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) 2017, Commissariat a l'Energie Atomique et aux Energies
26 * Author: Henri Doreau <henri.doreau@cea.fr>
29 #define DEBUG_SUBSYSTEM S_MDC
31 #include <linux/init.h>
32 #include <linux/kthread.h>
33 #include <linux/poll.h>
34 #include <linux/miscdevice.h>
36 #include <lustre_log.h>
38 #include "mdc_internal.h"
42 * -- Changelog delivery through character device --
46 * Mutex to protect chlg_registered_devices below
48 static DEFINE_MUTEX(chlg_registered_dev_lock);
51 * Global linked list of all registered devices (one per MDT).
53 static LIST_HEAD(chlg_registered_devices);
56 struct chlg_registered_dev {
57 /* Device name of the form "changelog-{MDTNAME}" */
59 /* Misc device descriptor */
60 struct miscdevice ced_misc;
61 /* OBDs referencing this device (multiple mount point) */
62 struct list_head ced_obds;
63 /* Reference counter for proper deregistration */
65 /* Link within the global chlg_registered_devices */
66 struct list_head ced_link;
69 struct chlg_reader_state {
70 /* Shortcut to the corresponding OBD device */
71 struct obd_device *crs_obd;
72 /* Producer thread (if any) */
73 struct task_struct *crs_prod_task;
74 /* An error occurred that prevents from reading further */
76 /* EOF, no more records available */
78 /* Desired start position */
79 __u64 crs_start_offset;
80 /* Wait queue for the catalog processing thread */
81 wait_queue_head_t crs_waitq_prod;
82 /* Wait queue for the record copy threads */
83 wait_queue_head_t crs_waitq_cons;
84 /* Mutex protecting crs_rec_count and crs_rec_queue */
85 struct mutex crs_lock;
86 /* Number of item in the list */
88 /* List of prefetched enqueued_record::enq_linkage_items */
89 struct list_head crs_rec_queue;
92 struct chlg_rec_entry {
93 /* Link within the chlg_reader_state::crs_rec_queue list */
94 struct list_head enq_linkage;
95 /* Data (enq_record) field length */
97 /* Copy of a changelog record (see struct llog_changelog_rec) */
98 struct changelog_rec enq_record[];
102 /* Number of records to prefetch locally. */
103 CDEV_CHLG_MAX_PREFETCH = 1024,
107 * ChangeLog catalog processing callback invoked on each record.
108 * If the current record is eligible to userland delivery, push
109 * it into the crs_rec_queue where the consumer code will fetch it.
111 * @param[in] env (unused)
112 * @param[in] llh Client-side handle used to identify the llog
113 * @param[in] hdr Header of the current llog record
114 * @param[in,out] data chlg_reader_state passed from caller
116 * @return 0 or LLOG_PROC_* control code on success, negated error on failure.
118 static int chlg_read_cat_process_cb(const struct lu_env *env,
119 struct llog_handle *llh,
120 struct llog_rec_hdr *hdr, void *data)
122 struct llog_changelog_rec *rec;
123 struct chlg_reader_state *crs = data;
124 struct chlg_rec_entry *enq;
129 LASSERT(crs != NULL);
130 LASSERT(hdr != NULL);
132 rec = container_of(hdr, struct llog_changelog_rec, cr_hdr);
134 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
136 CERROR("%s: not a changelog rec %x/%d in llog "DFID" rc = %d\n",
137 crs->crs_obd->obd_name, rec->cr_hdr.lrh_type,
139 PFID(lu_object_fid(&llh->lgh_obj->do_lu)), rc);
143 /* Skip undesired records */
144 if (rec->cr.cr_index < crs->crs_start_offset)
147 CDEBUG(D_HSM, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID" %.*s\n",
148 rec->cr.cr_index, rec->cr.cr_type,
149 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
150 rec->cr.cr_flags & CLF_FLAGMASK,
151 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
152 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
154 wait_event_interruptible(crs->crs_waitq_prod,
155 crs->crs_rec_count < CDEV_CHLG_MAX_PREFETCH ||
156 kthread_should_stop());
158 if (kthread_should_stop())
159 RETURN(LLOG_PROC_BREAK);
161 len = changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
162 OBD_ALLOC(enq, sizeof(*enq) + len);
166 INIT_LIST_HEAD(&enq->enq_linkage);
167 enq->enq_length = len;
168 memcpy(enq->enq_record, &rec->cr, len);
170 mutex_lock(&crs->crs_lock);
171 list_add_tail(&enq->enq_linkage, &crs->crs_rec_queue);
172 crs->crs_rec_count++;
173 mutex_unlock(&crs->crs_lock);
175 wake_up_all(&crs->crs_waitq_cons);
181 * Remove record from the list it is attached to and free it.
