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 * Copyright (c) 2017, Intel Corporation.
28 * Author: Henri Doreau <henri.doreau@cea.fr>
31 #define DEBUG_SUBSYSTEM S_MDC
33 #include <linux/init.h>
34 #include <linux/kthread.h>
35 #include <linux/poll.h>
36 #include <linux/device.h>
37 #include <linux/cdev.h>
38 #include <linux/idr.h>
40 #include <lustre_log.h>
41 #include <uapi/linux/lustre/lustre_ioctl.h>
43 #include "mdc_internal.h"
47 * -- Changelog delivery through character device --
51 * Mutex to protect chlg_registered_devices below
53 static DEFINE_MUTEX(chlg_registered_dev_lock);
56 * Global linked list of all registered devices (one per MDT).
58 static LIST_HEAD(chlg_registered_devices);
61 struct chlg_registered_dev {
62 /* Device name of the form "changelog-{MDTNAME}" */
64 /* changelog char device */
66 struct device ced_device;
67 /* OBDs referencing this device (multiple mount point) */
68 struct list_head ced_obds;
69 /* Reference counter for proper deregistration */
71 /* Link within the global chlg_registered_devices */
72 struct list_head ced_link;
75 struct chlg_reader_state {
76 /* Shortcut to the corresponding OBD device */
77 struct obd_device *crs_obd;
78 /* the corresponding chlg_registered_dev */
79 struct chlg_registered_dev *crs_ced;
80 /* Producer thread (if any) */
81 struct task_struct *crs_prod_task;
82 /* An error occurred that prevents from reading further */
84 /* EOF, no more records available */
86 /* Desired start position */
87 __u64 crs_start_offset;
88 /* Wait queue for the catalog processing thread */
89 wait_queue_head_t crs_waitq_prod;
90 /* Wait queue for the record copy threads */
91 wait_queue_head_t crs_waitq_cons;
92 /* Mutex protecting crs_rec_count and crs_rec_queue */
93 struct mutex crs_lock;
94 /* Number of item in the list */
96 /* List of prefetched enqueued_record::enq_linkage_items */
97 struct list_head crs_rec_queue;
98 unsigned int crs_last_catidx;
99 unsigned int crs_last_idx;
103 struct chlg_rec_entry {
104 /* Link within the chlg_reader_state::crs_rec_queue list */
105 struct list_head enq_linkage;
106 /* Data (enq_record) field length */
108 /* Copy of a changelog record (see struct llog_changelog_rec) */
109 struct changelog_rec enq_record[];
113 /* Number of records to prefetch locally. */
114 CDEV_CHLG_MAX_PREFETCH = 1024,
117 DEFINE_IDR(mdc_changelog_minor_idr);
118 static DEFINE_SPINLOCK(chlg_minor_lock);
120 static int chlg_minor_alloc(int *pminor)
122 void *minor_allocated = (void *)-1;
125 idr_preload(GFP_KERNEL);
126 spin_lock(&chlg_minor_lock);
127 minor = idr_alloc(&mdc_changelog_minor_idr, minor_allocated, 0,
128 MDC_CHANGELOG_DEV_COUNT, GFP_NOWAIT);
129 spin_unlock(&chlg_minor_lock);
139 static void chlg_minor_free(int minor)
141 spin_lock(&chlg_minor_lock);
142 idr_remove(&mdc_changelog_minor_idr, minor);
143 spin_unlock(&chlg_minor_lock);
146 static void chlg_device_release(struct device *dev)
148 struct chlg_registered_dev *entry = dev_get_drvdata(dev);
150 chlg_minor_free(MINOR(entry->ced_cdev.dev));
155 * Deregister a changelog character device whose refcount has reached zero.
157 static void chlg_dev_clear(struct kref *kref)
159 struct chlg_registered_dev *entry;
162 entry = container_of(kref, struct chlg_registered_dev,
165 list_del(&entry->ced_link);
166 cdev_device_del(&entry->ced_cdev, &entry->ced_device);
167 put_device(&entry->ced_device);
171 static inline struct obd_device* chlg_obd_get(struct chlg_registered_dev *dev)
173 struct obd_device *obd;
175 mutex_lock(&chlg_registered_dev_lock);
176 if (list_empty(&dev->ced_obds))
179 obd = list_first_entry(&dev->ced_obds, struct obd_device,
180 u.cli.cl_chg_dev_linkage);
181 class_incref(obd, "changelog", dev);
182 mutex_unlock(&chlg_registered_dev_lock);
186 static inline void chlg_obd_put(struct chlg_registered_dev *dev,
187 struct obd_device *obd)
189 class_decref(obd, "changelog", dev);
193 * ChangeLog catalog processing callback invoked on each record.
