsigned long ras_consecutive_requests;
956 * Parameters of current read-ahead window. Handled by
957 * ras_update(). On the initial access to the file or after a seek,
958 * window is reset to 0. After 3 consecutive accesses, window is
959 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
960 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
962 pgoff_t ras_window_start_idx;
963 pgoff_t ras_window_pages;
965 /* Page index where min range read starts */
966 pgoff_t ras_range_min_start_idx;
967 /* Page index where mmap range read ends */
968 pgoff_t ras_range_max_end_idx;
969 /* number of mmap pages where last time detected */
970 pgoff_t ras_last_range_pages;
971 /* number of mmap range requests */
972 pgoff_t ras_range_requests;
975 * Optimal RPC size in pages.
976 * It decides how many pages will be sent for each read-ahead.
978 unsigned long ras_rpc_pages;
980 * Where next read-ahead should start at. This lies within read-ahead
981 * window. Read-ahead window is read in pieces rather than at once
982 * because: 1. lustre limits total number of pages under read-ahead by
983 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
984 * not covered by DLM lock.
986 pgoff_t ras_next_readahead_idx;
988 * Total number of ll_file_read requests issued, reads originating
989 * due to mmap are not counted in this total. This value is used to
990 * trigger full file read-ahead after multiple reads to a small file.
992 unsigned long ras_requests;
994 * The following 3 items are used for detecting the stride I/O
996 * In stride I/O mode,
997 * ...............|-----data-----|****gap*****|--------|******|....
998 * offset |-stride_bytes-|-stride_gap-|
999 * ras_stride_offset = offset;
1000 * ras_stride_length = stride_bytes + stride_gap;
1001 * ras_stride_bytes = stride_bytes;
1002 * Note: all these three items are counted by bytes.
1004 loff_t ras_stride_offset;
1005 loff_t ras_stride_length;
1006 loff_t ras_stride_bytes;
1008 * number of consecutive stride request count, and it is similar as
1009 * ras_consecutive_requests, but used for stride I/O mode.
1010 * Note: only more than 2 consecutive stride request are detected,
1011 * stride read-ahead will be enable
1013 unsigned long ras_consecutive_stride_requests;
1014 /* index of the last page that async readahead starts */
1015 pgoff_t ras_async_last_readpage_idx;
1016 /* whether we should increase readahead window */
1017 bool ras_need_increase_window;
1018 /* whether ra miss check should be skipped */
1019 bool ras_no_miss_check;
1022 struct ll_readahead_work {
1023 /** File to readahead */
1024 struct file *lrw_file;
1025 pgoff_t lrw_start_idx;
1026 pgoff_t lrw_end_idx;
1029 /* async worker to handler read */
1030 struct work_struct lrw_readahead_work;
1031 char lrw_jobid[LUSTRE_JOBID_SIZE];
1034 extern struct kmem_cache *ll_file_data_slab;
1035 struct lustre_handle;
1036 struct ll_file_data {
1037 struct ll_readahead_state fd_ras;
1038 struct ll_grouplock fd_grouplock;
1042 /* openhandle if lease exists for this file.
1043 * Borrow lli->lli_och_mutex to protect assignment */
1044 struct obd_client_handle *fd_lease_och;
1045 struct obd_client_handle *fd_och;
1046 struct file *fd_file;
1047 /* Indicate whether need to report failure when close.
1048 * true: failure is known, not report again.
1049 * false: unknown failure, should report. */
1050 bool fd_write_failed;
1051 bool ll_lock_no_expand;
1052 /* Used by mirrored file to lead IOs to a specific mirror, usually
1053 * for mirror resync. 0 means default. */
1054 __u32 fd_designated_mirror;
1055 /* The layout version when resync starts. Resync I/O should carry this
1056 * layout version for verification to OST objects */
1057 __u32 fd_layout_version;
1058 struct pcc_file fd_pcc_file;
1059 /* striped directory may read partially if some stripe inaccessible,
1060 * -errno is saved here, and will return to user in close().