183 static void enq_record_delete(struct chlg_rec_entry *rec)
185 list_del(&rec->enq_linkage);
186 OBD_FREE(rec, sizeof(*rec) + rec->enq_length);
190 * Record prefetch thread entry point. Opens the changelog catalog and starts
193 * @param[in,out] args chlg_reader_state passed from caller.
194 * @return 0 on success, negated error code on failure.
196 static int chlg_load(void *args)
198 struct chlg_reader_state *crs = args;
199 struct obd_device *obd = crs->crs_obd;
200 struct llog_ctxt *ctx = NULL;
201 struct llog_handle *llh = NULL;
205 ctx = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
207 GOTO(err_out, rc = -ENOENT);
209 rc = llog_open(NULL, ctx, &llh, NULL, CHANGELOG_CATALOG,
212 CERROR("%s: fail to open changelog catalog: rc = %d\n",
217 rc = llog_init_handle(NULL, llh,
220 LLOG_F_EXT_EXTRA_FLAGS,
223 CERROR("%s: fail to init llog handle: rc = %d\n",
228 rc = llog_cat_process(NULL, llh, chlg_read_cat_process_cb, crs, 0, 0);
230 CERROR("%s: fail to process llog: rc = %d\n", obd->obd_name, rc);
240 wake_up_all(&crs->crs_waitq_cons);
243 llog_cat_close(NULL, llh);
248 wait_event_interruptible(crs->crs_waitq_prod, kthread_should_stop());
254 * Read handler, dequeues records from the chlg_reader_state if any.
255 * No partial records are copied to userland so this function can return less
256 * data than required (short read).
258 * @param[in] file File pointer to the character device.
259 * @param[out] buff Userland buffer where to copy the records.
260 * @param[in] count Userland buffer size.
261 * @param[out] ppos File position, updated with the index number of the next
263 * @return number of copied bytes on success, negated error code on failure.
265 static ssize_t chlg_read(struct file *file, char __user *buff, size_t count,
268 struct chlg_reader_state *crs = file->private_data;
269 struct chlg_rec_entry *rec;
270 struct chlg_rec_entry *tmp;
271 size_t written_total = 0;
276 if (file->f_flags & O_NONBLOCK && crs->crs_rec_count == 0) {
277 if (crs->crs_err < 0)
278 RETURN(crs->crs_err);
279 else if (crs->crs_eof)
285 rc = wait_event_interruptible(crs->crs_waitq_cons,
286 crs->crs_rec_count > 0 || crs->crs_eof || crs->crs_err);
288 mutex_lock(&crs->crs_lock);
289 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage) {
290 if (written_total + rec->enq_length > count)
293 if (copy_to_user(buff, rec->enq_record, rec->enq_length)) {
298 buff += rec->enq_length;
299 written_total += rec->enq_length;
301 crs->crs_rec_count--;
302 list_move_tail(&rec->enq_linkage, &consumed);
304 crs->crs_start_offset = rec->enq_record->cr_index + 1;
306 mutex_unlock(&crs->crs_lock);
308 if (written_total > 0) {
310 wake_up_all(&crs->crs_waitq_prod);
311 } else if (rc == 0) {
315 list_for_each_entry_safe(rec, tmp, &consumed, enq_linkage)
316 enq_record_delete(rec);
318 *ppos = crs->crs_start_offset;
324 * Jump to a given record index. Helper for chlg_llseek().
326 * @param[in,out] crs Internal reader state.
327 * @param[in] offset Desired offset (index record).
328 * @return 0 on success, negated error code on failure.
330 static int chlg_set_start_offset(struct chlg_reader_state *crs, __u64 offset)
332 struct chlg_rec_entry *rec;
333 struct chlg_rec_entry *tmp;
335 mutex_lock(&crs->crs_lock);
336 if (offset < crs->crs_start_offset) {
337 mutex_unlock(&crs->crs_lock);
341 crs->crs_start_offset = offset;
342 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage) {
343 struct changelog_rec *cr = rec->enq_record;
345 if (cr->cr_index >= crs->crs_start_offset)
348 crs->crs_rec_count--;
349 enq_record_delete(rec);
352 mutex_unlock(&crs->crs_lock);
353 wake_up_all(&crs->crs_waitq_prod);
358 * Move read pointer to a certain record index, encoded as an offset.
360 * @param[in,out] file File pointer to the changelog character device
361 * @param[in] off Offset to skip, actually a record index, not byte count
362 * @param[in] whence Relative/Absolute interpretation of the offset
363 * @return the resulting position on success or negated error code on failure.