194 * If the current record is eligible to userland delivery, push
195 * it into the crs_rec_queue where the consumer code will fetch it.
197 * @param[in] env (unused)
198 * @param[in] llh Client-side handle used to identify the llog
199 * @param[in] hdr Header of the current llog record
200 * @param[in,out] data chlg_reader_state passed from caller
202 * @return 0 or LLOG_PROC_* control code on success, negated error on failure.
204 static int chlg_read_cat_process_cb(const struct lu_env *env,
205 struct llog_handle *llh,
206 struct llog_rec_hdr *hdr, void *data)
208 struct llog_changelog_rec *rec;
209 struct chlg_reader_state *crs = data;
210 struct chlg_rec_entry *enq;
215 LASSERT(crs != NULL);
216 LASSERT(hdr != NULL);
218 rec = container_of(hdr, struct llog_changelog_rec, cr_hdr);
220 crs->crs_last_catidx = llh->lgh_hdr->llh_cat_idx;
221 crs->crs_last_idx = hdr->lrh_index;
223 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
225 CERROR("%s: not a changelog rec %x/%d in llog "DFID" rc = %d\n",
226 crs->crs_obd->obd_name, rec->cr_hdr.lrh_type,
228 PFID(lu_object_fid(&llh->lgh_obj->do_lu)), rc);
232 /* Skip undesired records */
233 if (rec->cr.cr_index < crs->crs_start_offset)
236 CDEBUG(D_HSM, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID" %.*s\n",
237 rec->cr.cr_index, rec->cr.cr_type,
238 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
239 rec->cr.cr_flags & CLF_FLAGMASK,
240 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
241 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
243 wait_event_interruptible(crs->crs_waitq_prod,
244 crs->crs_rec_count < CDEV_CHLG_MAX_PREFETCH ||
245 kthread_should_stop());
247 if (kthread_should_stop())
248 RETURN(LLOG_PROC_BREAK);
250 len = changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
251 OBD_ALLOC(enq, sizeof(*enq) + len);
255 INIT_LIST_HEAD(&enq->enq_linkage);
256 enq->enq_length = len;
257 memcpy(enq->enq_record, &rec->cr, len);
259 mutex_lock(&crs->crs_lock);
260 list_add_tail(&enq->enq_linkage, &crs->crs_rec_queue);
261 crs->crs_rec_count++;
262 mutex_unlock(&crs->crs_lock);
264 wake_up(&crs->crs_waitq_cons);
270 * Remove record from the list it is attached to and free it.
272 static void enq_record_delete(struct chlg_rec_entry *rec)
274 list_del(&rec->enq_linkage);
275 OBD_FREE(rec, sizeof(*rec) + rec->enq_length);
279 * Record prefetch thread entry point. Opens the changelog catalog and starts
282 * @param[in,out] args chlg_reader_state passed from caller.
283 * @return 0 on success, negated error code on failure.
285 static int chlg_load(void *args)
287 struct chlg_reader_state *crs = args;
288 struct chlg_registered_dev *ced = crs->crs_ced;
289 struct obd_device *obd = NULL;
290 struct llog_ctxt *ctx = NULL;
291 struct llog_handle *llh = NULL;
295 crs->crs_last_catidx = 0;
296 crs->crs_last_idx = 0;
299 obd = chlg_obd_get(ced);
305 ctx = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
307 GOTO(err_out, rc = -ENOENT);
309 rc = llog_open(NULL, ctx, &llh, NULL, CHANGELOG_CATALOG,
312 CERROR("%s: fail to open changelog catalog: rc = %d\n",
318 rc = llog_init_handle(NULL, llh,
321 LLOG_F_EXT_EXTRA_FLAGS |
322 LLOG_F_EXT_X_UIDGID |
328 CERROR("%s: fail to init llog handle: rc = %d\n",
333 rc = llog_cat_process(NULL, llh, chlg_read_cat_process_cb, crs,
334 crs->crs_last_catidx, crs->crs_last_idx);
336 CERROR("%s: fail to process llog: rc = %d\n", obd->obd_name, rc);
339 if (!kthread_should_stop() && crs->crs_poll) {
340 llog_cat_close(NULL, llh);
342 class_decref(obd, "changelog", crs);
343 schedule_timeout_interruptible(cfs_time_seconds(1));
353 wake_up(&crs->crs_waitq_cons);
356 llog_cat_close(NULL, llh);
362 chlg_obd_put(ced, obd);
363 wait_event_interruptible(crs->crs_waitq_prod, kthread_should_stop());
369 * Read handler, dequeues records from the chlg_reader_state if any.