1062 int fd_partial_readdir_rc;
1063 /* mdtest unique/shared dir stat mode: per process statahead struct. */
1064 struct ll_statahead_info *fd_sai;
1067 void llite_tunables_unregister(void);
1068 int llite_tunables_register(void);
1070 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
1072 return &lli->lli_vfs_inode;
1075 __u32 ll_i2suppgid(struct inode *i);
1076 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
1078 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
1080 #if BITS_PER_LONG == 32
1082 #elif defined(CONFIG_COMPAT)
1083 if (unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags)))
1086 # ifdef CONFIG_X86_X32
1087 /* in_compat_syscall() returns true when called from a kthread
1088 * and CONFIG_X86_X32 is enabled, which is wrong. So check
1089 * whether the caller comes from a syscall (ie. not a kthread)
1090 * before calling in_compat_syscall(). */
1091 if (current->flags & PF_KTHREAD)
1095 return unlikely(in_compat_syscall());
1097 return unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1101 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
1103 return test_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1106 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
1108 return test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1111 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
1113 return test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1116 static inline bool ll_sbi_has_foreign_symlink(struct ll_sb_info *sbi)
1118 return test_bit(LL_SBI_FOREIGN_SYMLINK, sbi->ll_flags);
1121 static inline bool ll_sbi_has_parallel_dio(struct ll_sb_info *sbi)
1123 return test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1126 void ll_ras_enter(struct file *f, loff_t pos, size_t bytes);
1128 /* llite/lcommon_misc.c */
1129 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
1130 enum obd_notify_event ev, void *owner);
1131 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
1132 struct ll_grouplock *lg);
1133 void cl_put_grouplock(struct ll_grouplock *lg);
1135 /* llite/lproc_llite.c */
1136 int ll_debugfs_register_super(struct super_block *sb, const char *name);
1137 void ll_debugfs_unregister_super(struct super_block *sb);
1138 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
1139 void ll_free_rw_stats_info(struct ll_sb_info *sbi);
1142 LPROC_LL_READ_BYTES,
1143 LPROC_LL_WRITE_BYTES,
1170 LPROC_LL_GETXATTR_HITS,
1172 LPROC_LL_REMOVEXATTR,
1173 LPROC_LL_INODE_PERM,
1175 LPROC_LL_INODE_OCOUNT,
1176 LPROC_LL_INODE_OPCLTM,
1177 LPROC_LL_FILE_OPCODES
1181 enum get_default_layout_type {
1182 GET_DEFAULT_LAYOUT_ROOT = 1,
1185 extern const struct file_operations ll_dir_operations;
1186 extern const struct inode_operations ll_dir_inode_operations;
1187 #ifdef HAVE_DIR_CONTEXT
1188 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1189 struct dir_context *ctx, int *partial_readdir_rc);
1191 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1192 void *cookie, filldir_t filldir, int *partial_readdir_rc);
1194 int ll_get_mdt_idx(struct inode *inode);
1195 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1196 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1197 __u64 offset, int *partial_readdir_rc);
1198 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1199 int quotactl_ioctl(struct super_block *sb, struct if_quotactl *qctl);
1202 extern const struct inode_operations ll_special_inode_operations;
1204 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1205 struct lustre_md *lic);
1206 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1207 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1208 void *data, int flag);
1209 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1210 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1211 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1214 int ll_writepage(struct page *page, struct writeback_control *wbc);
1215 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1216 int ll_readpage(struct file *file, struct page *page);
1217 #ifdef HAVE_AOPS_READ_FOLIO
1218 int ll_read_folio(struct file *file, struct folio *folio);
1220 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1221 struct cl_page *page, struct file *file);
1222 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1223 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1226 void ll_cl_add(struct inode *inode, const struct lu_env *env, struct cl_io *io,
1227 enum lcc_type type);
1228 void ll_cl_remove(struct inode *inode, const struct lu_env *env);
1229 struct ll_cl_context *ll_cl_find(struct inode *inode);
1231 extern const struct address_space_operations ll_aops;