365 static loff_t chlg_llseek(struct file *file, loff_t off, int whence)
367 struct chlg_reader_state *crs = file->private_data;
376 pos = file->f_pos + off;
383 /* We cannot go backward */
384 if (pos < file->f_pos)
387 rc = chlg_set_start_offset(crs, pos);
396 * Clear record range for a given changelog reader.
398 * @param[in] crs Current internal state.
399 * @param[in] reader Changelog reader ID (cl1, cl2...)
400 * @param[in] record Record index up which to clear
401 * @return 0 on success, negated error code on failure.
403 static int chlg_clear(struct chlg_reader_state *crs, __u32 reader, __u64 record)
405 struct obd_device *obd = crs->crs_obd;
406 struct changelog_setinfo cs = {
411 return obd_set_info_async(NULL, obd->obd_self_export,
412 strlen(KEY_CHANGELOG_CLEAR),
413 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs, NULL);
416 /** Maximum changelog control command size */
417 #define CHLG_CONTROL_CMD_MAX 64
420 * Handle writes() into the changelog character device. Write() can be used
421 * to request special control operations.
423 * @param[in] file File pointer to the changelog character device
424 * @param[in] buff User supplied data (written data)
425 * @param[in] count Number of written bytes
426 * @param[in] off (unused)
427 * @return number of written bytes on success, negated error code on failure.
429 static ssize_t chlg_write(struct file *file, const char __user *buff,
430 size_t count, loff_t *off)
432 struct chlg_reader_state *crs = file->private_data;
439 if (count > CHLG_CONTROL_CMD_MAX)
442 OBD_ALLOC(kbuf, CHLG_CONTROL_CMD_MAX);
446 if (copy_from_user(kbuf, buff, count))
447 GOTO(out_kbuf, rc = -EFAULT);
449 kbuf[CHLG_CONTROL_CMD_MAX - 1] = '\0';
451 if (sscanf(kbuf, "clear:cl%u:%llu", &reader, &record) == 2)
452 rc = chlg_clear(crs, reader, record);
458 OBD_FREE(kbuf, CHLG_CONTROL_CMD_MAX);
459 return rc < 0 ? rc : count;
463 * Find the OBD device associated to a changelog character device.
464 * @param[in] cdev character device instance descriptor
465 * @return corresponding OBD device or NULL if none was found.
467 static struct obd_device *chlg_obd_get(dev_t cdev)
469 int minor = MINOR(cdev);
470 struct obd_device *obd = NULL;
471 struct chlg_registered_dev *curr;
473 mutex_lock(&chlg_registered_dev_lock);
474 list_for_each_entry(curr, &chlg_registered_devices, ced_link) {
475 if (curr->ced_misc.minor == minor) {
476 /* take the first available OBD device attached */
477 obd = list_first_entry(&curr->ced_obds,
479 u.cli.cl_chg_dev_linkage);
483 mutex_unlock(&chlg_registered_dev_lock);
488 * Open handler, initialize internal CRS state and spawn prefetch thread if
490 * @param[in] inode Inode struct for the open character device.
491 * @param[in] file Corresponding file pointer.
492 * @return 0 on success, negated error code on failure.
494 static int chlg_open(struct inode *inode, struct file *file)
496 struct chlg_reader_state *crs;
497 struct obd_device *obd = chlg_obd_get(inode->i_rdev);
498 struct task_struct *task;
510 crs->crs_err = false;
511 crs->crs_eof = false;
513 mutex_init(&crs->crs_lock);
514 INIT_LIST_HEAD(&crs->crs_rec_queue);
515 init_waitqueue_head(&crs->crs_waitq_prod);
516 init_waitqueue_head(&crs->crs_waitq_cons);
518 if (file->f_mode & FMODE_READ) {
519 task = kthread_run(chlg_load, crs, "chlg_load_thread");
522 CERROR("%s: cannot start changelog thread: rc = %d\n",
526 crs->crs_prod_task = task;
529 file->private_data = crs;
538 * Close handler, release resources.
540 * @param[in] inode Inode struct for the open character device.
541 * @param[in] file Corresponding file pointer.
542 * @return 0 on success, negated error code on failure.
544 static int chlg_release(struct inode *inode, struct file *file)
546 struct chlg_reader_state *crs = file->private_data;
547 struct chlg_rec_entry *rec;
548 struct chlg_rec_entry *tmp;
551 if (crs->crs_prod_task)
552 rc = kthread_stop(crs->crs_prod_task);
554 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage)
555 enq_record_delete(rec);
563 * Poll handler, indicates whether the device is readable (new records) and
566 * @param[in] file Device file pointer.