370 * No partial records are copied to userland so this function can return less
371 * data than required (short read).
373 * @param[in] file File pointer to the character device.
374 * @param[out] buff Userland buffer where to copy the records.
375 * @param[in] count Userland buffer size.
376 * @param[out] ppos File position, updated with the index number of the next
378 * @return number of copied bytes on success, negated error code on failure.
380 static ssize_t chlg_read(struct file *file, char __user *buff, size_t count,
383 struct chlg_reader_state *crs = file->private_data;
384 struct chlg_rec_entry *rec;
385 struct chlg_rec_entry *tmp;
386 size_t written_total = 0;
391 if (file->f_flags & O_NONBLOCK && crs->crs_rec_count == 0) {
392 if (crs->crs_err < 0)
393 RETURN(crs->crs_err);
394 else if (crs->crs_eof)
400 rc = wait_event_interruptible(crs->crs_waitq_cons,
401 crs->crs_rec_count > 0 || crs->crs_eof || crs->crs_err);
403 mutex_lock(&crs->crs_lock);
404 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage) {
405 if (written_total + rec->enq_length > count)
408 if (copy_to_user(buff, rec->enq_record, rec->enq_length)) {
413 buff += rec->enq_length;
414 written_total += rec->enq_length;
416 crs->crs_rec_count--;
417 list_move_tail(&rec->enq_linkage, &consumed);
419 crs->crs_start_offset = rec->enq_record->cr_index + 1;
421 mutex_unlock(&crs->crs_lock);
423 if (written_total > 0) {
425 wake_up(&crs->crs_waitq_prod);
426 } else if (rc == 0) {
430 list_for_each_entry_safe(rec, tmp, &consumed, enq_linkage)
431 enq_record_delete(rec);
433 *ppos = crs->crs_start_offset;
439 * Jump to a given record index. Helper for chlg_llseek().
441 * @param[in,out] crs Internal reader state.
442 * @param[in] offset Desired offset (index record).
443 * @return 0 on success, negated error code on failure.
445 static int chlg_set_start_offset(struct chlg_reader_state *crs, __u64 offset)
447 struct chlg_rec_entry *rec;
448 struct chlg_rec_entry *tmp;
450 mutex_lock(&crs->crs_lock);
451 if (offset < crs->crs_start_offset) {
452 mutex_unlock(&crs->crs_lock);
456 crs->crs_start_offset = offset;
457 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage) {
458 struct changelog_rec *cr = rec->enq_record;
460 if (cr->cr_index >= crs->crs_start_offset)
463 crs->crs_rec_count--;
464 enq_record_delete(rec);
467 mutex_unlock(&crs->crs_lock);
468 wake_up(&crs->crs_waitq_prod);
473 * Move read pointer to a certain record index, encoded as an offset.
475 * @param[in,out] file File pointer to the changelog character device
476 * @param[in] off Offset to skip, actually a record index, not byte count
477 * @param[in] whence Relative/Absolute interpretation of the offset
478 * @return the resulting position on success or negated error code on failure.
480 static loff_t chlg_llseek(struct file *file, loff_t off, int whence)
482 struct chlg_reader_state *crs = file->private_data;
491 pos = file->f_pos + off;
498 /* We cannot go backward */
499 if (pos < file->f_pos)
502 rc = chlg_set_start_offset(crs, pos);
511 * Clear record range for a given changelog reader.
513 * @param[in] crs Current internal state.
514 * @param[in] reader Changelog reader ID (cl1, cl2...)
515 * @param[in] record Record index up which to clear
516 * @return 0 on success, negated error code on failure.
518 static int chlg_clear(struct chlg_reader_state *crs, __u32 reader, __u64 record)
520 struct obd_device *obd = NULL;
521 struct changelog_setinfo cs = {
527 obd = chlg_obd_get(crs->crs_ced);
531 rc = obd_set_info_async(NULL, obd->obd_self_export,
532 strlen(KEY_CHANGELOG_CLEAR),
533 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs, NULL);
535 chlg_obd_put(crs->crs_ced, obd);
539 /** Maximum changelog control command size */
540 #define CHLG_CONTROL_CMD_MAX 64
543 * Handle writes() into the changelog character device. Write() can be used
544 * to request special control operations.