1234 extern const struct inode_operations ll_file_inode_operations;
1235 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi);
1236 extern int ll_have_md_lock(struct obd_export *exp, struct inode *inode,
1237 __u64 *bits, enum ldlm_mode l_req_mode);
1238 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1239 struct lustre_handle *lockh, __u64 flags,
1240 enum ldlm_mode mode);
1242 int ll_file_open(struct inode *inode, struct file *file);
1243 int ll_file_release(struct inode *inode, struct file *file);
1244 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1245 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1246 void ll_track_file_opens(struct inode *inode);
1247 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1248 struct ll_file_data *file, loff_t pos,
1249 size_t count, int rw);
1250 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
1251 int ll_getattr(struct mnt_idmap *, const struct path *path,
1252 struct kstat *stat, u32 request_mask, unsigned int flags);
1254 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1255 #endif /* HAVE_USER_NAMESPACE_ARG */
1256 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1257 unsigned int flags, bool foreign);
1258 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1259 struct posix_acl *ll_get_acl(
1260 #ifdef HAVE_ACL_WITH_DENTRY
1261 struct mnt_idmap *, struct dentry *, int);
1262 #elif defined HAVE_GET_ACL_RCU_ARG
1263 struct inode *inode, int type, bool rcu);
1265 struct inode *inode, int type);
1266 #endif /* HAVE_GET_ACL_RCU_ARG */
1268 int ll_set_acl(struct mnt_idmap *mnt_userns,
1269 #ifdef HAVE_ACL_WITH_DENTRY
1270 struct dentry *dentry,
1272 struct inode *inode,
1274 struct posix_acl *acl, int type);
1275 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1276 #define ll_get_acl NULL
1277 #define ll_set_acl NULL
1278 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1280 static inline int ll_xflags_to_inode_flags(int xflags)
1282 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1283 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1284 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1285 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1288 static inline int ll_inode_flags_to_xflags(int inode_flags)
1290 return ((inode_flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1291 ((inode_flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1292 ((inode_flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1293 ((inode_flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1296 int ll_migrate(struct inode *parent, struct file *file,
1297 struct lmv_user_md *lum, const char *name, __u32 flags);
1298 int ll_get_fid_by_name(struct inode *parent, const char *name,
1299 int namelen, struct lu_fid *fid, struct inode **inode);
1300 int ll_inode_permission(struct mnt_idmap *, struct inode *inode, int mask);
1301 int ll_ioctl_check_project(struct inode *inode, __u32 xflags, __u32 projid);
1302 int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid);
1303 #ifndef HAVE_FILEATTR_GET
1304 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1306 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1309 int ll_ioctl_project(struct file *file, unsigned int cmd, void __user *uarg);
1310 int ll_ioctl_ahead(struct file *file, struct llapi_lu_ladvise2 *ladvise);
1312 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1313 __u64 flags, struct lov_user_md *lum,
1315 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1316 struct lov_mds_md **lmm, int *lmm_size,
1317 struct ptlrpc_request **request);
1318 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1320 int ll_dir_get_default_layout(struct inode *inode, void **plmm, int *plmm_size,
1321 struct ptlrpc_request **request, u64 valid,
1322 enum get_default_layout_type type);
1323 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1324 int *lmm_size, struct ptlrpc_request **request,
1325 struct ptlrpc_request **root_request, u64 valid);
1326 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1327 struct ptlrpc_request **request, u64 valid);
1328 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1329 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1330 int ll_merge_attr_try(const struct lu_env *env, struct inode *inode);
1331 int ll_fid2path(struct inode *inode, void __user *arg);
1332 int __ll_fid2path(struct inode *inode, struct getinfo_fid2path *gfout,
1333 size_t outsize, __u32 pathlen_orig);
1334 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1335 int ll_hsm_release(struct inode *inode);