567 * @param[in] wait (opaque)
568 * @return combination of the poll status flags.
570 static unsigned int chlg_poll(struct file *file, poll_table *wait)
572 struct chlg_reader_state *crs = file->private_data;
573 unsigned int mask = 0;
575 mutex_lock(&crs->crs_lock);
576 poll_wait(file, &crs->crs_waitq_cons, wait);
577 if (crs->crs_rec_count > 0)
578 mask |= POLLIN | POLLRDNORM;
583 mutex_unlock(&crs->crs_lock);
587 static const struct file_operations chlg_fops = {
588 .owner = THIS_MODULE,
589 .llseek = chlg_llseek,
593 .release = chlg_release,
598 * This uses obd_name of the form: "testfs-MDT0000-mdc-ffff88006501600"
599 * and returns a name of the form: "changelog-testfs-MDT0000".
601 static void get_chlg_name(char *name, size_t name_len, struct obd_device *obd)
605 snprintf(name, name_len, "changelog-%s", obd->obd_name);
607 /* Find the 2nd '-' from the end and truncate on it */
608 for (i = 0; i < 2; i++) {
609 char *p = strrchr(name, '-');
618 * Find a changelog character device by name.
619 * All devices registered during MDC setup are listed in a global list with
620 * their names attached.
622 static struct chlg_registered_dev *
623 chlg_registered_dev_find_by_name(const char *name)
625 struct chlg_registered_dev *dit;
627 list_for_each_entry(dit, &chlg_registered_devices, ced_link)
628 if (strcmp(name, dit->ced_name) == 0)
634 * Find chlg_registered_dev structure for a given OBD device.
635 * This is bad O(n^2) but for each filesystem:
636 * - N is # of MDTs times # of mount points
637 * - this only runs at shutdown
639 static struct chlg_registered_dev *
640 chlg_registered_dev_find_by_obd(const struct obd_device *obd)
642 struct chlg_registered_dev *dit;
643 struct obd_device *oit;
645 list_for_each_entry(dit, &chlg_registered_devices, ced_link)
646 list_for_each_entry(oit, &dit->ced_obds,
647 u.cli.cl_chg_dev_linkage)
654 * Changelog character device initialization.
655 * Register a misc character device with a dynamic minor number, under a name
656 * of the form: 'changelog-fsname-MDTxxxx'. Reference this OBD device with it.
658 * @param[in] obd This MDC obd_device.
659 * @return 0 on success, negated error code on failure.
661 int mdc_changelog_cdev_init(struct obd_device *obd)
663 struct chlg_registered_dev *exist;
664 struct chlg_registered_dev *entry;
668 OBD_ALLOC_PTR(entry);
672 get_chlg_name(entry->ced_name, sizeof(entry->ced_name), obd);
674 entry->ced_misc.minor = MISC_DYNAMIC_MINOR;
675 entry->ced_misc.name = entry->ced_name;
676 entry->ced_misc.fops = &chlg_fops;
678 kref_init(&entry->ced_refs);
679 INIT_LIST_HEAD(&entry->ced_obds);
680 INIT_LIST_HEAD(&entry->ced_link);
682 mutex_lock(&chlg_registered_dev_lock);
683 exist = chlg_registered_dev_find_by_name(entry->ced_name);
685 kref_get(&exist->ced_refs);
686 list_add_tail(&obd->u.cli.cl_chg_dev_linkage, &exist->ced_obds);
687 GOTO(out_unlock, rc = 0);
690 /* Register new character device */
691 rc = misc_register(&entry->ced_misc);
693 GOTO(out_unlock, rc);
695 list_add_tail(&obd->u.cli.cl_chg_dev_linkage, &entry->ced_obds);
696 list_add_tail(&entry->ced_link, &chlg_registered_devices);
698 entry = NULL; /* prevent it from being freed below */
701 mutex_unlock(&chlg_registered_dev_lock);
708 * Deregister a changelog character device whose refcount has reached zero.
710 static void chlg_dev_clear(struct kref *kref)
712 struct chlg_registered_dev *entry = container_of(kref,
713 struct chlg_registered_dev,
717 list_del(&entry->ced_link);
718 misc_deregister(&entry->ced_misc);
724 * Release OBD, decrease reference count of the corresponding changelog device.
726 void mdc_changelog_cdev_finish(struct obd_device *obd)
728 struct chlg_registered_dev *dev = chlg_registered_dev_find_by_obd(obd);
731 mutex_lock(&chlg_registered_dev_lock);
732 list_del_init(&obd->u.cli.cl_chg_dev_linkage);
733 kref_put(&dev->ced_refs, chlg_dev_clear);
734 mutex_unlock(&chlg_registered_dev_lock);