546 * @param[in] file File pointer to the changelog character device
547 * @param[in] buff User supplied data (written data)
548 * @param[in] count Number of written bytes
549 * @param[in] off (unused)
550 * @return number of written bytes on success, negated error code on failure.
552 static ssize_t chlg_write(struct file *file, const char __user *buff,
553 size_t count, loff_t *off)
555 struct chlg_reader_state *crs = file->private_data;
562 if (count > CHLG_CONTROL_CMD_MAX)
565 OBD_ALLOC(kbuf, CHLG_CONTROL_CMD_MAX);
569 if (copy_from_user(kbuf, buff, count))
570 GOTO(out_kbuf, rc = -EFAULT);
572 kbuf[CHLG_CONTROL_CMD_MAX - 1] = '\0';
574 if (sscanf(kbuf, "clear:cl%u:%llu", &reader, &record) == 2)
575 rc = chlg_clear(crs, reader, record);
581 OBD_FREE(kbuf, CHLG_CONTROL_CMD_MAX);
582 return rc < 0 ? rc : count;
586 * Open handler, initialize internal CRS state and spawn prefetch thread if
588 * @param[in] inode Inode struct for the open character device.
589 * @param[in] file Corresponding file pointer.
590 * @return 0 on success, negated error code on failure.
592 static int chlg_open(struct inode *inode, struct file *file)
594 struct chlg_reader_state *crs;
595 struct chlg_registered_dev *dev;
596 struct task_struct *task;
600 dev = container_of(inode->i_cdev, struct chlg_registered_dev, ced_cdev);
606 kref_get(&dev->ced_refs);
608 crs->crs_err = false;
609 crs->crs_eof = false;
611 mutex_init(&crs->crs_lock);
612 INIT_LIST_HEAD(&crs->crs_rec_queue);
613 init_waitqueue_head(&crs->crs_waitq_prod);
614 init_waitqueue_head(&crs->crs_waitq_cons);
616 if (file->f_mode & FMODE_READ) {
617 task = kthread_run(chlg_load, crs, "chlg_load_thread");
620 CERROR("%s: cannot start changelog thread: rc = %d\n",
624 crs->crs_prod_task = task;
627 file->private_data = crs;
631 kref_put(&dev->ced_refs, chlg_dev_clear);
637 * Close handler, release resources.
639 * @param[in] inode Inode struct for the open character device.
640 * @param[in] file Corresponding file pointer.
641 * @return 0 on success, negated error code on failure.
643 static int chlg_release(struct inode *inode, struct file *file)
645 struct chlg_reader_state *crs = file->private_data;
646 struct chlg_rec_entry *rec;
647 struct chlg_rec_entry *tmp;
650 if (crs->crs_prod_task)
651 rc = kthread_stop(crs->crs_prod_task);
653 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage)
654 enq_record_delete(rec);
656 kref_put(&crs->crs_ced->ced_refs, chlg_dev_clear);
663 * Poll handler, indicates whether the device is readable (new records) and
666 * @param[in] file Device file pointer.
667 * @param[in] wait (opaque)
668 * @return combination of the poll status flags.
670 static unsigned int chlg_poll(struct file *file, poll_table *wait)
672 struct chlg_reader_state *crs = file->private_data;
673 unsigned int mask = 0;
675 mutex_lock(&crs->crs_lock);
676 poll_wait(file, &crs->crs_waitq_cons, wait);
677 if (crs->crs_rec_count > 0)
678 mask |= POLLIN | POLLRDNORM;
683 mutex_unlock(&crs->crs_lock);
687 static long chlg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
691 struct chlg_reader_state *crs = file->private_data;
693 case OBD_IOC_CHLG_POLL:
694 crs->crs_poll = !!arg;
704 static const struct file_operations chlg_fops = {
705 .owner = THIS_MODULE,
706 .llseek = chlg_llseek,
710 .release = chlg_release,
712 .unlocked_ioctl = chlg_ioctl,
716 * This uses obd_name of the form: "testfs-MDT0000-mdc-ffff88006501600"
717 * and returns a name of the form: "changelog-testfs-MDT0000".