1336 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1337 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1339 /* llite/dcache.c */
1341 extern const struct dentry_operations ll_d_ops;
1343 bool ll_d_setup(struct dentry *de, bool do_put);
1345 static inline bool lld_is_init(struct dentry *dentry)
1347 return ll_d2d(dentry);
1350 #define ll_d_setup(de, do_put) (true)
1351 #define lld_is_init(dentry) (true)
1354 void ll_intent_drop_lock(struct lookup_intent *);
1355 void ll_intent_release(struct lookup_intent *);
1356 void ll_prune_aliases(struct inode *inode);
1357 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1358 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1359 struct lookup_intent *it, struct dentry *de);
1361 /* llite/llite_lib.c */
1362 extern const struct super_operations lustre_super_operations;
1364 void ll_lli_init(struct ll_inode_info *lli);
1365 int ll_fill_super(struct super_block *sb);
1366 void ll_put_super(struct super_block *sb);
1367 void ll_kill_super(struct super_block *sb);
1368 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1369 void ll_dir_clear_lsm_md(struct inode *inode);
1370 void ll_clear_inode(struct inode *inode);
1371 int volatile_ref_file(const char *volatile_name, int volatile_len,
1372 struct file **ref_file);
1373 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1374 enum op_xvalid xvalid, bool hsm_import);
1375 int ll_setattr(struct mnt_idmap *, struct dentry *de, struct iattr *attr);
1376 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1377 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1379 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1380 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags);
1381 void ll_update_dir_depth_dmv(struct inode *dir, struct dentry *de);
1382 int ll_read_inode2(struct inode *inode, void *opaque);
1383 void ll_truncate_inode_pages_final(struct inode *inode);
1384 void ll_delete_inode(struct inode *inode);
1385 #ifdef HAVE_FILEATTR_GET
1386 int ll_fileattr_get(struct dentry *dentry, struct fileattr *fa);
1387 int ll_fileattr_set(struct mnt_idmap *mnt_userns,
1388 struct dentry *dentry, struct fileattr *fa);
1390 int ll_iocontrol(struct inode *inode, struct file *file,
1391 unsigned int cmd, void __user *uarg);
1392 int ll_flush_ctx(struct inode *inode);
1393 void ll_umount_begin(struct super_block *sb);
1394 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1395 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1396 void ll_dirty_page_discard_warn(struct inode *inode, int ioret);
1397 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
1398 struct super_block *sb, struct lookup_intent *it);
1399 int ll_obd_statfs(struct inode *inode, void __user *arg);
1400 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1401 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1402 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1404 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1405 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1406 struct inode *i1, struct inode *i2,
1407 const char *name, size_t namelen,
1408 __u32 mode, enum md_op_code opc,
1410 void ll_finish_md_op_data(struct md_op_data *op_data);
1411 int ll_get_obd_name(struct inode *inode, unsigned int cmd, void __user *arg);
1412 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1413 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1414 struct lov_user_md **kbuf);
1415 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill);
1417 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1419 /* Compute expected user md size when passing in a md from user space */
1420 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1422 switch (lum->lmm_magic) {
1423 case LOV_USER_MAGIC_V1:
1424 return sizeof(struct lov_user_md_v1);
1425 case LOV_USER_MAGIC_V3:
1426 return sizeof(struct lov_user_md_v3);
1427 case LOV_USER_MAGIC_SPECIFIC:
1428 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1431 return lov_user_md_size(lum->lmm_stripe_count,
1432 LOV_USER_MAGIC_SPECIFIC);
1433 case LOV_USER_MAGIC_COMP_V1:
1434 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1435 case LOV_USER_MAGIC_FOREIGN:
1436 return foreign_size(lum);
1442 /* llite/llite_nfs.c */
1443 extern const struct export_operations lustre_export_operations;
1444 __u32 get_uuid2int(const char *name, int len);
1445 struct inode *search_inode_for_lustre(struct super_block *sb,
1446 const struct lu_fid *fid);
1447 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1449 /* llite/symlink.c */
1450 extern const struct inode_operations ll_fast_symlink_inode_operations;
1453 * IO arguments for various VFS I/O interfaces.
1455 struct vvp_io_args {
1456 /** normal/sendfile/splice */
1459 struct kiocb *via_iocb;
1460 struct iov_iter *via_iter;
1470 struct ll_cl_context {
1471 struct list_head lcc_list;
1473 const struct lu_env *lcc_env;
1474 struct cl_io *lcc_io;
1475 struct cl_page *lcc_page;
1476 enum lcc_type lcc_type;
1477 pgoff_t lcc_end_index;
1480 struct ll_thread_info {
1481 struct vvp_io_args lti_args;
1482 struct ra_io_arg lti_ria;
1483 struct ll_cl_context lti_io_ctx;
1486 extern struct lu_context_key ll_thread_key;
1488 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1490 struct ll_thread_info *lti;
1492 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1493 LASSERT(lti != NULL);
1498 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1500 return &ll_env_info(env)->lti_args;
1503 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1504 struct vvp_io_args *args);
1506 /* llite/llite_mmap.c */
1508 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1509 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1510 unsigned long addr, size_t count);
1511 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1514 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1516 /* don't need an addref as the sb_info should be holding one */
1517 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1519 return ll_s2sbi(sb)->ll_dt_exp;
1522 /* don't need an addref as the sb_info should be holding one */
1523 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1525 return ll_s2sbi(sb)->ll_md_exp;
1528 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1530 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1536 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1537 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1539 return ll_s2sbi(inode->i_sb);
1542 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1544 return ll_s2dtexp(inode->i_sb);
1547 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1549 return ll_s2mdexp(inode->i_sb);
1552 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1556 LASSERT(inode != NULL);
1557 fid = &ll_i2info(inode)->lli_fid;
1562 static inline bool ll_dir_striped(struct inode *inode)
1567 if (!S_ISDIR(inode->i_mode))
1570 down_read(&ll_i2info(inode)->lli_lsm_sem);
1571 rc = !!(ll_i2info(inode)->lli_lsm_obj &&
1572 lmv_dir_striped(ll_i2info(inode)->lli_lsm_obj));
1573 up_read(&ll_i2info(inode)->lli_lsm_sem);
1578 static inline loff_t ll_file_maxbytes(struct inode *inode)
1580 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1583 return MAX_LFS_FILESIZE;
1585 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1589 extern const struct xattr_handler *ll_xattr_handlers[];
1591 #define XATTR_USER_T 1
1592 #define XATTR_TRUSTED_T 2
1593 #define XATTR_SECURITY_T 3
1594 #define XATTR_ACL_ACCESS_T 4
1595 #define XATTR_ACL_DEFAULT_T 5
1596 #define XATTR_LUSTRE_T 6
1597 #define XATTR_OTHER_T 7
1598 #define XATTR_ENCRYPTION_T 9
1600 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1601 int ll_xattr_list(struct inode *inode, const char *name, int type,
1602 void *buffer, size_t size, u64 valid);
1603 const struct xattr_handler *get_xattr_type(const char *name);
1606 * Common IO arguments for various VFS I/O interfaces.
1608 int cl_sb_init(struct super_block *sb);
1609 int cl_sb_fini(struct super_block *sb);
1611 enum ras_update_flags {
1615 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1616 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1620 #define LL_SA_RPC_MIN 8
1621 #define LL_SA_RPC_DEF 128
1622 #define LL_SA_RPC_MAX 2048
1624 /* XXX: If want to support more concurrent statahead instances,
1625 * please consider to decentralize the RPC lists attached
1626 * on related import, such as imp_{sending,delayed}_list.
1628 #define LL_SA_RUNNING_MAX 256
1629 #define LL_SA_RUNNING_DEF 16
1631 #define LL_SA_BATCH_MAX 1024
1632 #define LL_SA_BATCH_DEF 64
1634 #define LL_SA_CACHE_BIT 6
1635 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1636 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1638 /* statahead controller, per process struct, for dir only */
1639 struct ll_statahead_info {
1640 struct dentry *sai_dentry;
1641 atomic_t sai_refcount; /* when access this struct, hold
1643 unsigned int sai_max; /* max ahead of lookup */
1644 __u64 sai_sent; /* stat requests sent count */
1645 __u64 sai_replied; /* stat requests which received
1647 __u64 sai_index; /* index of statahead entry */
1648 __u64 sai_index_wait; /* index of entry which is the
1649 * caller is waiting for */
1650 __u64 sai_hit; /* hit count */
1651 __u64 sai_miss; /* miss count:
1652 * for "ls -al" case, includes
1653 * hidden dentry miss;
1654 * for "ls -l" case, it does not
1655 * include hidden dentry miss.