719 static void get_target_name(char *name, size_t name_len, struct obd_device *obd)
723 snprintf(name, name_len, "%s", obd->obd_name);
725 /* Find the 2nd '-' from the end and truncate on it */
726 for (i = 0; i < 2; i++) {
727 char *p = strrchr(name, '-');
736 * Find a changelog character device by name.
737 * All devices registered during MDC setup are listed in a global list with
738 * their names attached.
740 static struct chlg_registered_dev *
741 chlg_registered_dev_find_by_name(const char *name)
743 struct chlg_registered_dev *dit;
745 LASSERT(mutex_is_locked(&chlg_registered_dev_lock));
746 list_for_each_entry(dit, &chlg_registered_devices, ced_link)
747 if (strcmp(name, dit->ced_name) == 0)
753 * Find chlg_registered_dev structure for a given OBD device.
754 * This is bad O(n^2) but for each filesystem:
755 * - N is # of MDTs times # of mount points
756 * - this only runs at shutdown
758 static struct chlg_registered_dev *
759 chlg_registered_dev_find_by_obd(const struct obd_device *obd)
761 struct chlg_registered_dev *dit;
762 struct obd_device *oit;
764 LASSERT(mutex_is_locked(&chlg_registered_dev_lock));
765 list_for_each_entry(dit, &chlg_registered_devices, ced_link)
766 list_for_each_entry(oit, &dit->ced_obds,
767 u.cli.cl_chg_dev_linkage)
774 * Changelog character device initialization.
775 * Register a misc character device with a dynamic minor number, under a name
776 * of the form: 'changelog-fsname-MDTxxxx'. Reference this OBD device with it.
778 * @param[in] obd This MDC obd_device.
779 * @return 0 on success, negated error code on failure.
781 int mdc_changelog_cdev_init(struct obd_device *obd)
783 struct chlg_registered_dev *exist;
784 struct chlg_registered_dev *entry;
788 OBD_ALLOC_PTR(entry);
792 get_target_name(entry->ced_name, sizeof(entry->ced_name), obd);
794 kref_init(&entry->ced_refs);
795 INIT_LIST_HEAD(&entry->ced_obds);
796 INIT_LIST_HEAD(&entry->ced_link);
798 mutex_lock(&chlg_registered_dev_lock);
799 exist = chlg_registered_dev_find_by_name(entry->ced_name);
801 kref_get(&exist->ced_refs);
802 list_add_tail(&obd->u.cli.cl_chg_dev_linkage, &exist->ced_obds);
803 GOTO(out_unlock, rc = 0);
806 list_add_tail(&obd->u.cli.cl_chg_dev_linkage, &entry->ced_obds);
807 list_add_tail(&entry->ced_link, &chlg_registered_devices);
809 rc = chlg_minor_alloc(&minor);
811 GOTO(out_unlock, rc);
813 device_initialize(&entry->ced_device);
814 entry->ced_device.devt = MKDEV(MAJOR(mdc_changelog_dev), minor);
815 entry->ced_device.class = mdc_changelog_class;
816 entry->ced_device.release = chlg_device_release;
817 dev_set_drvdata(&entry->ced_device, entry);
818 rc = dev_set_name(&entry->ced_device, "%s-%s", MDC_CHANGELOG_DEV_NAME,
823 /* Register new character device */
824 cdev_init(&entry->ced_cdev, &chlg_fops);
825 entry->ced_cdev.owner = THIS_MODULE;
826 rc = cdev_device_add(&entry->ced_cdev, &entry->ced_device);
828 GOTO(out_device_name, rc);
830 entry = NULL; /* prevent it from being freed below */
831 GOTO(out_unlock, rc = 0);
834 kfree_const(entry->ced_device.kobj.name);
837 chlg_minor_free(minor);
839 list_del_init(&obd->u.cli.cl_chg_dev_linkage);
840 list_del(&entry->ced_link);
843 mutex_unlock(&chlg_registered_dev_lock);
850 * Release OBD, decrease reference count of the corresponding changelog device.
852 void mdc_changelog_cdev_finish(struct obd_device *obd)
854 struct chlg_registered_dev *dev;
857 mutex_lock(&chlg_registered_dev_lock);
858 dev = chlg_registered_dev_find_by_obd(obd);
859 list_del_init(&obd->u.cli.cl_chg_dev_linkage);
860 kref_put(&dev->ced_refs, chlg_dev_clear);
861 mutex_unlock(&chlg_registered_dev_lock);