1656 * "sai_miss_hidden" is used for
1659 unsigned int sai_consecutive_miss; /* consecutive miss */
1660 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1661 * is not a hidden one */
1662 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1664 unsigned int sai_ls_all:1; /* "ls -al", do stat-ahead for
1666 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1667 struct task_struct *sai_task; /* stat-ahead thread */
1668 struct task_struct *sai_agl_task; /* AGL thread */
1669 struct list_head sai_entries; /* completed entries */
1670 struct list_head sai_agls; /* AGLs to be sent */
1671 atomic_t sai_cache_count; /* entry count in cache */
1672 struct lu_batch *sai_bh;
1673 __u32 sai_max_batch_count;
1674 __u64 sai_index_end;
1678 char sai_fname[NAME_MAX];
1681 /* Per inode statahead information */
1682 struct ll_statahead_context {
1683 struct inode *sax_inode;
1684 atomic_t sax_refcount;
1686 struct list_head sax_cache[LL_SA_CACHE_SIZE];
1687 spinlock_t sax_cache_lock[LL_SA_CACHE_SIZE];
1690 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1692 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1693 void ll_authorize_statahead(struct inode *dir, void *key);
1694 void ll_deauthorize_statahead(struct inode *dir, void *key);
1697 blkcnt_t dirty_cnt(struct inode *inode);
1699 int cl_glimpse_size0(struct inode *inode, int agl);
1700 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1701 struct inode *inode, struct cl_object *clob, int agl);
1703 static inline int cl_glimpse_size(struct inode *inode)
1705 return cl_glimpse_size0(inode, 0);
1708 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1709 * part of statahead */
1710 static inline int cl_agl(struct inode *inode)
1712 return cl_glimpse_size0(inode, 1);
1715 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1717 int cl_io_get(struct inode *inode, struct lu_env **envout,
1718 struct cl_io **ioout, __u16 *refcheck);
1720 static inline int ll_glimpse_size(struct inode *inode)
1722 struct ll_inode_info *lli = ll_i2info(inode);
1725 down_read(&lli->lli_glimpse_sem);
1726 rc = cl_glimpse_size(inode);
1727 lli->lli_glimpse_time = ktime_get();
1728 up_read(&lli->lli_glimpse_sem);
1732 /* dentry may statahead when statahead is enabled and current process has opened
1733 * parent directory, and this dentry hasn't accessed statahead cache before */
1735 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1737 struct ll_inode_info *lli;
1738 struct ll_dentry_data *ldd;
1740 if (ll_i2sbi(dir)->ll_sa_max == 0)
1743 lli = ll_i2info(dir);
1745 /* statahead is not allowed for this dir, there may be three causes:
1746 * 1. dir is not opened.
1747 * 2. statahead hit ratio is too low.
1748 * 3. previous stat started statahead thread failed. */
1749 if (!lli->lli_sa_enabled)
1753 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1754 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1755 * For patchless client, lookup intent is not accurate, which may
1756 * misguide statahead. For example:
1757 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1758 * have the same intent -- IT_GETATTR, while one dentry should access
1759 * statahead cache once, otherwise statahead windows is messed up.
1760 * The solution is as following:
1761 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1762 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1763 * bypass interacting with statahead cache by checking
1764 * 'lld_sa_generation == lli->lli_sa_generation'.
1766 ldd = ll_d2d(dentry);
1767 if (ldd != NULL && lli->lli_sa_generation &&
1768 ldd->lld_sa_generation == lli->lli_sa_generation)
1771 if (lli->lli_sa_pattern == LSA_PATTERN_FNAME)
1774 /* not the same process, don't statahead */
1775 if (lli->lli_opendir_pid != current->pid)
1781 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1782 enum cl_fsync_mode mode, int ignore_layout);
1784 static inline int ll_file_nolock(const struct file *file)
1786 struct ll_file_data *fd = file->private_data;
1787 struct inode *inode = file_inode((struct file *)file);
1789 LASSERT(fd != NULL);
1790 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1791 test_bit(LL_SBI_NOLCK, ll_i2sbi(inode)->ll_flags));
1794 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1795 struct lookup_intent *it, __u64 *bits)
1797 if (!it->it_lock_set) {
1798 struct lustre_handle handle;
1800 /* If this inode is a remote object, it will get two
1801 * separate locks in different namespaces, Master MDT,
1802 * where the name entry is, will grant LOOKUP lock,
1803 * remote MDT, where the object is, will grant
1804 * UPDATE|PERM lock. The inode will be attched to both
1805 * LOOKUP and PERM locks, so revoking either locks will
1806 * case the dcache being cleared */
1807 if (it->it_remote_lock_mode) {
1808 handle.cookie = it->it_remote_lock_handle;
1809 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1810 "(%p) for remote lock %#llx\n",
1811 PFID(ll_inode2fid(inode)), inode,
1813 md_set_lock_data(exp, &handle, inode, NULL);
1816 handle.cookie = it->it_lock_handle;
1818 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1819 " for lock %#llx\n",
1820 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1822 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1823 it->it_lock_set = 1;
1827 *bits = it->it_lock_bits;
1830 static inline int d_lustre_invalid(const struct dentry *dentry)
1832 return !ll_d2d(dentry) || ll_d2d(dentry)->lld_invalid;
1836 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1837 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1838 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1839 * unhash this dentry and kill it.
1841 static inline void d_lustre_invalidate(struct dentry *dentry)
1843 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1845 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1847 spin_lock(&dentry->d_lock);
1848 if (lld_is_init(dentry))
1849 ll_d2d(dentry)->lld_invalid = 1;
1850 spin_unlock(&dentry->d_lock);
1853 static inline void d_lustre_revalidate(struct dentry *dentry)
1855 spin_lock(&dentry->d_lock);
1856 LASSERT(ll_d2d(dentry));
1857 ll_d2d(dentry)->lld_invalid = 0;
1858 spin_unlock(&dentry->d_lock);
1861 static inline dev_t ll_compat_encode_dev(dev_t dev)
1863 /* The compat_sys_*stat*() syscalls will fail unless the
1864 * device majors and minors are both less than 256. Note that
1865 * the value returned here will be passed through
1866 * old_encode_dev() in cp_compat_stat(). And so we are not
1867 * trying to return a valid compat (u16) device number, just
1868 * one that will pass the old_valid_dev() check. */
1870 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1873 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1874 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1875 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1876 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1877 struct lu_extent *ext);
1879 int ll_xattr_init(void);
1880 void ll_xattr_fini(void);
1882 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1883 struct cl_page *page, enum cl_req_type crt);
1885 int ll_getparent(struct file *file, struct getparent __user *arg);
1888 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1889 enum op_xvalid xvalid, unsigned int attr_flags);
1891 extern struct lu_env *cl_inode_fini_env;
1892 extern __u16 cl_inode_fini_refcheck;
1894 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1895 void cl_inode_fini(struct inode *inode);
1897 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1898 u32 cl_fid_build_gen(const struct lu_fid *fid);
1900 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1902 return &ll_i2sbi(inode)->ll_pcc_super;
1905 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1907 return ll_i2pccs(ll_info2i(lli));
1911 /* The digested form is made of a FID (16 bytes) followed by the second-to-last
1912 * ciphertext block (16 bytes), so a total length of 32 bytes.
1913 * That way, llcrypt does not compute a digested form of this digest.
1915 struct ll_digest_filename {
1916 struct lu_fid ldf_fid;
1917 char ldf_excerpt[LL_CRYPTO_BLOCK_SIZE];
1920 int ll_prepare_lookup(struct inode *dir, struct dentry *de,
1921 struct llcrypt_name *fname, struct lu_fid *fid);
1922 int ll_setup_filename(struct inode *dir, const struct qstr *iname,
1923 int lookup, struct llcrypt_name *fname,
1924 struct lu_fid *fid);
1925 int ll_fname_disk_to_usr(struct inode *inode,
1926 u32 hash, u32 minor_hash,
1927 struct llcrypt_str *iname, struct llcrypt_str *oname,
1928 struct lu_fid *fid);
1929 int ll_file_open_encrypt(struct inode *inode, struct file *filp);
1931 static inline char *xattr_for_enc(struct inode *inode)
1933 if (ll_sbi_has_name_encrypt(ll_i2sbi(inode)))
1934 return LL_XATTR_NAME_ENCRYPTION_CONTEXT;
1936 return LL_XATTR_NAME_ENCRYPTION_CONTEXT_OLD;
1939 #ifdef HAVE_LUSTRE_CRYPTO
1940 extern const struct llcrypt_operations lustre_cryptops;
1943 /* llite/llite_foreign.c */
1944 int ll_manage_foreign(struct inode *inode, struct lustre_md *lmd);
1945 bool ll_foreign_is_openable(struct dentry *dentry, unsigned int flags);
1946 bool ll_foreign_is_removable(struct dentry *dentry, bool unset);
1948 int ll_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
1950 #endif /* LLITE_INTERNAL_H */